- Library Home /
- Search Collections /
- Open Collections /
- Browse Collections /
- UBC Theses and Dissertations /
- Machine learning systems for obstetric ultrasonography
Open Collections
UBC Theses and Dissertations
UBC Theses and Dissertations
Machine learning systems for obstetric ultrasonography Porto, Lucas Resque
Abstract
Prenatal screening and ultrasound-guided epidurals are two common applications of ultrasound imaging in obstetrics that help with disease prevention and pain relief. While urban settings typically provide the expertise to perform these procedures, rural and underserved settings suffer from a lack of ultrasound equipment, training, and expertise that precludes a similar quality of care. This thesis seeks to address gaps in equipment and expertise by presenting machine learning systems for automating the analysis of conventional ultrasound images without the use of specialized equipment. The first system is a method for automatic segmentation of the placenta from 2D ultrasound sweeps acquired during first-trimester prenatal screening. We analyzed the performance and speed of four different deep learning architectures for spatiotemporal segmentation applied to 133 ultrasound sweeps from a diverse patient population. Compared to manual segmentations, the top-performing architecture achieved a Dice coefficient of 92.11 +/- 7.5 % and was able to segment at a rate of 100 frames per second. The second system is a method for 2D ultrasound image augmentation for improved interpretability during ultrasound-guided epidurals. This system relies on registering a 3D statistical shape model of the lumbar vertebrae constructed from computerized tomography scans of the lumbar spine to automatically classified and segmented 2D ultrasound images. The classification and segmentation of ultrasound images achieved an accuracy of 90% and a mean Dice coefficient of 74.9 +/- 4.9%, respectively. The registration to the segmented regions was evaluated on 43 ultrasound images, and the achieved a root mean squared error of 1.4 +/- 0.3 mm when compared to the ground truth. We showcase the ability of machine learning systems to automate ultrasound image analysis in common obstetric applications. The results show the potential for these systems to be further developed in a translational research setting.
Item Metadata
Title |
Machine learning systems for obstetric ultrasonography
|
Creator | |
Publisher |
University of British Columbia
|
Date Issued |
2020
|
Description |
Prenatal screening and ultrasound-guided epidurals are two common applications of ultrasound imaging in obstetrics that help with disease prevention and pain relief. While urban settings typically provide the expertise to perform these procedures, rural and underserved settings suffer from a lack of ultrasound equipment, training, and expertise that precludes a similar quality of care. This thesis seeks to address gaps in equipment and expertise by presenting machine learning systems for automating the analysis of conventional ultrasound images without the use of specialized equipment. The first system is a method for automatic segmentation of the placenta from 2D ultrasound sweeps acquired during first-trimester prenatal screening. We analyzed the performance and speed of four different deep learning architectures for spatiotemporal segmentation applied to 133 ultrasound sweeps from a diverse patient population. Compared to manual segmentations, the top-performing architecture achieved a Dice coefficient of 92.11 +/- 7.5 % and was able to segment at a rate of 100 frames per second. The second system is a method for 2D ultrasound image augmentation for improved interpretability during ultrasound-guided epidurals. This system relies on registering a 3D statistical shape model of the lumbar vertebrae constructed from computerized tomography scans of the lumbar spine to automatically classified and segmented 2D ultrasound images. The classification and segmentation of ultrasound images achieved an accuracy of 90% and a mean Dice coefficient of 74.9 +/- 4.9%, respectively. The registration to the segmented regions was evaluated on 43 ultrasound images, and the achieved a root mean squared error of 1.4 +/- 0.3 mm when compared to the ground truth. We showcase the ability of machine learning systems to automate ultrasound image analysis in common obstetric applications. The results show the potential for these systems to be further developed in a translational research setting.
|
Genre | |
Type | |
Language |
eng
|
Date Available |
2020-10-29
|
Provider |
Vancouver : University of British Columbia Library
|
Rights |
Attribution-NonCommercial-NoDerivatives 4.0 International
|
DOI |
10.14288/1.0394860
|
URI | |
Degree | |
Program | |
Affiliation | |
Degree Grantor |
University of British Columbia
|
Graduation Date |
2021-05
|
Campus | |
Scholarly Level |
Graduate
|
Rights URI | |
Aggregated Source Repository |
DSpace
|
Item Media
Item Citations and Data
Rights
Attribution-NonCommercial-NoDerivatives 4.0 International