Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Digital dental radiographs : influence of image quality and software manipulation on their interpretation Delantoni, Antigoni 2002

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_2002-0062.pdf [ 11.18MB ]
Metadata
JSON: 831-1.0090121.json
JSON-LD: 831-1.0090121-ld.json
RDF/XML (Pretty): 831-1.0090121-rdf.xml
RDF/JSON: 831-1.0090121-rdf.json
Turtle: 831-1.0090121-turtle.txt
N-Triples: 831-1.0090121-rdf-ntriples.txt
Original Record: 831-1.0090121-source.json
Full Text
831-1.0090121-fulltext.txt
Citation
831-1.0090121.ris

Full Text

Digital dental radiographs: Influence of image quality and software manipulation on their interpretation by Antigoni Delantoni D.M.D. Aristotelion University of Thessaloniki, Greece, 1998 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE MASTER OF SCIENCE in THE FACULTY OF GRADUATE STUDIES Department of Oral Biological and Medical Sciences We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH COLUMBIA February 2002 © Antigoni Delantoni, 2002 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Department The University of British Columbia Vancouver, Canada Date 4 1(32. DE-6 (2/88) 11 Abstract: With the technologic advances in digital systems there has been a development in the field of digital imaging. Systems have developed that allow us to take radiographs with a reduced patient exposure and higher convenience compared to conventional film. In many cases the information is electronically transferred either to get a second opinion or to share information. We need to know to what extent there is significant information loss with digital systems. It is therefore useful to compare the image quality of digital systems to conventional film. The aim of this study was to evaluate the amount of diagnostic information loss during the scanning of conventional radiographs by comparing images of different qualities for periodontal defects. Radiographs were taken and scanned at different resolutions (600,300,150 and 75dpi). A visual analog scale was used for the subjective assessment of image quality. Observers were asked to mark points on the selected images. Analysis of Variance was used to demonstrate significant differences. From the results we obtained from our data, no significant differences were found between the resolutions of 600 and 300 dpi, while there were differences between low quality images (75 dpi) and the higher ones. We drew conclusions as to whether there is significant amount of information loss during the scanning of radiographs in cases of periodontal defects. Ill In addition to the periodontal model, further study on image manipulation by blurring was done using an endodontic model, and the subjective and objective performances and measurements of the observers were compared. The feature of blurring was selected since it is a feature present in Microsoft PowerPoint and no studies had been done on it in the past. The images were blurred using a blurring feature provided by software by the NIH image, and a comparison was made between blurred and non-blurred images for good and poor quality images. Differences were found in the points' distribution and the measurements between the two features. We concluded that blurring worsens good quality images, while poor quality images are improved. iv Table of contents Abstract ii Table of contents iv List of tables vii List of figures viii Acknowledgements xii Chapter One Introduction and review of the literature 1 1.1 Introduction 2 1.2 Review of the literature 3 1.2.1 Image quality 3 1.2.1.1 Conventional radiographs 3 1.2.1.2 Digital Imaging 8 A Direct digital imaging 10 B Indirect digital imaging 12 C Devices used for the acquisition of a digital image from a conventional radiograph 13 D Digital imaging in dentistry (a summary) 17 1.2.2 Periodontal defects imaging 20 1.2.3 Caries models imaging 27 1.2.4 Endodontic file models imaging 30 1.3 Research problem 36 Chapter two Materials and methods 38 2.1 Section I (Clinical study) 39 2.1.1 Test objects 39 2.1.2 Radiography 41 2.1.2.1 Film Types 41 2.1.2.2 Digitizing 41 2.1.2.3 Randomization 44 2.1.2.4 Image Presentation 44 2.1.2.5 Visual Analog Scale 46 2.1.3 Observation and Interpretation 46 2.1.3.1 Selection of Observers 46 2.1.3.2 Measurements 47 2.2 Section II (In vitro model) 49 2.2.1 Radiography 50 2.2.2 Scanning 53 2.2.3 Measurements 53 2.3 Section III ( Endodontic model section) 56 2.3.1 Test objects 56 2.3.2 Radiography 56 2.3.3 Digitizing 57 2.3.4 Image manipulation 57 2.3.5 Observers 59 2.3.6 Measurements 60 vi Chapter Three Analysis and Results 61 3.1 Section I (Clinical Study) 62 3.2 Section II (Cadaver Study) 81 3.3 Section III ( Endodontic Model) 99 Chapter Four Discussion and Conclus ions 118 Discussion 119 Conclusions 122 References 123 Appendix I Pilot Study Paper 133 Appendix II Volunteer sheet with instructions 136 List of tables Table 1 Analysis of Variance for the various points when case factor is excluded for the clinical study. 72 Table 2 Analysis of Variance for the various points when the interactions between the independent variables are considered for the clinical study. 74 Table 3 Analysis of Variance for all points when two-way interaction between each pair of independent variables is considered for the clinical study. 76 Table 4 Bonferroni P values for the observers' assessment of image quality for each digital resolution. 78 Table 5 Analysis of Variance for the various points for the cadaver study. 90 Table 6 Analysis of Variance for the various points when the interactions between the independent variables are considered for the cadaver study. 92 Table 7 Analysis of Variance for all points when two-way interaction between each pair of independent variables is considered for the cadaver study. 94 Table 8 Bonferroni's Correction test to show where the differences in the measurements are located. 96 Table 9 Analysis if Variance for each point for the endodontic model study 105 Table 10 Analysis of Variance for the distance between the tip of the file and the apex of the root. 106 Table 11 Bonferroni's Correction test to check for differences amongst the different file sizes. 107 Vlll List of figures 1. An example of the selected cases used for the experiments. 40 2. An example of the four different image qualities used in the study. 43 3. Difference between the different image qualities 45 4. An example of a marked case at all the different image qualities 48 5. The specimen selected for the study. 49 6. The specimen fixed on the acrylic plate with pins. 49 7. Test column used to standardize the relationships between x-ray source, object and film. 51 8. The film placed at a constant position. 51 9. The cadaver specimen positioned in the column and held in place by metal pins. 52 10. An example of the images of the cadaver at the four different qualities used 55 11 .An example of a good and bad quality image from the selected ones. 57 12. An example of a good quality image and a bad quality image after the blurring. 58 13. a Scatterplot for the cemento-enamel junction for image quality 1 (600dpi). 65 b Scatterplot for the cemento-enamel junction for image quality 2(300dpi). 65 c Scatterplot for the cemento-enamel junction for image quality 3(150dpi). 66 ix d Scatterplot for the cemento-enamel junction for image quality 4(75dpi). 66 14. a Scatterplot for the alveolar crest for image quality 1 (600dpi). 67 b Scatterplot for the alveolar crest for image quality 2(300dpi). 67 c Scatterplot for the alveolar crest for image quality 3(150dpi). 68 d Scatterplot for the alveolar crest for image quality 4(75dpi). 68 15. a Scatterplot for the pocket depth for image quality 1 (600dpi). 69 b Scatterplot for the pocket depth for image quality 2(300dpi). 69 c Scatterplot for the pocket depth for image quality 3(150dpi)." 70 d Scatterplot for the pocket depth for image quality 4(75dpi). 70 16. Plot of means of difference in mm of points for different qualities. 71 17. Bar graph of means of points against quality. 71 18. Scatter plot of VAS against quality. 79 19. Means of the VAS for each quality. 79 20. a Scatterplot for the cemento-enamel junction for image quality 1 (600dpi). 82 b Scatterplot for the cemento-enamel junction for image quality 2(300dpi) 83 c Scatterplot for the cemento-enamel junction for image quality 3(150dpi). 83 d Scatterplot for the cemento-enamel junction for image quality 4(75dpi). 84 21. a Scatterplot for the alveolar crest for image quality 1 (600dpi). 85 b Scatterplot for the alveolar crest for image quality 2(300dpi). 85 c Scatterplot for the alveolar crest for image quality 3 (150dpi). 86 d Scatterplot for the alveolar crest for image quality 4 (75dpi). 86 22. a Scatterplot for the depth of the pocket for image quality 1 (600dpi). 87 b Scatterplot for the depth of the pocket for image quality 2 (300dpi). 87 c Scatterplot for the depth of the pocket for image quality 3(150dpi). 88 d Scatterplot for the depth of the pocket for image quality 4 (750dpi). 88 23. Plot of the means of points for different qualities 89 24. Bar graph of means of points against quality. 89 25. Scatter plot of VAS against quality. 97 26. Means of the VAS for each quality 97 27. a Bar graph of the actual distance against file size. 100 b Bar graph of the actual distance against file position. 101 c Bar graph of the actual distance against image quality. 101 28. a Bar graph of the actual distance against file size for the second reading. 102 b Bar graph of the actual distance against file position for the second reading. 102 c Bar graph of the actual distance against image quality for the second reading. 103 29. a Bar graph of the actual distance against observers for the first reading. 104 xi b Bar graph of the actual distance against observers for the second reading. 104 30. Scatter plot of the VAS against file size. 108 31. Scatterplot of the VAS against the file positions. 108 32. Scatter plot of the VAS against the image quality. 109 33. Box plot of the VAS against quality. 109 34. Bar graph of the VAS against quality. 110 35. Bar graphs of the VAS against quality for the second reading. 110 Xll Acknowledgements I would like to express my gratitude to my supervisor Dr. Colin Price whose overall guidance and encouragement throughout this thesis has proven invaluable. Primarily, he mentored me throughout the processes of research and literature review. The numerous hours that he spent with me in discussion and in draft corrections were also most helpful, considering his meticulous attention to detail. I especially appreciate his time, effort, and patience without which the writing of this thesis would be incomplete. I would also like to thank my committee members Dr. Alan Hannam, Dr. Ian Matthew, and Dr. Edward Putnins for their time and insight in the restructuring and reorganization of my drafts. Their suggestions and comments added clarity and helped me gain a better understanding of my project. Finally, I wish to take this opportunity to thank my parents for their continual fidelity and encouraging me when in need of support. Without them and their believing in me, I wouldn't have accomplished this thesis. Chapter 1 Introduction Review of the Literature 2 1.1 Introduction In clinical dentistry the use of radiographs as a diagnostic tool has become routine. The presence and extent of hard tissues pathologic conditions can be traced by means of radiographs. In radiology, there has been a need for alternative imaging systems to reduce patients' exposure and increase conventional image quality. With the introduction of computer technology systems in the past years, radiographic imaging has evolved. The new systems allow us to manipulate and alter the images. The obtained images can be both improved and degraded. It is desirable to analyze and highlight factors that affect the quality of both conventional radiographs as well as digital images. It is also essential to evaluate the amount of information lost during the conversion of a conventional radiograph to a digital image, since this application is commonly done. 3 1.2 Literature review Previous work has been done on image quality of both conventional and digital systems whether direct or indirect. The literature review is divided into the major sections: • Image quality • Periodontal defects imaging (major part of the study) • Other models (caries and endodontic models) 1.2.1 Image quality 1.2.1.1 Conventional radiographs Much work has been done on radiographic image quality and its improvement in the cases of conventional radiographs. Within radiology, images are typically evaluated for contrast, resolution, noise, and sharpness. Contrast describes the range of densities in a radiograph and is defined as the difference between dark and light regions of the radiograph. It is therefore the degree of image density differentiation seen in a radiograph. A high contrast radiograph has light and dark areas with fewer grey shades between them, whereas a low contrast radiograph has many grey shades. Resolution limits how well a radiograph can reveal small objects that are close together. It implies the clarity with which outlines or boundaries of individual features are delineated in a radiograph. For digital images it depends on the number of pixels in a given area. For conventional radiographs the analog image has data in a range of different grey shades, and is primarily determined by factors geometric in nature. Those factors depend on penumbral phenomena due to the finite size of the focal spot and the grain size of the emulsion, which is analyzed further. Noise is the appearance of uneven densities in a uniformly exposed film. It is seen in areas of the film as small, localized variations, and it is basically anything that is not part of the real image. Sharpness is the ability of a radiograph to define an edge precisely (White 2000). These features have been studied for conventional radiographs in the past and mentioned as factors affecting image quality in most radiology textbooks. They can also be used to evaluate digital image quality besides conventional radiographic quality. This is a valuable approach for the work done here, which concerns image quality and digitization of conventional radiographs. A 35mm scanner was tested as an intraoral dental radiographic digitizer in 1993, and all those factors are mentioned and explained with regard to digitized images. The same study was expanded to include the effects of alterations of brightness and contrast, which result in optical densities differences (Shrout, Potter et al. 1993; Shrout, Potter ef al. 1993). This is significant since changes in optical densities affect image quality and as a result influenced our decision to scan the clinical cases at their default settings of brightness and contrast and not at a preset value. Loss of image quality A) Blurring Radiographic blurring or image unsharpness is constrained by the system's ability to record information accurately and is caused by image receptor blurring, motion blurring, and geometric blurring. Image receptor blurring: This type of blurring depends on the size of silver grains in the film emulsion. In general, the larger the size of the grains, the worse the sharpness becomes. Image graininess is more noticeable in higher speed films that have larger size of silver grains. The density of the grains depends on the intensity of the applied X-ray beam. Exposure to higher X-ray photon energy tends to aggravate the graininess of the image presented (Barr 1980). When a radiograph is viewed on a light box, the pattern of the different densities of the grains is transferred to the eye and is conceived as different shades of grey. 6 Motion blurring: Any movement of the X-ray source, the patient, or the film itself may cause image unsharpness. The movement of the X-ray source has an effect on the focal spot and, therefore, on the image projection on the film. Film movement can cause the same effect. Geometric blurring: There are a few reports in the literature regarding geometric unsharpness for conventional radiographs. The effect of focal spot size and the radiographic accuracy was investigated, and it was concluded that the annual focal spot size measurement, which is part of the dental radiographic quality program, is not necessary (Platin, Mauriello et al. 1996). This finding is debatable since most textbooks mention focal spot size as a significant factor affecting the image clarity (Barr 1980; White 2000). An important reference to the work done here is the experiments done by Dr Radan (Radan 1999) for her Master's thesis. She evaluated the effects of geometric unsharpness in the case of endodontic files and concluded that geometric unsharpness is an overestimated factor when regarding digitization and display of intraoral radiographs in the model used. In relation to the above, the factors that improve image quality in the case of conventional radiographs should also be mentioned. B) Minimization of loss of image clarity Three methods minimize the geometric unsharpness and improve the quality of the radiographs (White 2000). Those include: Using as small an effective focal spot as practically possible: The use of a small focal spot results in increased image sharpness and resolution. Using a large focal spot gives a partial shadow at the edge of the image, which is not sharp due to the superimposition of various sections of the X-ray beam. There are multiple adjacent images giving the appearance of unsharpness. The resulting unsharp area obtained is referred to as penumbra. Increasing the distance between the focal spot and the object: This is primarily achieved with a longer open-ended cylinder. Its effect is to reduce the penumbra. Decreasing the distance between the object and the film: This also reduces the penumbra and gives a clearer image. 8 1.2.1.2 Digital Imaging There has been a need for alternative imaging systems in order to reduce patients' exposure and increase conventional image quality. With the introduction of computer technology systems in the past years, radiographic imaging has improved. Digital systems allow us to manipulate and alter the images in many ways that were previously impossible. To provide adequate diagnostic information a digital image, whether acquired by the direct or indirect method, must have adequate grey level resolution. The human eye can distinguish roughly 100 grey levels (van der Stelt 2000). Consequently, the 256 grey levels most commonly used is sufficient. However, digital imaging is not just a procedure to simulate or replace film based imaging. It is a new way of diagnostic imaging and opens up possibilities, which are not available with film based imaging. That is, images can be manipulated and altered to either improve or degrade the image. Computer software exists that can improve image viewing regardless of whether the image is direct digital or indirect digital. This does not mean though that digital images are superior to conventional film radiographs, but only that they are more flexible in their manipulation. Comparison of the sensitometric properties of digital intraoral systems were made (Araki, Endo et al. 2000). Variations were found in both resolution and 9 latitude among the systems whether Charge Coupled Device (CCD) or Photostimulable Phosphor Plate (PSP) based. Therefore the specific properties of each system should be considered to avoid errors and image repetitions. Image quality has been compared between conventional film and digitally acquired images where a phantom model was used to quantitatively measure differences in image quality (Yoshiura, Kawazu et al. 1999; Yoshiura, Kawazu et al. 1999). Further investigation is recommended on the particular phantom since it approximated clinical conditions and helped quantitatively evaluate image quality. Finally, work has been done on image improvement by means of computer systems where mention is given to the extension of the applications of digital imaging and the future of dental imaging (Wenzel 2000). 10 A. Direct Digital Imaging This method uses standard radiographic equipment (for exposure of the film to the x-rays) with a receptor instead of conventional intraoral films. The image production in these systems includes the following steps: • Acquisition of data by an electronic sensor (receptor). • Presentation of the data to the computer as analog information. • Transformation of data by an analog-to-digital converter to a binary code that a computer can recognize. • Presentation of the data on the screen of the computer (the information is made visible to the user). The digital receptors that are available and most commonly used are Charge Coupled Device (CCD) or Photostimulable Phosphor Plate (PSP) receptors. Charge coupled devices (CCD) receptors: The radiation intensity of the source is measured directly by an electronic device, which has many light-sensitive elements. Upon stimulus the output of the elements is converted into an electric signal and then transferred to a computer where it is digitized by the use of an analog-to-digital converter. Then X-ray photons are converted into light photons, which increase the detector's efficacy. The size of the electronic sensors is often smaller than a size 2 conventional film. Due to the smaller receptor size, which is present in most systems, the smaller area to irradiate and the higher absorbance 11 of the photons aimed at them, the radiation amount given to the patient is significantly decreased. Photostimulable Phosphor Plate(PSP) receptors: PS plates have a semi-direct digital image. The X-ray photons excite electrons on the phosphor crystals, which results in a latent image on the photostimulable phosphor plates. A laser beam then scans the plate, and the electrons return to the original energy level. A photo multiplier device captures the visible light energy liberated from this process. The output of the device is converted into picture elements (pixels), each with a value that contains the image information. Pixels are a two-dimensional array of sensors having potentially 256 grey shades but one value according to the information of the binary system that they include. 12 B. Indirect Digital Imaging Indirect imaging systems are still most widely accepted by dentists despite the disadvantages when compared to digital systems. Conventional film systems absorb only part of the photons aimed at them, provide a static image that cannot be manipulated or changed, require processing during which errors can occur, require a relatively high radiation dose, and are sensitive to variations in exposure and time. Dentists frequently need to exchange information and obtain advice on radiographic image interpretation. Images may be scanned and sent by e-mail or by another digital format. The scanning process is as follows: The original image, which is recorded on a radiographic film, is an analog image that contains continuous data in a range of grey shades. The information in this analog image is converted to discrete information based on the binary system in which two digits (0 and 1) are used to represent information. This is done by means of computer systems. The most common unit of information in computer systems is 8 bits long and is called a byte. One byte can represent 256 shades of grey ranging from 0 to 255 in value. After computer manipulation, the assigned numbers that represent different levels of grey are converted into an image. This new digitized image consists of different pixels, each of which is assigned its own shade of grey according to the 256 system's value. The diagnostic information is included in the original image. 13 Since the converted image is a second generation one, further errors can be introduced during the process. This can lead to information loss in the image produced. C. Devices used for the acquisition of a digital image from a conventional radiograph: Conventional radiographs digitized using a Flatbed Scanner and a transparency adapter. In most cases the spatial resolution is chosen so that the diagnostic details are preserved in the newly formed digital image (Chen, Hollender et al. 1997). Typical resolutions used for digitizing radiographs are 300 or 600 dots per inch (dpi). Conventional radiographs digitized using a charge coupled device camera. This method is quite similar to the previous method, but instead of the scanner a charge coupled device (CCD) camera is used. The quality of the resulting image depends on the software used to control the camera settings. Key information may be lost during the conversion of a radiograph to a digital image. We do not know the image resolution at which there is"no diagnostically significant information loss. File size is another factor, but modern computer systems have large hard drives which can be further increased, although data acquisition and transfer may still cause limitations. Ideally, we seek a resolution at which there is 14 no significant diagnostic information loss without having an image of a very large file size. There is much reference in the literature to image file size and compression when digital images are considered. Although with the newer computer systems the problem of image storage has been eased, it is still significant when a large amount of information is considered. Image compression was studied with a caries model for the measuring of the effect of compression on accuracy (Wenzel, Borg et al. 1995). Differences were found in both compression size and file size of the resulting images, but the results were similar in that they detected loss of diagnostic information by the application of compression algorithms. The effects of image magnification of digitized bitewing radiographs and its impact on caries detection was measured. Smaller zooming was found to be superior to larger magnifications such as x 18 or x 30. Limitations for various diagnostic tasks were noted, and guidelines for the appropriate magnification should therefore be given. This is quite significant since most people would think that image magnification would increase the visibility of details. Measurements on endodontic file location to determine the effects of image size reduction of the images have also been used. Size reduction was found to possibly cause less detectability and loss of diagnostic information when the nearest neighbour interpolation was used to simulate zoom out. This means that when the nearest pixel is removed from an area to zoom out, there was information loss. Previously, the same researchers had performed a similar study on the effects of altering image size. Similar results were obtained, and therefore they concluded that diagnostic information is lost when images are reduced in size (Versteeg C. H., SanderinkG. C. era/. 1997; Versteeg, Sanderink er al. 1998). Joint Photographic Experts Group (JPEG) compression has also been investigated for the case of intraoral radiographs. The JPEG format uses a standard compression algorithm. However, its applications resulted in some quality loss, and it was confirmed that smaller pixel size was required to obtain sufficient quality (Yuasa, Ariji et al. 1999). The researchers of the study recommended that intraoral radiographs should be digitized at a sampling rate of 400 dpi or higher. In the study, there seems to be no difference between resolutions of 400 and 600 dpi although in the figures they supply the difference between 300 and 400 dpi, which seems to be small. The mean grey level had been used to measure changes in region of interest densities together with two more statistics (mean grey level, largest cluster of pixels exhibiting loss of density and the total number of pixels in the region of interest exhibiting loss of density) (Cohen and Roddy 1995). The above region of interest changes were compared to pixel size, and it was concluded that selection of the appropriate statistics (in reference to the factors the word 16 statistics is used) for identifying changes between radiographs was crucial. All factors offered advantages under some conditions, with the mean grey level being the most powerful. They suggest further research on the factor used according to the different cases involved. The detection of small changes in bone mass when using a digital system was used to evaluate image quality in digital images (Chen, Hollender et al. 1997). The image produced was affected significantly by alterations in the mass. As we see from the above examples, most researchers concluded that alterations in image file size led to changes in image quality and therefore to the diagnostic accuracy of the digital radiographs. The image quality was decreased when compression was used and so was the ability of readers to detect details. The ability of dentists to identify manipulated images has also been evaluated. It was shown that image manipulation, even when altering the diagnostic content of a radiograph, was unlikely to be detected by dentists (Visser and Kruger 1997). The idea of image manipulation opens up a discussion with respect to ethics. 17 D. Digital imaging in dentistry (a summary) The factors that determine the accuracy of digitized conventional radiographs by using three different digitizers were studied. The best diagnostic accuracy was in images scanned with a drum scanner and not the laser scanner or the TV camera. The image size was also examined and was found that digital images with a pixel size of 100pm and 32 grey levels (equivalent to 256dpi) were acceptable for diagnostic purposes (Ohki, Okano et al. 1994). This is important considering that most of the scanning of images is still performed with conventional scanners. A similar study comparing different digital systems for the detection of small mass changes and appearance of burn-out effects and blooming phenomena was done, and it was concluded that higher image quality was achieved over a wider exposure range with the PSP system than with either the film or the CCD system they compared (Borg and Grondahl 1996). Altering the exposure in PSP systems and conventional film radiography was studied, and a direct linear relationship between exposure and digital grey levels was found over the narrow range of exposures studied for PSP images, but not for conventional radiographs (Hildebolt, Pilgram et al. 1998). In a similar study, which compared exposure effects in PSP systems and digital films, it was concluded that PSP had a wider dynamic range than conventional film and a better low contrast detectability (Huda, Rill et al. 1997). Both papers are interesting because PSP systems seemed to perform better than conventional film in regard to their characteristics; their diagnostic performance was similar but not superior. Also, in 1996 linear measurements from PSP systems and conventional radiographs were compared (Conover, Hildebolt et al. 1996). No significant differences between linear measurements taken from both modalities were found, in agreement with most researchers in the literature. Another factor whose effect on image quality was studied is background lighting (Cederberg, Frederiksen et al. 1998). Lighting conditions did not have any effect on the ability of observers to detect artificially developed approximal lesions. A large study involving four digital systems and two film systems was done (Yoshiura, Kawazu et al. 1999) to compare the exposure conditions. All digital systems, except one, showed lower optimum exposures than E speed film. In the case of contrast enhancement, all digital systems, except one, showed visibility superior to film systems. They concluded that digital systems perform equally well as conventional film, although when properly used, their performance could exceed that of film. This study included more systems than most studies have used, and therefore its findings are valuable. In conclusion, we can say that although digital systems are being used in larger scales, their diagnostic performance is similar to that of film. Although there are advantages in their usage such as exposure reduction, easier handling, and ability to manipulate the images, there are still factors that have to be examined. 19 These factors include number of retakes and the legal implications of images that may have been altered. 20 1.2.2 Periodontal defects imaging Studies on digital imaging of periodontal defects go back to 1983, when a paper used subtraction radiography to better image periodontal bone lesions (Grondahl and Grondahl 1983). Most of the work regarding periodontal defects and digital imaging has been on subtraction radiography. Various researchers have tried to improve the subtraction process by altering image quality or the angles at which the images are taken. Different methods and studies have been done, most of which discuss image improvement and how it can be achieved. They do not, however, apply it to their results most of the time by further investigations. Pattern recognition in the diagnosis of periodontal bone defects has been investigated by mean density comparison (van der Stelt, van der Linden et al. 1985; van der Stelt, van der Linden et al. 1985), and they concluded that by computer-aided systems an objective and reproducible assessment of the parts of the radiographic image that were considered to be of interest for the interpretation of bone defects could be carried out. The detectability of smaller in vitro periodontal defects was made easier with quantitative digital subtraction radiography when compared to other conventional techniques (Janssen, van Palenstein Helderman era/. 1989). Therefore, smaller 21 bone changes were made easier to detect on the quantitatively subtracted images when compared to conventional radiographs or to simple subtracted radiographs. There was no effect, however, by alterations of the exposure factors on the observation threshold of the images. A correction method for the difference in the optical densities of two consecutive radiographs was applied and tested in terms of lesion extent and average optical density in the region of interest (Okano, Ohki et al. 1988). By this method, the images to be subtracted were made more consistent and comparable by altering their contrast and brightness. This has not been verified by further studies. In a review of periodontal radiography, emphasis was given in the development of digital imaging systems for dental radiography and the effectiveness of their applications in the cases of periodontal defects (Bragger 1988). Again, even in this review paper, most images are those related to subtraction radiography. It is mentioned that the application of digital imaging methods offers objective quantitative and non-invasive ways to obtain additional diagnostic information from conventional radiographs. The conclusion drawn is that, although not yet routinely used in the clinic, promising and exciting progress in digital imaging of dental and periodontal tissues have been reported. Digital subtraction radiography enables smaller changes in alveolar bone density to be detected and quantified, automatically providing interpretation variation. A comparison on the accuracy of detecting bone loss was done (Stassinakis, Bragger et al. 1995) comparing conventional radiographs and subtracted direct digital images. The diagnostic characteristics of the digital systems were improved with subtraction, but the problem of positioning the patient and standardizing the exposures still remained. One study found that angulation changes up to 10° did not have an effect in the diagnosis of the bone changes (Verdonschot, Sanders et al. 1991). This means that misangulation errors did not make much difference in the diagnosis of the disease and the measurements performed, contrary to what many claimed. What should be noted is that this study did not use a conventional subtraction technique but a new image analysis system. Most researchers, however, believe that correct angulation is crucial for the proper application of the subtraction technique. Despite the potential advances, the quality of direct digital images is inferior to those obtained on conventional film, and there is a lack of controlled studies taking into account the different features of direct digital imaging. A review article (Gutteridge 1995) emphasized the use of radiographic techniques in the diagnosis and management of periodontal disease. The importance of digital systems and their value in the diagnosis and management of periodontal disease is discussed. A significant finding is that the best resolution and repeatability are still given by bitewing radiographs in agreement with most authors. 23 A digital device was used to quantitatively analyze periodontal defects in an in vitro model with the use of subtraction radiography (Young, Chaibi et al. 1996). The findings were that the subtraction method frequently underestimated the lesions' sizes by various percentages according to the type of lesion present. Therefore, they concluded that the radiovisiography (RVG) imaging system used is not sufficiently accurate to establish absolute determinations of the bone loss, but is still clinically useful to establish relative changes in bone loss. It was suggested that caution should be taken when interpreting a given change in calculated bone loss. In another review paper, a comparison between the available radiographic methods for detecting progressive alveolar bone loss was included (Jeffcoat 1992). Comparison was made between digital subtraction radiology and the clinical probing depth of the pockets. In 81.2% of the cases, there was concordance between the clinical pocket depth measurement and the radiographic assessment of the defect depth which constitutes a very high degree of concordance. A study was done in 1995 to measure the accuracy of observers' diagnostic performance in locating bone levels using periapical radiographs (Gurgan, Grondahl et al. 1995). The major finding was a lack of accuracy and precision between the observers. This is significant considering that most papers do not mention observer variation. The lack of observer accuracy increases 24 respectively to pocket depth, although there is a general tendency to underestimate bone loss by nearly all observers when using conventional periapical radiographs. More recently, a comparison was made between colour coded radiographs and simple subtraction images (Shi, Eklund et al. 1999), and it was concluded that colour coding of radiographic differences by means of image addition can be used as an alternative to the subtraction technique for the evaluation of periodontal defects with a better inter-observer agreement. Their results could be explained by the enhancement of the difference that colour coding gives. However, radiographs to be compared should be exposed with a reproducible technique making registration possible. A valuable point presented in this particular study is that colour coding radiographs not only enhances the differences between the originals by colouring them, but also retains the information that is common to them. This is an advantage compared to subtraction technique. New subtraction methods and techniques have been attempted for periodontal bone detection (Nummikoski, Steffensen et al. 2000). A technique enhancing the accuracy of alveolar crestal bone detection, when compared to conventional film imaging, was developed and was considered for use in clinical practice. However, this technique failed to provide significant difference when compared to the conventional subtraction technique. 25 Besides subtraction radiography for evaluation of periodontal defects, other aspects of periodontal digital imaging have been considered and reviewed. A study was performed (Borg, Grondahl etal. 1997) comparing digital images from a CCD based and a PSP based machine in the cases of periodontal defects in mandibular molars. Error in bone loss evaluation was found, but smaller than that reported for film radiographs in the past. Therefore, it was concluded that digital radiographs are comparable to film-based radiographs for measurements of buccal bone loss, but they require less exposure for producing the same result as the film-based images. A study on the reliability of linear alveolar bone loss measurements of mandibular posterior teeth from digitized bitewing radiographs was done (Hildebolt, Pilgram et al. 1998) in which it was concluded that inter-observer variation was greater than intra-observer, but despite that variation, an overall difference of 0.3mm could be detected on average in the cases of periodontal defects by all observers in digitized images. The periodontitis detection efficacy of film and digital images was studied, and the digital imaging system examined was not significantly different from regular film for crestal bone evaluation (Nair, Ludlow et al. 1998). Significant differences between observers and their readings were found, but no difference between modalities. 26 The use of different filters was evaluated for the digital display of interproximal bone loss (Eickholz, Riess et al. 1999). The filters used were of basic image processing models, such as enhancement of grey level differences and histogram correction. No difference between the various filters used and the original unchanged digitized images was found, and all radiographic assessments and modalities came close to the surgical gold standard. The digitization and display of intraoral films was evaluated (Attaelmanan, Borg et al. 2000). They tested for the optimal parameters for the conversion of an intraoral radiograph to a digital one and its optimal display. They concluded that the best results were provided by the use of masked films, at a resolution of 400ppi. The display settings were optimal with 256 colours and a screen size of 1152x 864 pixels. What was important in this study was the finding that image quality of higher than 600ppi did not get as high of a rating by the observers. On the contrary, images of that quality were even rejected by many observers. This was explained by the belief that images of that size provide limitations upon viewing and cannot be fully viewed on the monitor. Also in a review paper, a description and analysis of the different techniques available for periodontal defects radiology is done (Hausmann 2000). Reference is given to the fact that although digitization of the images is not necessary, it enhances the measurements taken in periodontal cases. CT is mentioned as an alternative technique in periodontal radiology. 27 1.2.3 Caries models The caries model is not part of this study, but has been used extensively in the past for the evaluation of image quality in both film-based and digital systems. It is therefore appropriate to review some of the relevant literature. An in vitro caries model was used to evaluate various direct digital systems (Wenzel, Borg er al. 1995). They concluded that their performance was similar whether CCD or PSP receptors based. A caries model was used to compare conventional films to direct digital PSP images, and no significant differences were found between the two systems (Nielsen, Hoernoe etal. 1996). Considering the emphasis given on digital systems lately, this result is significant. The same year, a newly developed digital system was tested and compared to conventional film radiography for caries detection and image magnification in relation to observer performance (Svanaes, Moystad et al. 1996). No significant difference between the use of digital system and conventional radiographs was found. However, magnification increased the diagnostic accuracy of the observers when digital images were used. In 1997, Price and Ergul (Price and Ergul 1997) used an in vitro caries model to compare conventional film to a CCD based digital system. Their conclusions 28 were that the film was superior to the digital system for the interpretation of caries, but the magnitude of difference was small. Radiovisiography (RVG) systems showed differences in enamel density when compared to healthy enamel (Forner, Llena et al. 1999). Therefore, this method was potentially useful for the diagnosis of approximal caries. The evidence on the diagnostic efficacy of digital systems for the detection and evaluation of caries lesions was reviewed (Wenzel 1998). The conclusions were that the digital systems perform equally well, but there has been very little work done on their utilization and their clinical application. Attention was directed to the absence of literature regarding cross infection controls and the enhancement of the digital images in treatment planning. A comparison on the consistency and observer agreement for caries detection, both in the case of direct conventional radiography and digital systems, has also been investigated (Naitoh, Yuasa et al. 1998). Digital systems did not result in any deterioration in observer agreement and should not limit the diagnosis of caries when used. A review paper on digital imaging for dental caries was published in which the whole process of digital imaging was explained and analyzed from the taking of the radiographs, their storage and display to patients acceptance, number of 29 retakes, and their precision (Wenzel 2000). She concluded that current digital systems provided a diagnostic outcome as precise as conventional film, but more studies were needed to explore the difficulties met by dentists when converting an image from analog to digital and when interpreting a digital image. 30 1.2.4 Endodontic file models A considerable number of papers regarding digital imaging have used endodontic file models due to their accuracy and precision. Since part of the experiments we performed included use of those models, the literature about it was relevant. Our experiments using the endodontic model were based on previous research done by Dr Radan for her MSc thesis (Radan 1999). She compared the effects of digital and geometric unsharpness using a range of different file sizes, file positions, and image qualities. She found that there was no difference in diagnostic yield between digital resolutions of 600 and 300 dpi regardless of the file sizes used. This is important since the sizes used were between file sizes 6 and 15, which are relatively small. Therefore she concluded that there is no need for the quest for digital resolutions greater than 300 dpi. Studies using endodontic file models in digital imaging go back to 1990 when radiovisiography was compared to conventional film for imaging root canals in vitro (Shearer, Horner et al. 1990). Radiovisiography was found to perform similarly to conventional film for root canal imaging. The study was continued in 1991 and similar results were obtained (Shearer, Horner etal. 1991). Most of the work on endodontic files has been done since 1994 when a study on image quality of direct digital systems in assessing root canal length was done 31 (Sanderink, Huiskens et al. 1994). When endodontic files of size 15 were used in root canals the results obtained by digital systems were comparable to conventional film images, while with the use of file size 10, film was superior to digital systems. A study, which examined the diagnostic accuracy of direct digital radiography for the detection of periapical bone lesions, found that processing of digital images had a limited effect on the diagnostic accuracy of the lesion. Processing included alterations in contrast and brightness for the detection of periapical lesions. The study was expanded further to include a comparison between digital systems and E speed film for the detection of periapical bone lesions. The systems used were comparable to conventional film (Kullendorff and Nilsson 1996; Kullendorff, Nilsson et al. 1996). A paper on estimating distances on direct digital images and conventional radiographs was published that concluded that measurements on digital images were comparable to those of conventional radiographs and even facilitated by experience, which is an advantage for their usage (Versteeg K. H., Sanderink G. C. et al. 1997). The means of scanning radiographs and its effect on endodontic file detection with both conventional film and scanned images has also been examined. They found that the images scanned at five different settings were inferior to the original radiograph and there was even a high agreement between the evaluators (Fuge, Stuck et al. 1998). These results do not support the idea of scanned 32 images, but the findings may possibly be explained by the fact that the files created were of very small sizes, and not easy to trace even on conventional images. The same year periapical artificial lesions were compared between conventional radiographs, direct digital and telephonically transmitted images and no statistical difference between the measurements in all cases were found (Mistak, Loushine era/. 1998). Comparing direct digital and conventional radiography for the detection of artificial periapical lesions concluded that the two imaging modalities were comparable and that digital imaging did not enhance lesion detectability (Barbat and Messer 1998). When PSP radiography and conventional radiography were compared for the detection of pathologic periradicular bone loss in cadavers it was concluded that the digital systems were comparable to the images obtained by conventional radiography (Holtzmann DJ 1998). A similar study was done (Farman, Avant et al. 1998) using a CCD system and the results were slightly different in that CCD systems were preferred to film-based radiographs for measuring periapical lesions dimensions. However it is mentioned in the paper that diagnostically there is no difference between the two modalities. It is in the case of subtle early density changes that occur as an initial stage in the disease that the effectiveness of the device was shown. 33 Another comparison between direct digital systems and conventional films for the estimation of canal length in the case of curved canals was done. The results were that all techniques resulted in similar measurements amongst them, but that the total measurements were different from the actual true canal length, which could be explained by the canal curvatures (Burger, Mork et al. 1999). Besides the comparison of the different systems used, studies have been done using the endodontic model to investigate benefits of image enhancement for the assessment of periradicular lesion dimensions (Scarfe, Norton et al. 1995). What was found was that image enhancement had little value in the estimation of the lesions' dimensions. Another factor examined using an endodontic model in digital imaging is the number of retakes taken with digital imaging systems. A paper on the evaluation of periapical radiography with a CCD system in comparison to the standards of conventional periapical radiography found something quite different to other studies (Versteeg, Sanderink et al. 1998), that periapical radiography using a CCD leads to more errors and thus more retakes than conventional film. This is very interesting since one of the advantages of digital imaging, reduced exposure, could be questioned with the increase in the number of retakes. What should be noted in the literature is that most authors agree that digital systems are just as effective as conventional films for the detection of most 34 lesions. Although there is a small difference in opinion, most researchers obtained similar results when comparing conventional films to various digital systems and the digital systems amongst them. There is no additional information given by the images in case of enhancement, nor is the observers' performance increased by it. This is valuable for the section of this study involving the blurring of images, a feature not studies before. Conclusively we can say that digital imaging has the potential for improving diagnostic accuracy and making quantitative diagnoses. In addition it makes image archiving and communication systems feasible. However it is necessary in the first instance to establish that the accuracy of digital images is sufficient for a specific diagnostic purpose. The diagnostic accuracy of digital imaging, whether direct or indirect, has been investigated in dental radiographs in terms of the detectability of various defects. A number of studies have failed to show any clear advantage of computer based image processing and analysis over conventional radiographs for improving the detectability of most defects. However the fact that a digitized image is a secondary image remains. As a secondary image it can only include in the best case an equivalent amount of information when compared to the original film image. The digitization process is especially important so as not to loose small changes in radiographic densities that may exist in the original image. 35 In this study we have performed a further investigation into the diagnostic accuracy of digitized intraoral radiographs, to clarify the effects of image quality on the amount of information loss that might occur during the process. We used four different image scanning qualities and one periodontal model to measure and evaluate those effects. In addition to image quality, the feature of blurring was evaluated based on images from previous work. The feature was selected since during the initial study we discovered that PowerPoint Program alters image quality and appearance of the image upon entry 36 1.3 Research problem A scanned image is a second generation and can only be as good as, or worse than, the original image. It cannot include additional information that was not part of the original image. The aim of this study was to measure the amount of information lost during scanning of conventional radiographs at different resolutions, and to detect whether this loss was significant. Similar studies have been published, but none focused on image quality and periodontal defects displayed on indirect digital images. We thought this merited further investigation. Three models were used to study and evaluate the information loss during the scanning of the particular defects. The first model used was based on periodontal infrabony defects in cases obtained from clinic files at the Faculty of Dentistry of the University of British Columbia. In those clinical cases observers marked certain points. The points' coordinates were measured and compared among the different image qualities. Our null hypothesis in this section was that higher digital resolutions have better diagnostic quality for measurements performed in the case of infrabony defects. This model was used to resolve the problem of image scanning and periodontal defects which has not been extensively studied in the past. The second model used a cadaver specimen. In the specimen we created infrabony defects of certain known depths. Radiographs were obtained and 37 digitized at different resolutions with the same contrast and brightness. The observers then marked the perceived origin and depth of the defects. The results were used as a basis for comparison to the periodontal model from the clinically obtained cases. This was done since in the case of the clinical study one could argue having different exposure conditions that affect image quality. This way we avoided that by comparing the clinical results to known in vitro defects we created. Our null hypothesis in this section was that higher digital resolutions have better diagnostic quality for measurements performed in the case of artificially created infrabony defects, as we supposed similarly for the clinical cases. Finally, archival material of an endodontic file model used in the past by a former graduate student in the Faculty (Radan 1999) was used. This model was used for further image manipulation since, during the previous research based on it, differences in qualities and observer performance were found. For our study, images were selected of good (600 dpi) and of poor (75 dpi) quality, then blurred using the mean 7 x 7 Kernel of NIH Image, (taking the mean of every seven pixels in both directions and replacing their values with it). The feature was selected since during the initial study we discovered that PowerPoint program alters image quality and appearance of the image upon entry. We used both the blurred and non-blurred films for comparison. Our null hypothesis was that there should be no effect on image quality by the algorithm the program uses, which we had no control over. Chapter 2 Materials and Methods 39 2.1 SECTION I (Clinical study) 2.1.1 Test objects Radiographs from the University of British Columbia dental clinic files were selected as test objects. The radiographs were used after approval was granted from the Clinical Research Ethics Board of the University of British Columbia. The radiographs were all of posterior teeth of both the upper and lower jaw, and they were selected from the clinic files according to certain criteria. The criteria for the selection of the cases were the following: Image quality. The radiographs had to be clear, easy to interpret, and of good processing quality. Periodontal defects. The cases all had to have infrabony defects in a posterior teeth area without any particular additional features in the selected defect (e.g., Endo perio combined defects were rejected). All posterior teeth present. To prevent any pattern matching by the observers, we selected cases that had all posterior teeth present. All teeth in the radiographs had to be without any restorations (intact). That way, we tried to eliminate additional pattern matching of the observers, which could be caused by characteristic restorations or any additional features of the teeth. A total of 20 cases were selected from the clinic files after going through all the files that were present in the clinics. The number selected was large enough to provide precise and reliable results, since the total number of images would 80 after the scanning at the four selected different qualities. Fig 1 An example of the selected cases used for the experiments. 2.1.2 Radiography Having selected clinical cases that were taken in the past, we had no knowledge of the radiographic conditions under which the films were obtained. Knowing that exposure factors make relatively little difference on image quality, provided that the overall image quality is reasonable, our selection was based on the overall quality of the cases (Manson-Hing 1979; White 2000). Furthermore, adding the radiographic variables would have rendered the model statistically unmanageable. 2.1.2.1 Film types The radiographs selected comprised 17 Ektaspeed Plus films and 3 Ultra-speed films (Kodak Canada Inc, Toronto, Ontario, Canada). All the films were size 2. Adding the 3 D-speed films to the Ektaspeed ones would not alter the results since we are testing for scanning and image quality interrelationships and those factors are not affected by film type, but by the scanning parameters and the quality of the radiographs. The selected cases were chosen to enable the digital qualities to be compared. They were the best cases for our model and were examined and scanned regardless of film type. 2.1.2.2 Digitizing The 20 radiographs were scanned with a Hewlett Packard 3c flatbed scanner (Hewlett Packard Company, Greeley, CO, USA) as sharp black and white 8-bit images. The resolutions selected for the scanning of the images were 75, 150, 42 300, and 600 dpi. This produced 4 different quality images of each case. The different qualities were all produced from the same original radiograph, thus eliminating radiographic variables from the statistical analysis. During the scanning, the radiographs were placed on the scanner in such a way that the image obtained would be the one as viewed when placed in a mount. The images, being of different exposures and qualities, were scanned at the default settings for each, therefore providing the best image quality according to the scanner's software for each case. This is usually the case for scanning images since in most cases the default settings are accepted as the best for the image selected. The size of the image was kept stable at 4 x 3 cm so that all images would be of the same size when opened, although they would be of different digital image file sizes (due to the different resolutions). a Resolution of 600 dpi b Resolution of 300 dpi c Resolution of 150 dpi d Resolution of 75 dpi Fig 2 An example of the four different image qualities used in the study. 2.1.2.3 Randomization The final number of the 80 images consisted of 4 different image qualities of the 20 selected cases. They were numbered according to image and quality and entered in an Excel Program Worksheet (Excel, Microsoft Corporation, Redmond, WA, USA). The images were first randomized and then organized in two groups of 40 images according to the random order. 2.1.2.4 Image Presentation All of the images were opened in NIH Image (Wayne Rasband, National Institutes of Health, USA) and viewed. Then they were inserted into the older version of PowerPoint Program (Microsoft Corporation, Redmond, WA, USA) in the random order. The more recent version of PowerPoint smoothed the edges upon insertion of the image. This PowerPoint feature was not desired since we wanted to maintain the original image quality. We did not want to have additional parameters involved such as image blurring, especially since there was no control of the action, which was performed automatically by the program. During image insertion the images went to the middle of the slide. They were then dragged to the top left corner and increased in size to fill in the screen. All the images were manipulated and dragged to fill the slide in the exact same way, which made them comparable for the purpose of the study. They were numbered to aid identification and were then masked. Masking has been shown to improve performance of the observers (Attaelmanan, Borg era/. 2000). 45 Finally, before presentation to the observers, a visual analog scale was added to each slide. What should be noted is that the magnified images were presented to the observers on a 72 dpi monitor. This does not mean that the images' resolutions were 72 dpi, but rather that an image scanned at a resolution of 600 dpi was presented on a monitor displaying 72 dpi. Therefore, each image pixel used a certain number of display pixels, depending on the scanned resolution. When an image is scanned at high resolution for each of its pixels, it uses only a small number of display pixels while, if scanned at low resolutions, it uses a larger number of display pixels. The difference of images on the screen is given by the following figure where the image subtraction shows the difference in pixels amongst the highest quality image (600 dpi) and the second best quality (300 dpi) as well as the worst quality (75 dpi). a Resolution of 600 dpi-300 dpi. b Resolution of 600 dpi-75 dpi. Fig 3. Difference between the different image qualities. 46 2.1.2.5 Visual Analog Scale A visual analog scale was created in SuperPaint program (Silicon Beach Software Inc, San Diego, CA, USA). The visual analog scale consisted of one horizontal line, which joined at each end to a vertical line. The two ends were assigned the values of 0 and 10. Zero represented the poor image quality while 10 was used as a measurement of a high quality image (a perfectly clear and sharp image). This visual analog scale was copied and pasted in each slide at the top part of each in identical positions so that the data collection in all the scales would be comparable (Figure 4a, page 48). 2.1.3 Observation and interpretation 2.1.3.1 Selection of observers One group of observers was asked to perform all the required measurements. The group was comprised of eight volunteer third year dental students. All the observers viewed the images under identical conditions on the computer screen in a room with subdued lighting. There were no time limits set on the observers and although the observer's distance from the screen was allowed to differ, they were not permitted to alter any of the screen brightness, contrast, or software features. They were given written instructions of the objectives and a list of the selected pockets in each slide (appendix 2). They were asked to mark selected points with the use of the marking facility of the PowerPoint program. 47 The points that they identified were the following: Cemento-enamel junction (CEJ): used as a reference point and the point expected to have the least variation in the measurements. Superior aspect of the alveolar crest bone: adjacent to the tooth and pocket of interest. Most inferior aspect of the periodontal pocket: adjacent to the tooth and pocket of interest. In addition to those three points, they were asked to mark a point on the visual analog scale to give a subjective opinion of image quality (figure 4b page 48). Before the observers marked the selected points on the images, they were calibrated with an example of a poor and a good quality image, and measurement performance was demonstrated on those. That way, they knew how to mark the images and what was meant by the factor of difference in quality. They therefore knew what to expect and how to perform their task of marking. The marked images were captured to memory by appropriate software after each marking. 2.1.3.2 Measurements The marked images were opened in NIH Image program. An example of the marked images is given below. All the points that were selected by the observers were four pixels in size, and the top left pixel was selected for the determination of the required measurements so that the results obtained would 48 be comparable. Out of all the points selected, the x and y coordinates were measured in pixels and further analysis of the data set was performed. Resolution of 600 dpi. Resolution of 300 dpi. Resolution of 150 dpi. Resolution of 75 dpi. Fig 4a An example of the visual analog scale after the marking for all four qualities. 0 -, 0 0 10 0 o 10 a Resolution of 600 dpi. b Resolution of 300 dpi c Resolution of 150 dpi. d Resolution of 75 dpi. Fig 4b. An example of a marked infrabony defect at all different image qualities. 49 2.2 SECTION II (In vitro model) A fresh cadaver specimen of a hemi-mandible was selected from the Anatomy Department of the University of British Columbia. All the teeth were present. Fig 5. The specimen selected for the study. The specimen was fixed on an acrylic plate with metal pins. Fig 6. The specimen fixed on the acrylic plate with pins. The area between the two premolars was selected to create artificial infrabony periodontal defects. A flap was raised with a surgical blade (number 10) on the buccal surface of the specimen, and different depth lesions were created with a round size 6 bur. Initially, there was no infrabony defect present. The lesions were created buccally between the two premolars. The lesions were of incremental depths of 2, 4, 6mm. After the 6mm pocket on the buccal side, lesions were created on the lingual side as well. These were also of 2, 4, 6mm increments. At each depth, images were taken at three different exposures. The factors under which the images were taken were kept constant. Therefore, we ended with seven different depths of the same infrabony defect (including the original images taken before the defects were created). 2.2.1 Radiography A General Electric GE 100 x-ray unit was used to expose the films (General Electric Company, Milwaukee, USA). The unit was secured in a reproducible and stable position at the top of a test column. The test column consisted of three vertical rods that were stable and aluminum platforms that were adjustable and with reproducible positions. On the side, there was a vertical steel ruler with an accuracy of 1mm. The distance between the source and the film was kept constant at all times at 1 meter. Fig 7. Test column used to standardize the relationships between x-ray source, object, and film. Fig 8. The film placed at a constant position. 52 In addition, the positions of the film on the plate and the specimen were kept constant. These relationships were identical for all exposures with all different pocket depths. Therefore, the distance between the source and the object, as well as the distance between the specimen and the film, were kept constant so that there would be identical conditions for all the exposures taken at the different pocket depths. Fig 9. The cadaver specimen positioned in the column and held in place by metal pins. The exposure factors used for the different pocket depths were 90kVp, 15mA, 3.5mm total aluminum filtration and exposure time of 0.8 sec, 1 sec, 1.25 seconds (0.8 sec= 48/60 , h of a second). Of the different exposures, the one selected was the 0.8 seconds since it gave us a better dynamic range of scanning. 53 The radiographs were processed with an AT 2000 processor (Air Techniques Inc., Hicksville, NY, USA) with Kodak Readymatic chemistry at 28°C in a darkroom with a safety light and a processing time of 5.5 minutes. 2.2.2 Scanning The images were scanned with a Hewlett Packard flatbed scanner HP6100 (Hewlett Packard Company, Greeley, CO, USA). During the scanning of the films, the size of the images was maintained constant at 3 x 4 cm so that the images would be of comparable size, and the contrast and brightness were also maintained at 125 for all the scanned images, which is the middle value of the range of contrast and brightness values (0-250). This was done to avoid having additional variables in the statistical analysis and to make all images comparable since they were of the same specimen and were taken under identical conditions. The variable selected to determine image quality was resolution. The images were all scanned at the four selected resolutions used in the clinical study (600, 300, 150, and 75 dpi). The result produced digital images of the same pocket with the different created depths at different qualities. 2.2.3 Measurements All scanned images were included for the performance of the measurements, including the original one (before any lesions were created). The observers were the same eight used in the clinical study. The images were randomized by use of an Excel randomization pattern and entered into PowerPoint 4 as in the clinical 54 study. They were then given to the observers to mark the points in the radiographs of interest. Those points were the same as in the clinical study and were the following: The cemento-enamel junction (CEJV. which is the point of the tooth where the enamel meets the dentin. The highest point on the alveolar crest bone: at the area of the defect of interest. The deepest point of the infrabony periodontal defect: of interest, between the first and second premolar in the specimen used. The visual analog scale: to indicate the subjective opinion of the volunteers on image quality. Then measurements were taken of the points' coordinates in the exact same way as in the clinical study. The marked images were opened in NIH Image program. All the points that were selected by the observers were four pixels in size, and the top left pixel was selected for the determination of the required measurements so that the results obtained would be comparable. Out of all the points selected, the x and y coordinates were measured in pixels and further analysis of the data set was performed. 55 a Resolution of 600 dpi. b Resolution of 300 dpi. c Resolution of 150 dpi. d Resolution of 75 dpi. Fig 10 An example of the images of the cadaver at the four different qualities used. The selected images are before the infrabony defects were created. 56 2.3.SECTION III (Endodontic model section) 2.3.1 Test objects The data were taken from pre-existing work that was done by Dr E. Radan [1999] for her MSc Thesis project. She had taken endodontic files of different sizes from 6 up to 15 (6, 8, 10, 15) that were inserted in a tooth at three different positions (1 mm over the apex, 1 mm and 3 mm short of the apex) and were kept in position by use of a wire that held the tooth and file in position. Then radiographs were taken using standard conditions so that the data would be comparable. Therefore, there were four different file sizes at the three positions in regard to the tooth. 2.3.2 Radiography Although the images were already scanned by Dr Radan, mention of the conditions under which the experiments were performed should be given. The x-ray source, object, and film positioning were standardized during the experiment. The films used were Ektaspeed Plus films (Kodak Canada Inc, Toronto, Ontario, Canada) size 2 and were kept in identical position at a 1 meter distance from the source. Exposure factors were 90kVp, 15mA, 3.5 total aluminum filtration and 0.6 seconds. 57 2.3.3 Digitizing The radiographs were scanned with a Hewlett Packard 3C flatbed scanner and a 4c transparency adapter (Hewlett Packard Company, Greeley, CO, USA) as 8 bit grey scale images at resolutions of 75, 150, 300, and 600 dpi. Of all the radiographs that were used in the original experiment, only the 75 and 600 dpi images were selected for our study. 2.3.4 Image manipulation We took only the images of good quality (600 dpi) and of poor image quality (75 dpi), giving us a total of 24 images. Fig 11 An example of a good and bad quality image from the selected ones. The file size in the images is 15, the file position is 1mm over the apex, and the images qualities are the selected ones of 600 and 75 dpi. 58 The selected images of the two qualities were taken and then blurred. The images were blurred using 7 x 7 mean software, which is a kernel in NIH Image, taking the mean value of a 7 x 7 pixel area and substituting it with the average of all the measurements. We therefore had a total of 48 images. Fig 12 An example of a good quality image and a bad quality image after blurring. The images are the same as figure 11, but blurred. The blurring feature was selected since it smoothes the original pixelated images and softens the edges, which, in the case of the 75 dpi, results in a subjectively better quality image. Another reason for selecting this feature was that the current version of PowerPoint automatically blurs the image upon import to a slide. We tried to find out the feature that PowerPoint uses when blurring the 59 images, but we couldn't since it is determined by the software and is not mentioned in the manual of the office 2000 version, nor was there any information about it on the Microsoft web page. Not being able to have control of the images obtained by use of this feature, we selected the feature that we considered most appropriate since its application resulted in the best quality images. Before we decided to use that particular feature, we tried other features as well, such as 3 x 3 or 11 x11 mean and Gaussian 7 x 7 and 11 x 11. Our selection included those features since they are the only features readily available for image smoothing. The blurring was very exaggerated in the other selections, reducing our choice to the 7 x 7 mean kernel. 2.3.5 Observers The observers used in this part of the study were the same ones who were used in the previous two parts of the study. They performed the measurements on the endodontic files on a separate occasion and were given a total of 48 images in 24 PowerPoint slides. They were asked to mark the following points: The tip of the root apex. The tip of the endodontic file. The quality of the image on the visual analog scale. In this section of the study, they were asked to repeat the measurements a second time to determine intraobserver agreement. 60 2.3.6 Measurements Measurements were taken of the points' coordinates with the help of NIH Image, and the difference between blurred and not blurred images was measured with the visual analog scale and the distances obtained between file tip and root apex for all the images. The analysis of the data is included in further sections of the study. Chapter 3 Analysis and results 62 Section I (Clinical study) The measurements consisted of the coordinates of the three marked points, together with the x coordinate of the Visual Analog Scale. After the acquisition, the data were entered into an Excel spreadsheet and the order of the slides was brought back to the initial one, before the randomization. The variables were the following: 1. The dependent variables were the coordinates of the points that the observers marked in the radiographs as well as their subjective measurement in the Visual Analog Scale (VAS). 2. The independent variables were the case number, the image quality, and the observer. The entire data set consisted of the variables of slide number before the actual sorting to the initial stage, case number, observer image quality, and the measurements that we had obtained. All the measurements (each coordinate of the points) were plotted to check for normality, and they were all normally distributed (the classic bell-shaped curve). Though two of the curves produced were skewed we decided to proceed with parametric tests. Besides the plotting of the data we took all the informative statistics for each measurement. Since the variables in the data set were normally distributed, our observations were independent and the variance was similar between the groups, we selected to perform Analysis of Variance (ANOVA) as the statistic most suited for the analysis. 63 Since each case was different and the values would not correspond in positions, we took the mean value for each case and subtracted it from each independent value in the data set. The difference of each coordinate measurement from the mean was transferred to a data set in a spreadsheet of the SYSTAT Program (Intelligent software, Evanston, IL, USA). The values of the Visual Analog Scale (VAS), which represent the observers' subjective opinion of the image quality, were entered in a similar way. Analyses of Variance were done for each of the independent variables. The interactions between each pair of independent variables were also studied in the test by the same method. For a three-factor experiment we have three possible two-way interactions effects, corresponding to the three possible pairing of the main factors (in our case image quality, case and observer). The two-way interaction is defined as a two-factor experiment. The interaction effect for two factors is found by ignoring the remaining factor. For the Visual Analog Scale, all the data we had was converted to a 0-10 scale to simplify the plotting of the data against image quality, and the mean for each different quality was measured and plotted too. The analyses were also performed with the use of SAS program (Statistical Analysis System). After the performance of the ANOVA Bonferroni's correction tests were used to check where the differences were located between the groups. Basically this method, which is a multiple comparison procedure, guarantees that the probability of any false rejection amongst all comparisons is no greater than the significance level set. Therefore we have a much stronger protection than controlling the probability 64 of a false rejection at the set significance level for each separate comparison, since performing multiple tests increases the probability that we will commit a type I error (reject the null hypothesis H 0 when it is true) (Pagano 1993). Results The analyses and measurements were done using the difference of each value from the true mean and not the actual values. For each case the mean was calculated and was subtracted from all the values we had. Therefore each value consisted of the difference from the mean for the measurement we had. By doing so the case factor was eliminated. To get a general idea of the points distribution initially graphs were plotted of the distribution of the points for each quality and ellipses at a 95% confidence interval were superimposed on them. Therefore we could visualize the distribution of the points' coordinates for all four different qualities. 65 The results we got were the following: For the cej: Quality 1(600 dpi) L U Fig 13a Scatterplot for the cemento-enamel junction for image quality 1 (600 dpi), (the axes represent pixel values and there are 23.62 pixels per mm). Quality 2 (300 dpi) >-•-i L U l-l -75 Fig 13b Scatterplot for the cemento-enamel junction for image quality 2 (300 dpi). 66 Quality 3 (150 dpi) >-L U Fig 13c Scatterplot for the cemento-enamel junction for image quality 3 (150 dpi). Quality 4 (75 dpi) >-L U Fig 13d Scatterplot for the cemento-enamel junction for image quality 4 (75 dpi). For the top of the alveolar crest: Quality 1 (600 dpi) L U l - l Fig 14a Scatterplot for the alveolar crest for image quality 1 (600 dpi). Quality 2 (300 dpi) L U l - l Fig 14b Scatterplot for the alveolar crest for image quality 2 (300 dpi). Quality 3 (150 dpi) 75 ! j i » • /•! itt / Irf fc* • I i i i • -75 0 75 CRESTX Fig 14c Scatterplot for the alveolar crest for image quality 3 (150 dpi). Quality 4 (75 dpi) Fig 14d Scatterplot for the alveolar crest for image quality 4 (75 dpi). For the depth of the periodontal infrabony defect: Quality 1 (600 dpi) o D. - 7 5 Fig 15a Scatterplot for the depth of the defect for image quality 1 (600 dpi). Quality 2 (300 dpi) 7 5 ! j •! • j • i i i - 7 5 0 7 5 P 0 C K E T X Fig 15b Scatterplot for the depth of the defect for image quality 2 (300 dpi). Quality 3 (150 dpi) -75 75 0 P O C K E T X Fig 15c Scatterplot for the depth of the defect for image quality 3 (150 dpi). Quality 4 (75 dpi) P O C K E T X Fig 15d Scatterplot for the depth of the defect for image quality 4 (75 dpi). 71 To get a better understanding of the data and the points' distribution in relation to quality, a graph was plotted of the differences in the mean values of the points, independent of observers, for each quality. The result we got was as follows: PCTYIUW 1.00 QUALITY 2.00 3.00 4.00 Fig 16 Plot of mean difference in mm of points for different qualities. Qualityl: 600 dpi, 2: 300 dpi, 3: 150 dpi, 4: 75 dpi. To better visualize the above graph a representation of the means with bars was performed. The result we obtained was as follows: Fig 17 Bar graph of means of points against quality. The above demonstrate the differences of the markings between the better quality images and the worse quality images with most evident differences between a quality of 75 dpi and the qualities of 150 dpi and above. In the linear plot we can clearly see the horizontal direction of the points for the first three qualities while the bar graphs better demonstrates the negative values found in the case of quality four (75 dpi) which can be explained by a shift of the mean towards the high qualities. To verify the above graphs we proceeded to statistics. Knowing the difference between the various cases and based on the fact that the study was a clinical one, the elimination of the cases factor was done, since the difference between cases was a variable we didn't want to include in the statistics. The ANOVA obtained were the following: For the cemento-enamel junction: Cej-x Cej-y DF F value P value F value P value Observers 7 1.6328 0.1232 3.3800 0.0015 Quality 3 3.0811 0.0269 10.7094 0.0000 Table 1a Analysis of variance for the cej when case factor is excluded. 73 For the crest: Crest-x Crest-y DF F value P value F value P value Observers 7 1.4680 0.1757 2.4198 0.0188 Quality 3 5.5689 0.0009 3.4797 0.0157 Table 1 b Analysis of variance for the alveolar crest when case factor is excluded. For the depth of the infrabony defect: Pocket-x Pocket-y DF F value P value F value P value Observers 7 0.8197 0.5710 5.3064 0.0000 Quality 3 1.8290 0.1406 13.0000 0.0000 Table 1c Analysis of variance for the pocket depth when case factor is excluded. The above results showed the following: For the cemento-enamel junction though in the x coordinate there was no variation for the observers, there was for the image quality. The y coordinate showed big difference between the observer p value and the quality p value though both are statistically significant. The same was valid for the top of the alveolar crest, though in this case the difference between the observers and the quality for the y coordinate is not as wide. For the periodontal pocket the same was valid though in this case the p values are even smaller as was expected considering the fact that the pocket depth is not as accurate a measurement as the other two. After all the above statistics we checked for interactions between the different pairs of independent variables we had. During this process we initially checked for interactions with all the 3 independent variables included in the analysis and then we did the same for each independent pair of variables. We performed the statistics with the three factor interactions including the cases and taking both the original values as well as the difference from the mean. In both cases the results were identical and were the following: For the cemento-enamel junction: Cej-x Cej-y DF F value P value F value P value Cases* observers 133 1.3636 0.0117 1.5187 0.0011 Observers* quality 21 0.7457 0.7849 1.4717 0.0829 Cases* quality 57 1.5845 0.0066 1.5181 0.0124 Table 2a:Analysis of variance for the cej when the interactions between the independent variables are considered. 75 For the alveolar crest: Crest-x Crest-y DF F value P value F value P value Cases* observers 133 1.9579 0.0000 3.2286 0.0000 Observers* quality 21 1.4445 0.0935 0.8622 0.6411 Cases* quality 57 1.4639 0.0205 1.9264 0.0002 Table 2b:Analysis of variance for the alveolar crest when the interactions between the independent variables are considered. For the depth of the infrabony defect: Pocket-x Pocket-y DF F value P value F value P value Cases* observers 133 1.0959 0.2498 2.7399 0.0000 Observers* quality 21 1.1213 0.3221 1.1168 0.3267 Cases* quality 57 1.1114 0.2797 2.5500 0.0000 Table 2c:Analysis of variance for the infrabony defect deepest part when the interactions between the independent variables are considered. After that we considered each pair of independent variables (and their interaction) separately, excluding the third variable. This was done in order to avoid fitting a saturated model (i.e. there would be no estimate of MSError). If we 76 included all three variables in the ANOVA model, with two-factor interactions, and the three-factor interaction term, there would be no degrees of freedom left for error. The results we got were the following: For the interaction between case and observer: Case * observer DF F value P value Cej-x 133 1.27 0.0365 Cej-y 133 1.32 0.0194 Crest-x 133 1.76 0.0001 Crest-y 133 2.86 0.0001 Pocket-x 133 1.07 0.3013 Pocket-y 133 2.10 0.0001 Table 3a Analysis of variance for all poin case and observer is considered. :s when two-way interaction between 77 For the interactions between cases and quality: Case * quality DF F value P value Cej-x 57 1.46 0.0192 Cej-y 57 1.29 0.0818 Crest-x 57 1.17 0.1962 Crest-y 57 1.24 0.1222 Pocket-x 57 1.08 0.3198 Pocket-y 57 1.70 0.0017 Table 3b Analysis of variance for all poin case and quality is considered. s when two-way interaction between For the interactions between observers and quality: Observer * quality DF F value P value Cej-x 21 0.01 1.0000 Cej-y 21 0.03 1.0000 Crest-x 21 0.01 1.0000 Crest-y 21 0.08 1.0000 Pocket-x 21 0.03 1.0000 Pocket-y 21 0.14 1.0000 Table 3c Analysis of variance for all poin observer and quality is considered. s when two-way interaction between 78 The above tables demonstrate that though there are interactions between the cases and the observers in all the measurements, especially in the y coordinates, there are no significant interactions for the case x quality nor for the observer x quality. Besides the above statistics, corrections and type I error measurements were performed by Bonferroni's tests. That way the difference between the various variables could be located for each statistic we performed. The Bonferroni's tests gave significant differences for all points' coordinates between quality 1 (600 dpi) and quality 4 (75 dpi) except for the x coordinate of the depth of the pocket. Additional differences were shown in the case of the y coordinates among all four qualities. Bonferroni tests for the assessment of image quality: Quality 600 dpi 300 dpi 150 dpi 75 dpi 600 dpi 1.0000 300 dpi 1.0000 1.0000 150 dpi 1.0000 1.0000 1.0000 75 dpi Cej-x:0.0371 Cej-y:0.0000 Cst-x :0.0035 Cst-y :0.0354 Pct-y :0.0000 Cej-y :0.0000 Cst-x :0.0035 Cst-y :0.0677 Pct-y :0.0000 Cej-y :0.0000 Cst-x :0.0367 Pct-y :0.0003 1.0000 Table 4 Bonferroni P values for the observers' assessment of image quality for each digital resolution. 79 The Visual Analog Scale was considered separately. We converted the values to a scale of 1-10 and then plot them against the quality. The results for the graphs we got were the following: A scatterplot of the points for each quality: 10 8 6 4 0 1 2 3 4 QUALITY Fig 18 Scatterplot of VAS against quality. and from the above graph the means for each quality are plotted: CD 2 0 1 2 3 QUALITY Fig 19 Means of the VAS for each quality. The above graphs show the deterioration in measurement of the subjective opinion of the observers with regard to quality. The better quality images are given a higher rating on the 0 to 10 scale, while the lower quality images are given lower ratings. All the above figures and graphs will be analyzed in details in the Discussion section of the study. 81 Section II (In vitro study) The data in the cadaver study consisted of similar measurements to the clinical study. The only difference between the variables was that instead of cases we had a known infrabony defect depth of the one particular area we had selected to create the defects. Therefore the difference was in the independent variables that were the pocket depth, the image quality and the observer. Our dependent variables were the same as in the clinical study and consisted of the points' coordinates the observers had marked, and their subjective measurement in the Visual Analog Scale (VAS). Once again the measurements were plotted to check for normality and all the information on each statistic was taken. The observations were independent, normally distributed and the variance similar between the groups and therefore the statistic of choice was analysis of variance. Since we didn't have different pockets the actual data was used for the statistics and not the difference from the mean. The data was entered in a Systat spreadsheet and ANOVA were performed for each of the independent variables. After the ANOVA the interactions between the variables were investigated. Finally for both the initial ANOVA and for the interactions Bonferroni's tests were done to check where the differences found were located between the groups investigated. 82 Results Like in the case of the clinical study initially we plotted graphs of the distribution of the points at the various qualities at a 95% confidence interval. This helped us visualize the data and understand the points distribution for the different qualities. The results we got from the graphs were the following: For the cej: Quality 1 (600 dpi) I i j i j j i j I J -50 0 50 D I F F C E J X Fig 20a Scatterplot for the cemento-enamel junction for image quality 1 (600 dpi), (the axes represent pixel values and there are 21.00 pixels per mm). X 83 Quality 2 (300 dpi) Fig 20b Scatterplot for the cemento-enamel junction for image quality 2 (300 dpi). Quality 3 (150 dpi) 5 0 i X - 5 0 - 5 0 0 5 0 D I F F C E J X Fig 20c Scatterplot for the cemento-enamel junction for image quality 3 (150 dpi). 84 Quality 4 (75 dpi) Fig20 d Scatterplot for the cemento-enamel junction for image quality 4(75 dpi). For the alveolar crest: Quality 1 (600 dpi) 0 5 0 D I F F C S T X Fig 21a Scatterplot for the alveolar crest for image quality 1 (600 dpi). Quality 2 (300 dpi) - 5 0 0 5 0 D I F F C S T X Fig 21 b Scatterplot for the alveolar crest for image quality 2 (300 dpi). Quality 3 (150 dpi) LO 0 5 0 D I F F C S T X Fig 21c Scatterplot for the alveolar crest for image quality 3 (150 dpi). Quality 4 (75 dpi) 5 0 i D I F F C S T X Fig 21 d Scatterplot for the alveolar crest for image quality 4 (75 dpi). For the depth of the infrabony defect: Quality 1 (600 dpi) 5 0 I j h t-— % V ; | • m i j - 5 0 0 5 0 D I F F P O C X Fig 22a Scatterplot for the depth of the defect for image quality 1 (600 dpi). Quality 2 (300 dpi) D I F F P O C X Fig 22b Scatterplot for the depth of the defect for image quality 2 (300 dpi). Quality 3 (150 dpi) -50 0 50 DIFFPOCX Fig 22c Scatterplot for the depth of the defect for image quality 3 (150 dpi). Quality 4 (75 dpi) 50 ! i A i to .•1! • .! u i i i -50 0 50 DIFFPOCX Fig 22d Scatterplot for the depth of the defect for image quality 4 (75 dpi). 89 To get a better understanding of the above points coordinates distribution we plotted graphs of the differences in the mean values of the points independent of the observers and depth for each quality as we did for the clinical study: The results were as follows: •FCEJXM •FCEJYM •FCSTXM DFCSTYM •FPCTXM •FPCTYM QUALITY Fig 23 Plot of the means of points for different qualities. To better visualize this we plotted the same graph with bars and the results were as follows: |D!FCEJXM IDFCFJYM IDIFCSTXM IDIFCSTYM IDFFCTXM IDIFFCTYM QUALITY Fig 24 Bar graph of means of points against quality. 90 The above graphs indicate the deterioration of the measurements with quality for all the variables except for the cej-y. Like in the clinical section the bar graph helps us view the difference between the quality 4 (75 dpi) and the other three qualities. The negative values in the worst quality appear here again, due to the shift of the mean towards the higher three qualities. The linear representation helps us view the similar distribution of the points for the three higher qualities. To help us analyze the data we had after the above graphs were plotted statistical analysis if the data was done. Initially ANOVA were performed for the coordinates of each point to check for differences according to pocket depth, quality and observer. The results we got were the following for each point. For the cemento-enamel junction: Cej-x Cej-y DF F value P value F value P value Pocket 6 0.0000 1.0000 0.0000 1.0000 Observer 7 12.1667 0.0000 14.0060 0.0000 Quality 3 58.6522 0.0000 6.6216 0.0003 Table 5a Analysis of variance for the cej. 91 For the crest: Crest-x Crest-y DF F value P value F value P value Pocket 6 0.0000 1.0000 0.0000 1.0000 Observer 7 18.9853 0.0000 5.4985 00000 Quality 3 73.2568 0.0000 8.8982 0.0000 Table 5b Analysis of variance for the alveolar crest. For the depth of the infrabony defect: Pocket-x Pocket-y DF F value P value F value P value Pocket 6 0.0000 1.0000 0.0000 1.0000 Observer 7 8.8907 0.0000 7.6453 0.0000 Quality 3 42.3449 0.0000 1.3157 0.2703 Table 5c Analysis of variance for the depth of the infrabony defect. After the above statistics we performed interactions between the different independent variables were checked for. Initially we checked for interactions between all 3 variables together. 92 The results we got were the following: For the cemento-enamel junction: Cej-x Cej-y DF F value P value F value P value Pocket* observer 42 0.5036 0.9939 1.0607 0.3910 Pocket* quality 18 7.7518 0.0000 5.0087 0.0000 Observers* Quality 21 4.4548 0.0000 10.9227 0.0000 Table 6a Analysis of variance for the cej when the interactions between the independent variables are considered. For the alveolar crest: Crest-x Crest-y DF F value P value F value P value Pocket* observer 42 1.7381 0.0101 5.9939 0.0000 Pocket* quality 18 5.0708 0.0000 2.0323 0.0123 Observers* Quality 21 0.8469 0.6578 0.9484 0.5308 Table 6b Analysis of variance for the alveolar crest when the interactions between the independent variables are considered. 93 For the depth of the infrabony defect: Pocket-x Pocket-y DF F value P value F value P value Pocket* observer 42 1.6328 0.0199 4.0365 0.0000 Pocket* quality 18 4.9152 0.0000 1.4534 0.1185 Observers* Quality 21 0.7754 0.7444 0.5397 0.9484 Table 6c Analysis of variance for the deepest part of the infrabony defect when the interactions between the independent variables are considered. After the interactions with all the independent variables together we considered each pair of variables separately to allow degrees of freedom for error. The results we got were the following: 94 For the interactions between defect depth and observer: Defect depth * observer DF F value P value Cej-x 42 0.12 1.0000 Cej-y 42 0.36 0.9999 Crest-x 42 0.52 0.9929 Crest-y 42 4.22 0.0001 Pocket-x 42 0.66 0.9415 Pocket-y 42 3.99 0.0001 Table 7a Analysis of variance for all poin defect depth and observer is considered s when two-way interaction between For the interaction between defect depth and quality: Defect depth * quality DF F value P value Cej-x 18 3.83 0.0001 Cej-y 18 1.55 07.075 Crest-x 18 2.40 0.0018 Crest-y 18 0.83 0.6604 Pocket-x 18 3.20 0.0001 Pocket-y 18 0.72 0.7831 Table 7b Analysis of variance for all poin defect depth and quality is considered s when two-way interaction between 95 Finally for the interaction between quality and observer the results we got were the following: Quality * observer DF F value P value Cej-x 21 0.78 0.7367 Cej-y 21 2.11 0.0046 Crest-x 21 0.11 1.0000 Crest-y 21 0.03 1.0000 Pocket-x 21 0.13 1.0000 Pocket-y 21 0.05 1.0000 Table 7c Analysis of variance for all poin quality and observer is considered. :s when two-way interaction between 96 After all the above statistics Bonferonni's corrections tests were done to test for type I error and its measurements. That way the differences between the variables could be located. The Bonferroni's tests showed differences of statistical significance for the different qualities, for all the points coordinates between image quality 4 (75 dpi) and the other three except for the y coordinate of the pocket which showed no difference between any of the qualities. Additional differences were shown only for the y coordinate of the alveolar crest and the x coordinate of the pocket between qualities 1 (600 dpi) and 3 (150 dpi). Quality 600 dpi 300 dpi 150 dpi 75 dpi 600 dpi 1.0000 300 dpi 1.0000 1.0000 150 dpi Cst-y:0.0435 Pct-x:0.0283 1.0000 1.0000 75 dpi Cej-x:0.0000 Cst-x:0.0000 Cst-y:0.0000 Pct-x :0.0000 Cej-x:0.0000 Cej-y :0.0014 Cst-x :0.0000 Cst-y :0.0023 Pct-x :0.0000 Cej-x:0.0000 Cej-y :0.0008 Cst-x :0.0000 Pct-x :0.0000 1.0000 Table 8 Bonferroni's corrections test to show where the differences in the measurements are located. The Visual Analog Scale was considered separately. The data we had was converted to a scale of 0-10 and then was plotted against quality. The results were as follows: The scatterplot of the measurements of the VAS in relation to quality: QUALITY Fig 25 Scatterplot of VAS against quality. And the same graph for the mean values of the above: QUALITY Fig 26 Means of the VAS for each quality. 98 Again in the VAS in this section of the study we can see the same drop of the subjective opinion of the observers with relation to quality. This indicates that despite their performance in the different qualities their opinion on the image quality is significantly different between good and bad quality images. In this part of the study after the quality of 300 dpi there is a linear decrease of the measurement. All the above figures and graphs will be analyzed in detail further. 99 Section III (endodontic model) The endodontic study constituted a different section of the experiments we performed. In this model we used existing data and tried to see the effect of an additional factor (blurring). The factor was selected since it is a feature in PowerPoint's newer versions. Not being able to identify the type of blurring used in the program, we selected the best kernel from the NIH image kernels and used it as the one of choice. The Kernel we used was the 7 x 7 mean, in which the mean value of every 7 pixels in both directions is taken and used as a replacement of all the independent values. The data we obtained from the measurements were entered into a spreadsheet in Microsoft Excel Program and the difference in distance between the two points of interest was calculated. The values in pixels were converted into mm by calculating the magnification factor of PowerPoint upon entering an image. This in turn was measured by comparing distances between original images and the same images after insertion into PowerPoint. There were 2 image qualities but all the cases we had were of the same magnification factor in the original images. The data set we had consisted of the following variables: The independent variables were the slide number, the file size (there were four different file sizes used: 6,8,10,15), the file position (there were three different file positions used: 1mm over the apex of the root, 1mm short from the apex of the 100 root and 3 mm short from the apex of the root), the image quality (600 and 75 dpi or quality 1 and 2 respectively), and the blurring factor (original images and blurred images or blurring 1 and 2 respectively). The dependent variables we had were the points' coordinates and the VAS measurement, which again represented the subjective opinion of the observers regarding the quality of the image. Results Initially graphs were plotted of the mean values of the blurred and non-blurred images according to the file size and file position as well as image quality for both readings. The graphs we got were the following: According to file size: file Fig 27a Bar graph of the difference in distance against file size. File sizes are of 6,8,10 and 15. Blurring 1: original images and 2: blurred images. According to file position: blurting • I 1.00 Error Bars show 95.0% Cl of Mean Bars show Means 1.00 2.00 3,00 filepos Fig 27b Bar graph of the difference in distance against file position. File positions are 1:1mm over the root apex, 2: 1mm short of the apex and 3: 3mm short of the apex. Blurring is 1: original non blurred images, 2: blurred images. According to image quality: 1 00 2 00 quality Fig 27c Bar graph of the difference in distance against image quality. Image quality 1: 600dpi and 2: 75dpi. Blurring 1: original images, 2: blurred images. 102 The same graphs were plotted for the second reading of the observers and the results we got were comparable: Plots of the difference in distance against the file size file position and image quality for the second reading: Plots of the difference in distance against file size: blurting n — t o o Si" * 2 00 Error Bars show 9 5 . 0 % CI of Mean Bars show Means filesize Fig 28a Bar graph of the difference in distance against file size for the second reading. Plots of the difference in distance for the different file positions: 1.00 7 0 0 3.00 filepos Fig 28b Bar graph of the difference in distance against file position for the second reading. 103 Plots of the difference in distance for the different qualities: Error Bars show 95.0% Cl of Mean Bars show Means quality Fig 28c Bar graph of the difference in distance against image quality for the second reading. From the above graphs we can see the similarities between the two readings of the observers and in the measurements they performed. The graphs we obtained from the two measurements are comparable and the results very similar. Then we plotted the observers distribution for the two readings and found that there was a difference between the two readings in the observers' variability amongst them. Though there was only one observer in each reading that gave significant variability after the statistics were applied. For the first reading the graphs we got were the following: 104 1.6 1.4 1.2 1.0 co § 6 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 OBSERVER Fig 29a Bar graph of the difference in distance against observers for the first reading. While for the second reading the observers' distribution was as follows: 1.8 1.6 1.4 1.2 S 1.0 CO Q Q c ra 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 OBSERVER Fig 29b Bar graph of the difference in distance against observers for the second reading. 105 To verify the above graphs we got we performed ANOVA initially for all the independent variables. The results we got were the following: For the tip of the file: File x File y DF F value P value F value P value Observer 7 0.00 1.0000 3.89 0.0004 File size 3 1.51 1.2108 23.71 0.0001 File position 2 1.82 0.1637 332.27 0.0001 Quality 1 1.93 0.1657 245.14 0.0001 Blurring 1 11.70 0.0007 2.41 0.1217 Table 9a Analysis of variance for the tip of the endodontic file. 106 For the end of the root: Root-x Root-y DF F value P value F value P value Observer 7 0.00 1.0000 3.68 0.0007 File size 3 1.57 0.1959 8.07 0.0001 File position 2 2.37 0.0951 5.00 0.0072 Quality 1 2.34 0.1272 19.12 0.0001 Blurring 1 11.31 0.0009 17.32 0.0001 Table 9b Ana ysis of variance for the end of the root. Besides taking the coordinates of each point we took the distance between the two points and used an ANOVA to check for differences according to the dependent variables we had: Distance between points DF F value P value Observer 7 3.79 0.0005 File size 3 19.87 0.0001 File position 2 187.59 0.0001 Quality 1 187.28 0.0001 Blurring 1 1.70 0.1928 Table 10 Analysis of variance for the distance between the tip of the file and the apex of the root. 107 After the above statistics were performed we did Bonferroni tests to check for errors according to file sizes and positions. The results we got were as follows: For the different file sizes used: File sizes Size 6 Size 8 Size 10 Size 15 Size 6 Size 8 Read1:0.0000 Read2:0.0034 Size 10 Read1:0.0000 Read2:0.0000 Size 15 Read1:0.0005 Read2:0.0000 Read1:0.0000 Read2:0.0000 Read1:0.0017 Read2:0.0022 Table 11: Bonferroni's correction test to check for differences amongst the different file sizes. For the different file positions used Bonferroni's correction test was done also and no statistically differences were found according to file position. The only significant result was between file position 2 (1mm short of the apex) and file position 3 (3mm short of the apex) in the second reading of the observers. Additional Bonferroni's tests were done for image quality and blurring and in both cases there were significant differences for both readings. That result was expected considering the difference between the quality 1 images (600 dpi) and the quality 4 images (75 dpi) as well as between the blurred and non-blurred images. 108 The Visual Analog Scale was considered separately and was plotted both for blurred and non-blurred as well as for quality. The graphs we got for the visual analog scale were the following: Plot of VAS against the four file sizes: 4 6 8 10 12 14 16 FILE Fig 30 Scatterplot of the VAS against file size. File sizes 6, 8, 10 and 15. Plot of VAS against the different file positions: FILEPOS Fig 31 Scatterplot of the VAS against the file positions. 109 File position 1: 1 mm over the apex, file position 2: 1 mm short of the apex, file position 3: 3mm short of the apex. Plot of VAS against the image quality: 10 r 2 o I-o co £ o BLURRING A 2 o 1 QUALITY Fig 32 Scatterplot of the VAS against the image quality. Quality!: 600 dpi, Quality2: 75 dpi. Box plots of the VAS against quality: BLURRING • 1 • 2 QUALITY Fig 33 Box plot of the VAS against quality. 110 And a simple bar graph indicating the same data: blurring • • 2.00 Error Bars show 95.0% CI of Mean Bars show Means 1.00 2 00 quality Fig 34 Bar graph of the VAS against quality. For the second reading of the observers a similar graph was obtained indicating the consistency of the measurements and of their opinion regarding image quality: blurring • 1.00 • 2.00 Error Bars show Mean +/- 1.0 SE Bars show Means 1.00 2.00 quality Fig 35 Bar graphs of the VAS against quality for the second reading. I l l For the above figures and tables we can see clearly the differences between the blurred and the non-blurred images as well as the difference in the various file sizes positions and qualities. To obtain a better understanding of the results we got an analysis of the Tables and Figures should be performed for all three sections. Section I (clinical study): Initially scatterplots of the points were performed at a 95% confidence interval for all the points and all the image qualities. These scatterplots are demonstrated in Figures 13 to 15. As we see there is a broader scatter of the points as the quality decreases which indicates a less accurate measurement by the observers. This is expected but easier to visualize and understand by plotting in a graph. After the above representation to get a better understanding of all the points plotted together, we did a linear graph representation (Figure 16) and a bar graph of the same thing (Figure 17). In those two Figures the drop in the accuracy of the measurements after quality 3 is clear. Though the first part of the graph is at the same level of difference from the mean, the quality 4 measurements are appreciably lower. In the case of the bar graph, quality 4 is mostly represented by negative values, which is expected since there is a shift of the mean value of all the qualities towards the higher qualities. There is very little difference 112 between qualities 1 to 3 and the larger difference can be seen between the above three qualities and quality 4. After the graphs we plotted, we performed statistics, to better understand the results we got. In Table 1 ANOVA is performed without including the case factor. The case factor was excluded since it is a random case selection and the pockets in each case were not in the same position. In order to do so the difference from the mean of every case was taken for each measurement, and therefore we had differences from a point (mean) rather than coordinates. After the analysis, an interaction analysis was performed for all the variables together, which are demonstrated in Table 2. The biggest interaction for the cej was between the case and observer though there was statistically significant difference between the interaction of quality and cases. For the alveolar crest and the depth of the pocket the same interactions are obtained. Table 3 demonstrates the ANOVA performed for each pair of interactions separately to allow more for error and to avoid fitting a saturated model. Section II (cadaver study): Like in the clinical section, initially scatterplots were done to view the distribution of the points for the different qualities (Figures 20 to 22). Again here, as in the case of the clinical study, the difference in the distribution in the case of quality 4 113 is different from the other three qualities. The scatterplots for all the points examined are more widely scattered in the case of the low quality images (quality 4), which is expected since the measurements in lower qualities are not as precise. After the above scatterplots again as in the clinical section linear graphs and bar graphs were drawn for all the points together to visualize the difference between them (figures 23 and 24). In this section as in the previous one the quality 4 images have lower values of difference from the mean compared to the other three image qualities. The negative values in the case of quality 4 are also present here, as in the previous section, and can be explained from the mean shift towards the other three qualities. The only point that differs in the cadaver section compared to the clinical section is the y coordinate of the cemento-enamel junction, which is a point of no significance. This is only because we are interested in the pocket depth and alveolar crest rather than the cemento-enamel junction for periodontal defects. And even in this point there is a big difference in the measurements between quality 4 and the rest. The same statistics were performed for the cadaver study as in the clinical study. In this section, for the initial ANOVA, the case factor is excluded since we were referring to the same pocket of the same jaw specimen. 114 In Table 5 where the results of the performed ANOVA are demonstrated, there are statistically significant differences for the cemento-enamel junction and the alveolar crest according to both observer and image quality; while for the depth of the pocket there are statistically significant results only according to observer and not according to quality. For the interactions between the factors, which are given in Tables 6 and 7 we see that there are interactions of statistical significance between the infrabony defect depth and quality as well as for the defect depth and observer pairs in the case of the cemento-enamel junction. For the alveolar crest the pairs that have significant interactions are between pocket depth and observer and pocket depth and quality, while for the pocket depth the only pair that gave significant differences was between pocket depth and observer. The Visual Analog Scale for both sections: In both the clinical and the cadaver part of the study the Visual Analog Scale shows a characteristic drop with image quality after quality 2 (300 dpi). A comparison of Figures 18 (clinical study) and Figure 25 (cadaver study) demonstrate that drop, and the difference between image quality 4 and the other three qualities. Therefore independent of the measurements they performed in the radiographs, the subjective opinion of the observers regarding image quality is characteristic, showing a drop after quality 2 in both sections. The plot of the 115 means in Figure 19 (clinical study) and Figure 26 (cadaver study) indicates that clearly with a linear drop after quality 2. Section III (endodontic study): The endodontic section of the study is discussed separately from the other two sections since a different factor was considered in this model. In this section the additional factor considered was blurring. Graphs were plotted for both readings of the actual difference in distance between the points against all the dependent variables. The bar graphs for the mean distance of the two points according to the variables of file size and file position show the differences between the blurred and non-blurred images (Figures 27 and 28). The two readings gave us similar findings as is demonstrated in the above Figures. What should be mentioned is that in the initial experiments performed by Dr Radan for her MSc there was a difference found for file size 8, which was regarded as an error in the measurements. The same thing was seen here and there is no rational explanation as to why file size 8 would give a bigger distance from the mean when compared to both smaller and larger sizes. After the plotting of the graphs ANOVA were performed for the different points to see the effects of the dependent variables on the points' measurements. As we 116 see, the identification of the file tip gave us statistically significant results for the blurring in the x coordinate and for all the other four variables (quality, file size, file position, and observer) for the y coordinate. Similar results were obtained for the root apex, though in this case for the y coordinate a difference in the case of blurring was also observed. The above are clearly indicated in Table 9. In the case of quality we can see that blurring caused improvement in the accuracy for the poor quality images while deterioration in the measurement for the good quality images. This is of importance since the blurring feature of PowerPoint might effect image presentation and quality in a similar manner. The distribution of the observers with regard to image quality and blurring varied between the two readings despite the fact that the statistics were similar. The Visual Analog Scale was considered separately for the two readings and we can see in Figures 30 to 32 the distribution of the measurements for the observers' opinion with regard to file position, file size, and image quality. Similar results were obtained for the second reading. We clearly see the distribution of the points in the case of the non blurred images following a bimodal pattern according to qualities (quality 1 giving measurements in the values of 8-10 and quality 2 in the values of 0-2) while in the case of the blurred images it is more scattered in the middle values between 2 and 6 for both blurred and non-blurred images. This is better visualized in Figure 33 where the box plots of the mean distance between the two points is given according to image quality. 117 The bar graph of the mean difference in distance of the points against the quality clearly shows the improvement in the observers' subjective opinion for the poor quality images and the decrease in the case of the good quality images for both readings caused by blurring (Figures 34 and 35). Chapter 4 Discussion and conclusions 119 Discussion Summary of section I and II of the study: This study demonstrated that the periodontal model functions as a reliable model for the evaluation and comparison of image qualities. The major finding was that clinically there is no difference between images of different resolutions unless those are very low. In our cases there was no difference found for resolutions of 150 dpi and higher, while there were differences found between the poorest quality we used (75 dpi) and the better ones. This is consistent with a lot of work done in the literature though usually the resolutions used are higher (Versteeg, Sanderink era/. 1998; Attaelmanan, Borg etal. 2000). In both sections of the experiments (clinical study and cadaver specimen) the largest variation in all the measurements was produced by variation in image quality, although for certain points there were statistically significant observers' variations noted. Our null hypothesis was that higher digital resolutions have better diagnostic quality for measurements performed in the case of infrabony defects. Lack of such difference for resolutions of 150 dpi and above rejects it, and allows us to draw the conclusion that there is no need for digital resolutions greater than 300 dpi for diagnostic purposes. The results obtained from the clinical study and the cadaver studies are consistent. 120 In the clinical study the elimination of the case factor (since they were all different clinical cases) gave us more relevant statistics. The analysis of the data helped us interpret graphs we got. We see that in the scatter plots quality 4 (75 dpi) has a broader spread in all the points when compared to the other three qualities. That means that there is a bigger difference and a worse reading for the observers in the case of a bad quality image. Note should be taken that there were some points' scatterplots that gave us a wider spread for the resolution of 600 dpi when compared to the lower qualities of 300 and 150 dpi. This has been verified by other researchers in the past also though with the use of different models (Moystad, Svanaes et al. 1995; Fuge, Stuck et al. 1998; Attaelmanan, Borg et al. 2000)] and is explained by the limitations of the display monitor. The bar graphs that show the distribution of the points gave us a negative mean for all the points in the case of quality 4 (75 dpi) which can be explained by a shift of the mean towards the other qualities. This shift is expected when considering the worsening of quality. The shift of the mean is a factor that explains the graphs and leads us to the conclusion that there is no difference of significance between the three good qualities, which is also verified by the statistics. The fact that there were fewer differences between the other three qualities consists a similar finding to a previous experiment performed by Dr Radan for her MSc thesis in the Faculty for the case of an endodontic model (Radan 1999). 121 In the case of the cadaver study, which would be the standard we have for comparison, we obtained similar results from the statistics. The only point that gave a big difference when comparing the two graphs is the cej-y coordinate which is not an important point since the cemento-enamel junction is a point that is easily identifiable in all four qualities and not a point of clinical significance in periodontal defects. There is deterioration in all the measurements performed by the observers as quality worsens. This is also expected since a worse quality image would be expected to give us worse readings as it does in this study. Summary of Section III of the study: This section constituted a separate experiment that involved the blurring feature of PowerPoint. The feature we used to simulate the blurring was the 7 x 7 Kernel of NIH Image. The images were blurred only upon entry and not upon copying and pasting. Assuming that the blurring feature of PowerPoint performs in a similar function to the Kernel we used, we can conclude that image quality might be affected upon the entry into the program. This is an important finding considering the increased use of PowerPoint presentations both for educational studies as well as for demonstrations and general information approach. This effect has not been studied previously, and further work should be undertaken to confirm these findings. 122 Our study in all three sections involved the scanning of conventional radiographs. This allowed us a greater flexibility of image quality, since in the case of direct digital systems there is limited variability of quality. Conclusions: Section 1,11: There were no significant differences between a quality of 600 and 300 dpi or even 150 dpi in the performance of the observers. The results come as a verification of previous work done in the Division with using a different model. Section III: There was significant difference between the accuracy and opinion of the observers between good and bad quality images with cases entered into PowerPoint. The feature of blurring similar to the one so commonly used for presentation has an effect in image quality, and attention should be given to the initial quality of the radiograph before insertion into the program. References: Araki K., A. Endo and T. Okano (2000). "An objective comparison of four digital intra-oral radiographic systems: sensitometric properties and resolution." Dentomaxillofac Radiol 29(2): 76-80. Attaelmanan A., E. Borg and H. G. Grondahl (2000). "Digitisation and display of intra-oral films." Dentomaxillofac Radiol 29(2): 97-102. Barbat J . and H. H. Messer (1998). "Detectability of artificial periapical lesions using direct digital and conventional radiography." J Endod 24(12): 837-42. Barr S. (1980). Dental radiology. Pertinent basic concepts and their applications in clinical practice. Philadelphia. Borg E. and H. G. Grondahl (1996). "On the dynamic range of different X-ray photon detectors in intra-oral radiography. A comparison of image quality in film, charge-coupled device and storage phosphor systems." Dentomaxillofac Radiol 25(2): 82-8. Borg E., K. Grondahl and H. G. Grondahl (1997). "Marginal bone level buccal to mandibular molars in digital radiographs from charge-coupled device and storage phosphor systems. An in vitro study." J Clin Periodontol 24(5): 306-12. Bragger U. (1988). "Digital imaging in periodontal radiography. A review." Journal of clinical periodontology(15): 551-557. Burger C. L, T. O. Mork, J . W. Hutter and B. Nicoll (1999). "Direct digital radiography versus conventional radiography for estimation of canal length in curved canals." J Endod 25(4): 260-3. Cederberg R. A., N. L. Frederiksen, B. W. Benson and J . D. Shulman (1998). "Effect of different background lighting conditions on diagnostic performance of digital and film images." Dentomaxillofac Radiol 27(5): 293-7. Chen S. K., L. Hollenderand K. A. Omnell (1997). "Detection of small differences in mass using a direct digital dental X-ray system." Dentomaxillofac Radiol 26(1): 63-6. Cohen M. E. and W. C. Roddy (1995). "A comparison of three statistics for detecting differences in digitized dental radiographs: a simulation study." Dentomaxillofac Radiol 24(3): 179-84. Conover G. L, C. F. Hildebolt and N. Yokoyama-Crothers (1996). "Comparison of linear measurements made from storage phosphor and dental radiographs." Dentomaxillofac Radiol 25(5): 268-73. Eickholz P., T. Riess, M. Lenhard, S. Hassfeld and H. J . Staehle (1999). "Digital radiography of interproximal bone loss; validity of different filters." J Clin Periodontol 26(5): 294-300. Farman A. G., S. L. Avant, W. C. Scarfe, T. T. Farman and D. B. Green (1998). "In vivo comparison of Visualix-2 and Ektaspeed Plus in the assessment of periradicular lesion dimensions." Oral Surg Oral Med Oral Pathol Oral Radiol Endod 85(2): 203-9. Forner L, M. C. Llena, J . M. Almerich and F. Garcia-Godoy (1999). "Digital radiology and image analysis for approximal caries diagnosis." Oper Dent 24(5): 312-5. Fuge K. N., A. M. Stuck and R. M. Love (1998). "A comparison of digitally scanned radiographs with conventional film for the detection of small endodontic instruments." Int Endod J 31(2): 123-6. Grondahl H. G. and K. Grondahl (1983). "Subtraction radiography for the diagnosis of periodontal bone lesions." Oral Surg Oral Med Oral Pathol 55(2): 208-13. Gurgan C , K. Grondahl and Wennstrom (1995). "Observer variation in the radiographic assessment of the bone level on the buccal and lingual surfaces of mandibular molars." Dentomaxillofac Radiol 24(3): 165-8. Gutteridge D. L. (1995). "The use of radiographic techniques in the diagnosis and management of periodontal diseases." Dentomaxillofac Radiol 24(2): 107-13. Hausmann E. (2000). "Radiographic and digital imaging in periodontal practice." J Periodontol 71 (3): 497-503. Hildebolt C. F., T. K. Pilgram, N. Yokoyama-Crothers, G. Fletcher, J . L. Helbig, T. Q. Bartlett, M. Gravier, M. W. Vannier and M. K. Shrout (1998). "Reliability of linear alveolar bone loss measurements of mandibular posterior teeth from digitized bitewing radiographs." J Clin Periodontol 25(11 Pt 1): 850-6. Holtzmann DJ J . W., Southard TE, Khademi JA, Chang PJ, Rivera EM (1998). "Storage-phosphor computed radiography versus film radiography in the detection of pathologic periradicular bone loss in cadavers." Oral Surg Oral Med Oral Pathol Oral Radiol Endod 86(1). Huda W., L. N. Rill, D. K. Benn and J . C. Pettigrew (1997). "Comparison of a photostimulable phosphor system with film for dental radiology." Oral Surg Oral Med Oral Pathol Oral Radiol Endod 83(6): 725-31. Janssen P. T., W. H. van Palenstein Helderman and J. van Aken (1989). "The detection of in vitro produced periodontal bone lesions by conventional radiography and photographic subtraction radiography using observers and quantitative digital subtraction radiography." J Clin Periodontol 16(6): 335-41. Jeffcoat M. K. (1992). "Radiographic methods for the detection of progressive alveolar bone loss." J Periodontol 63(4 Suppl): 367-72. Kullendorff B. and M. Nilsson (1996). "Diagnostic accuracy of direct digital dental radiography for the detection of periapical bone lesions. II. Effects on diagnostic accuracy after application of image processing." Oral Surg Oral Med Oral Pathol Oral Radiol Endod 82(5): 585-9. Kullendorff B., M. Nilsson and M. Rohlin (1996). "Diagnostic accuracy of direct digital dental radiography for the detection of periapical bone lesions: overall comparison between conventional and direct digital radiography." Oral Surg Oral Med Oral Pathol Oral Radiol Endod 82(3): 344-50. Manson-Hing (1979). Fundamentals of dental radiography. Philadelphia. Mistak E. J . , R. J . Loushine, P. D. Primack, L. A. West and D. A. Runyan (1998). "Interpretation of periapical lesions comparing conventional, direct digital, and telephonically transmitted radiographic images." J Endod 24(4): 262-6. Moystad A., D. B. Svanaes, T. A. Larheim and H. G. Grondahl (1995). "Effect of image magnification of digitized bitewing radiographs on approximal caries detection: an in vitro study." Dentomaxillofac Radiol 24(4): 255-9. Nair M. K., J . B. Ludlow, D. A. Tyndall, E. Platin and G. Denton (1998). "Periodontitis detection efficacy of film and digital images." Oral Surg Oral Med Oral Pathol Oral Radiol Endod 85(5): 608-12. Naitoh M., H. Yuasa, M. Toyama, M. Shiojima, M. Nakamura, M. Ushida, H. lida, M. Hayashi and E. Ariji (1998). "Observer agreement in the detection of proximal caries with direct digital intraoral radiography." Oral Surg Oral Med Oral Pathol Oral Radiol Endod 85(1): 107-12. Nielsen L. L., M. Hoernoe and A. Wenzel (1996). "Radiographic detection of cavitation in approximal surfaces of primary teeth using a digital storage phosphor system and conventional film, and the relationship between cavitation and radiographic lesion depth: an in vitro study." Int J Paediatr Dent 6(3): 167-72. Nummikoski P. V., B. Steffensen, K. Hamilton and S. B. Dove (2000). "Clinical validation of a new subtraction radiography technique for periodontal bone loss detection." J Periodontol 71(4): 598-605. Ohki M., T. Okano and T. Nakamura (1994). "Factors determining the diagnostic accuracy of digitized conventional intraoral radiographs." Dentomaxillofac Radiol 23(2): 77-82. Okano T., M. Ohki, T. Mera, H. Soejima, I. Ishikawa and N. Yamada (1988). "Quantitative evaluation of proximal bone lesions using digital subtraction radiography." Dentomaxillofac Radiol 17(2): 99-103. Pagano G. (1993). Principles of Biostatistics, Duxbury. Platin E., S. Mauriello and J . B. Ludlow (1996). "Effects of focal spot size on caries diagnosis with D and E speed images." Oral Surgery, Oral Medicine. Oral Pathology, Oral Radiology. & Endodontics 81(2): 235-9. Price C. and N. Ergul (1997). "A comparison of a film-based and a direct digital dental radiographic system using a proximal caries model." Dentomaxillofac Radiol 26(1): 45-52. Radan (1999). Evaluation of Digital and Geometric Unsharpness in Dental radiographs. Department of Oral Biological and Medical Sciences. Vancouver, University of British Columbia. Sanderink G. C , R. Huiskens, P. F. van der Stelt, U. S. Welander and S. E. Stheeman (1994). "Image quality of direct digital intraoral x-ray sensors in assessing root canal length. The RadioVisioGraphy, Visualix/VIXA, Sens-A-Ray, and Flash Dent systems compared with Ektaspeed films." Oral Surg Oral Med Oral Pathol 78(1): 125-32. Scarfe W. C , S. Norton and A. G. Farman (1995). "Measurement accuracy: a comparison of two intra-oral digital radiographic systems, RadioVisiography-S and FlashDent, with analog film." Dentomaxillofac Radiol 24(4): 215-20. Shearer A. C , K. Horner and N. H. Wilson (1990). "Radiovisiography for imaging root canals: an in vitro comparison with conventional radiography." Quintessence International 21(10): 789-94. Shearer A. C , K. Horner and N. H.Wilson (1991). "Radiovisiography for length estimation in root canal treatment: an in-vitro comparison with conventional radiography." International Endodontic Journal 24(5): 233-9. Shi X. Q., I. Eklund, G. Tronje, U. Welander, H. C. Stamatakis, P. E. Engstrom and G. N. Engstrom (1999). "Comparison of observer reliability in assessing alveolar bone changes from color-coded with subtraction radiographs." Dentomaxillofac Radiol 28(1): 31-6. Shrout M. K., B. J . Potter, H. M. Yurgalavage, C. F. Hildebolt and M. W. Vannier (1993). "35-mm film scanner as an intraoral dental radiograph digitizer. I: A quantitative evaluation." Oral Surg Oral Med Oral Pathol 76(4): 502-9. Shrout M. K., B. J. Potter, M. H. Yurgalavage, C. F. Hildebolt and M. W. Vannier (1993). "35-mm film scanner as an intraoral dental radiograph digitizer. II: Effects of brightness and contrast adjustments." Oral Surg Oral Med Oral Pathol 76(4): 510-8. Stassinakis A., U. Bragger, M. Stojanovic, W. Burgin, A. Lussi and N. P. Lang (1995). "Accuracy in detecting bone lesions in vitro with conventional and subtracted direct digital imaging." Dentomaxillofac Radiol 24(4): 232-7. Svanaes D. B., A. Moystad, S. Risnes, T. A. Larheim and H. G. Grondahl (1996). "Intraoral storage phosphor radiography for approximal caries detection and effect of image magnification: comparison with conventional radiography." Oral Surg Oral Med Oral Pathol Oral Radiol Endod 82(1): 94-100. van der Stelt P. F. (2000). "Principles of digital imaging." Dental Clinics of North America 44(2): 237-48. van der Stelt P. F., L. W. van der Linden, W. G. Geraets and C. L. Alons (1985). "Digitized image processing and pattern recognition in dental radiographs with emphasis on the interdental bone." J Clin Periodontol 12(10): 815-21. van der Stelt P. F., L. W. van der Linden, W. G. Geraets and C. L. Alons (1985). "Digitized pattern recognition in the diagnosis of periodontal bone defects." J Clin Periodontol 12(10): 822-7. Verdonschot E. H., A. J . Sanders and A. J . Plasschaert (1991). "Applicability of an image analysis system in alveolar bone loss measurement." J Clin Periodontol 18(1): 30-6. Versteeg C. H., G. C. Sanderink, W. G. Geraets and P. F. van der Stelt (1997). "Impact of scale standardization on images of digital radiography systems." Dentomaxillofac Radiol 26(6): 337-43. Versteeg C. H., G. C. Sanderink, S. R. Lobach and P. F. van der Stelt (1998). "Reduction in size of digital images: does it lead to less detectability or loss of diagnostic information?" Dentomaxillofac Radiol 27(2): 93-6. Versteeg K. H., G. C. Sanderink, F. C. van Ginkel and P. F. van der Stelt (1997). "Estimating distances on direct digital images and conventional radiographs." J Am Dent Assoc 128(4): 439-43. Visser H. and W. Kruger (1997). "Can dentists recognize manipulated digital radiographs?" Dentomaxillofac Radiol 26(1): 67-9. Wenzel A. (1998). "Digital radiography and caries diagnosis." Dentomaxillofac Radiol 27(1): 3-11. Wenzel A. (2000). "Digital imaging for dental caries." Dent Clin North Am 44(2): 319-38, vi. Wenzel A., E. Borg, H. Hintze and H. G. Grondahl (1995). "Accuracy of caries diagnosis in digital images from charge-coupled device and storage phosphor systems: an in vitro study." Dentomaxillofac Radiol 24(4): 250-4. White P. (2000). Oral Radiology. Principles and Interpretation. St Lewis, John Schrefer. Yoshiura K., T. Kawazu, T. Chikui, M. Tatsumi, K. Tokumori, T. Tanaka and S. Kanda (1999). "Assessment of image quality in dental radiography, part 1: phantom validity." Oral Surg Oral Med Oral Pathol Oral Radiol Endod 87(1): 115-22. Yoshiura K., T. Kawazu, T. Chikui, M. Tatsumi, K. Tokumori, T. Tanaka and S. Kanda (1999). "Assessment of image quality in dental radiography, part 2: optimum exposure conditions for detection of small mass changes in 6 intraoral radiography systems." Oral Surg Oral Med Oral Pathol Oral Radiol Endod 87(1): 123-9. Young S. J . , M. S. Chaibi, D. T. Graves, Z. Majzoub, F. Boustany, D. Cochran and P. Nummikoski (1996). "Quantitative analysis of periodontal defects in 132 a skull model by subtraction radiography using a digital imaging device." J Periodontol 67(8): 763-9. Yuasa H., Y. Ariji, M. Ohki, M. Naitoh, M. Shiojima, M. Ushida and E. Ariji (1999). "Joint Photographic Experts Group compression of intraoral radiographs for image transmission on the World Wide Web." Oral Surg Oral Med Oral Pathol Oral Radiol Endod 88(1): 93-9. Appendix I Pilot study paper 134 Pilot study A pilot study was done to help plan the definitive study. Purpose: To evaluate whether there is any loss of diagnostic information during the scanning of conventional radiographs in images of different qualities. Case selection: Three cases were selected from the dental clinics files. The cases had no restorations, were of posterior teeth and had periodontal pockets of varying depths. Scanning: The images were scanned at 600 dpi with the use of a Hewlett Packard 6100 flatbed scanner (Hewlett Packard Company, Greeley, CO, USA) as 8 bit grey images. The digitized images were then degraded to 300, 150, 75 dpi therefore giving us a total of 12 images. All selected 12 images were rotated by 180° in the horizontal plane giving us mirror images of the initial 12 images and a total of 24 images to be presented to the observers. Interpretation: The images were randomized using Excel (Microsoft Corporation, Redmond, WA, USA). Three dentists were selected to view the 24 images The observers were all general dentists and had no previous knowledge of either the content of the study or the selected cases. They all viewed the images in identical conditions under subdued lighting. There were no time limits set and the observers were allowed to vary their position from the screen although they were not allowed to alter any of the screen settings or the software features. They were asked to mark the following points on the 24 images: • The CEJ ( cemento enamel junction) • The top of the alveolar crest adjacent to the tooth of interest • The deepest part of the periodontal pocket adjacent to the tooth of interest • Their subjective opinion on the quality of the image on a scale of 0-10. The marked images were photographed and captured to memory. 135 Analysis of the data: The images that were photographed were opened in NIH image program. After obtaining the coordinates of all the points for all the observers, the data was sorted according to case and repetition to check for the differences in image quality and inter-observer variation. To do that, the mean value of all the points was calculated for each image, its difference from each point was measured, and ANOVAs were performed. Graphs of the points' distribution around the mean were plotted to check for the variability by image quality and observer. Results: There was no difference in the distance of the points from the mean between the various image qualities. There was significant variation according to cases, which means that although there was no variation in image quality there is variation by case, which could very easily be attributed to the small number of cases. The visual analog scale, which measured the subjective opinion of the image quality by the observers showed a big difference with image quality. Conclusions: From the data analysis it can be concluded that images with a resolution between 75 and 600 dpi do not give any diagnostic significant variability for the case of periodontal defects detection. Observers even though they ranked the images of 75 dpi as having very poor quality, still performed well in their evaluation of pocket depths in them. Appendix II Volunteer sheet with instructions 137 Dear volunteer, Thank you for agreeing to participate in this study. During this study you will be asked to view a series of slides and mark certain points on them. You can look at each of the slides for as long as you feel is necessary. The slides are all clearly marked with a number at the top left corner. Once you have opened a slide you will have to move the mouse to the bottom right of the screen where you will see a little pencil. By clicking on the pencil you can convert the arrow of the pointer to a pencil. Once your cursor is a pencil, you are asked to mark the following points on each slide. • The Cemento-enamel junction (CEJ) of the specified tooth. • The deepest part of the specified pocket and • The highest point of the alveolar bone crest at the pocket adjacent to the tooth specified. • Above each radiograph there is a Visual Analog Scale where you are asked to rate the slide according to its quality on a scale between 0 and 10 where 0 indicates a very bad quality image and 10 an excellent quality image. Once you have marked all four points on the slide, you will photograph the screen by pressing command and shift together and then while keeping them pressed you will press the number 3 key. Before you photograph the screen. make sure that you have removed the pencil sign from the slide area. If you have made a mistake or want to remove the points you have marked, what you do is move to the previous slide by pressing P and then go to the slide of interest by pressing N. Note than when you move to the previous slide the points you had marked on it will be absent. If during this process you have any questions or problems with the marking I will be there to help. The pockets that you are required to mark will be given to you on a separate sheet of paper. 138 Perio Part 1 Slide 1 Mesial pocket of the 6 Slide 2 Mesial pocket of the 6 Slide 3 Mesial pocket of the 7 Slide 4 Mesial pocket of the 6 Slide 5 Mesial pocket of the 6 Slide 6 Mesial pocket of the 6 Slide 7 Mesial pocket of the 6 Slide 8 Mesial pocket of the 6 Slide 9 Mesial pocket of the 6 Slide 10 Mesial pocket of the 6 Slide 11 Distal pocket of the 6 Slide 12 Mesial pocket of the 6 Slide 13 Mesial pocket of the 6 Slide 14 Mesial pocket of the 6 Slide 15 Mesial pocket of the 6 Slide 16 Mesial pocket of the 6 Slide 17 Mesial pocket of the 6 Slide 18 Mesial pocket of the 6 Slide 19 Distal pocket of the 6 Slide 20 Mesial pocket of the 6 Slide 21 Mesial pocket of the 6 Slide 22 Mesial pocket of the 6 Slide 23 Distal pocket of the 6 Slide 24 Distal pocket of the 6 Slide 25 Mesial pocket of the 6 Slide 26 Mesial pocket of the 6 Slide 27 Distal pocket of the 6 139 Slide 28 Mesial pocket of the 6 Slide 29 Mesial pocket of the 5 Slide 30 Mesial pocket of the 7 Slide 31 Mesial pocket of the 7 Slide 32 Mesial pocket of the 5 Slide 33 Distal pocket of the 6 Slide 34 Mesial pocket of the 7 Slide 35 Mesial pocket of the 6 Slide 36 Distal pocket of the 6 Slide 37 Distal pocket of the 6 Slide 38 Mesial pocket of the 6 Slide 39 Mesial pocket of the 6 Slide 40 Mesial pocket of the 6 140 Perio Part 2 Slide 1 Mesial pocket of the 7 Slide 2 Distal pocket of the 6 Slide 3 Mesial pocket of the 6 Slide 4 Distal pocket of the 6 Slide 5 Mesial pocket of the 6 Slide 6 Mesial pocket of the 6 Slide 7 Distal pocket of the 6 Slide 8 Distal pocket of the 6 Slide 9 Mesial pocket of the 5 Slide 10 Distal pocket of the 6 Slide 11 Mesial pocket of the 6 Slide 12 Mesial pocket of the 7 Slide 13 Distal pocket of the 6 Slide 14 Mesial pocket of the 7 Slide 15 Mesial pocket of the 6 Slide 16 Mesial pocket of the 6 Slide 17 Mesial pocket of the 6 Slide 18 Mesial pocket of the 7 Slide 19 Mesial pocket of the 6 Slide 20 Mesial pocket of the 6 Slide 21 Distal pocket of the 6 Slide 22 Mesial pocket of the 7 Slide 23 Mesial pocket of the 5 Slide 24 Distal pocket of the 6 Slide 25 Mesial pocket of the 6 Slide 26 Mesial pocket of the 7 Slide 27 Mesial pocket of the 6 Slide 28 Mesial pocket of the 7 Slide 29 Mesial pocket of the 6 141 Slide 30 Mesial pocket of the 6 Slide 31 Mesial pocket of the 7 Slide 32 Mesial pocket of the 6 Slide 33 Mesial pocket of the 6 Slide 34 Distal pocket of the 6 Slide 35 Mesial pocket of the 6 Slide 36 Distal pocket of the 6 Slide 37 Mesial pocket of the 6 Slide 38 Distal pocket of the 6 Slide 39 Mesial pocket of the 6 Slide 40 Distal pocket of the 6 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0090121/manifest

Comment

Related Items