Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Algorithms for detecting and segmenting nucleated blood cells Poon, Steven Sui-Sang 1989

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

Item Metadata

Download

Media
831-UBC_1989_A7 P66.pdf [ 7.56MB ]
Metadata
JSON: 831-1.0064935.json
JSON-LD: 831-1.0064935-ld.json
RDF/XML (Pretty): 831-1.0064935-rdf.xml
RDF/JSON: 831-1.0064935-rdf.json
Turtle: 831-1.0064935-turtle.txt
N-Triples: 831-1.0064935-rdf-ntriples.txt
Original Record: 831-1.0064935-source.json
Full Text
831-1.0064935-fulltext.txt
Citation
831-1.0064935.ris

Full Text

ALGORITHMS FOR DETECTING AND SEGMENTING NUCLEATED BLOOD C E L L S  by STEVEN S U I - S A N G POON, A.Sc.,  The U n i v e r s i t y o f  P.Eng.  B r i t i s h Columbia,  1985  A T H E S I S SUBMITTED I N P A R T I A L F U L F I L L M E N T OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF A P P L I E D S C I E N C E  in THE FACULTY OF GRADUATE STUDIES Department of  Electrical  We a c c e p t t h i s to  the  thesis  required  Engineering  as  conforming  standard  THE U N I V E R S I T Y OF B R I T I S H August © Steven  COLUMBIA  1989  Sui-Sang  Poon,  1989  In presenting degree freely  this  thesis  in partial  fulfilment  of the requirements  at the University of British Columbia, I agree available for reference  copying  and study. I further  that the Library shall make it  agree that permission for extensive  of this thesis for scholarly purposes may be granted  department  or  by his or  her  representatives.  for an advanced  It  is  by the head of my  understood  that  copying or  publication of this thesis for financial gain shall not be allowed without my written permission.  Department  of  SLGar/e/C/jL  &J&  The University of British Columbia Vancouver, Canada  DE-6 (2/88)  ii  Abstract  The a n a l y s i s o f  the  different  in today's medical practice  types  of  cells  to g i v e an i n d i c a t i o n o f  health.  Many i m a g i n g  the  30 y e a r s i n a n a t t e m p t t o a u t o m a t e  last  systems cells is  the  still  A new  undergoing active  the  Cell  and p r o c e s s  algorithms  well  as  present  to  as  by  of  the high  correcting  i n the  those  which cells.  regions  of  extracted  from  determine  if  Results  are  Imaging  normal  show  that  Some o f  of  System, has  this  In  and  system to  cell  images by  greatly  quality  abnormal  work,  detect  cells  and  types.  aberrations  touching  and  clustering  are  and  of  and  a  algorithms  Simple  these  features  particular can  images  rest  of  detect,  segmented  features  class  are of  segment  as  effects  separated  s i n g l e c e l l s are f u r t h e r  and  noise  shading  from the  and  initial  A l l n u c l e a t e d c e l l s as selected  cytoplasm.  segment  The  reducing  to  some new  classification  Spectral information  segmented c e l l s  these  over  been developed  and segment n u c l e a t e d c e l l s f r o m the  just  of  these  abnormal  this  and abnormal  distortions,  nucleus  any  types  s t a i n e d b l o o d smears f o r  The r e s u l t i n g  the  process.  from a microscope.  a c q u i r e d images. to detect  been developed  between normal  various  ( n o n - n u c l e a t e d c e l l s and background).  individual the  among t h e  Analyzer  images  obtain  then u t i l i z e d scene  difference  used  a person's state  have  this  routinely  research.  in Wright's  sub-classification are  algorithms  have been developed u s i n g  nucleated c e l l s  steps  the  differentiation  system,  acquire  systems and  c a n now d i s t i n g u i s h  but  in blood is  are  is the  well into into then  compared  to  cell  exists.  and  classify  iii different errors  in  analyzed.  types of normal and abnormal n u c l e a t e d b l o o d c e l l s . segmentation  accounts  for  approximately  6%  of  The m a j o r the  cells  iv  Table of Contents Page Abstract  ii  Table of  Contents  List  of  Tables  List  of  Figures  iv vi vii  Acknowledgements  • .viii  1.  Introduction  2.  Automated Blood A n a l y z e r  3.  4.  1 ;  2.1  Image C y t o m e t r y  Systems  2.2  Flow Cytometry  2.3  G e n e r a l Imaging System D e s i g n  2.4  Cell Analyzer  Systems  6 6 7  Imaging System  Segmentation Techniques  ....8 11 15  3 .1  Overview  15  3.2  Thresholding  3.3  Edge D e t e c t i o n  18  3.4  Region E x t r a c t i o n  21  or C l u s t e r i n g  Blood C e l l Analysis Algorithms  15  23  4.1  Overview  23  4.2  Image A c q u i s i t i o n  25  4.3  Image C a l i b r a t i o n  27  4.4  Recognition  32  4.5  Boundary D e t e c t i o n  Single Cells  38  4.6  N u c l e u s and C y t o p l a s m Segmentation  40  4.7  Simple Feature  45  of  Nucleated C e l l of  Extraction  V  5.  6.  7.  D i s c u s s i o n and R e s u l t s  48  5.1  Data C o l l e c t i o n  48  5.2  Detection Accuracy  49  5 .3  Segmentation Accuracy  52  5.4  Feature  61  5.5  Cell Classification  Calculation Accuracies  C o n c l u s i o n and F u t u r e  63  Suggestions  67  6.1  Overview  67  6.2  System Performance  68  6.3  Future  68  6.4  Summary o f A u t h o r ' s  Plans Contributions  Bibliography  Appendix A.  C l a s s i f i c a t i o n Codes f o r  70 72  B l o o d C e l l Types  79  vi  List of Tables Page I.  Segmentation E r r o r s i n Non-Touching N u c l e a t e d C e l l s  58  II.  Percentage E r r o r s i n Non-Touching N u c l e a t e d C e l l s  58  III.  Segmentation E r r o r s i n Touching N u c l e a t e d C e l l s . . .  59  IV.  Percentage E r r o r s i n Non-Touching N u c l e a t e d C e l l s  59  V.  Percentage Segmentation E r r o r s i n N u c l e a t e d C e l l s  60  vii  List of Figures Page 1.  Typical Wright's  S t a i n e d Smear  3  2.  B l o c k Diagram of  a T y p i c a l Image C y t o m e t r y  3.  Layout of  4.  B l o c k Diagram of  C e l l A n a l y z e r Imaging System  13  5.  S e l e c t i n g a Threshold i n a Bi-modal Histogram  17  6.  Use of  19  7.  B l o c k Diagram of  8.  N o i s e R e d u c t i o n Due t o  9.  Noise Reduction Using Background S u b t r a c t i o n  31  10.  Cluster Plots of  33  11.  Segmentation of Nucleated C e l l s  12.  S m o o t h i n g the N u c l e a t e d C e l l Mask  13.  Boundary D i r e c t i o n  14.  Separation of Touching C e l l s  41  15.  Histogram of  42  16.  Process to  17.  M i n o r and Major  18.  Minor  19.  Minor Nucleus Errors  55  20.  Major  56  21  Major Nucleus Errors  57  22.  Feature C a l c u l a t i o n Accuracies  62  23  Cluster Plot of  t h e Mean I n t e n s i t i e s  64  24  C l u s t e r P l o t of  the  65  System  C e l l A n a l y z e r Imaging System  Clustering  the  i n Object the  Discrimination  Procedures to A n a l y z e B l o o d C e l l s Image A v e r a g i n g  S p e c t r a l Images o f B l o o d C e l l s  Codes  Segmented N u c l e a t e d C e l l s  Segment N u c l e u s a n d C y t o p l a s m Errors  i n Separating Touching C e l l s . .  Cytoplasm Errors  Cytoplasm Errors  P e r i m e t e r and R a t i o of A r e a s  9 12  24 28  35 ..37 38  44 51 54  viii  Acknowledgements  I  would  Cancer  like  to  thank  Research Centre  specifically  want  all  who h a v e  t o thank  their  time  and e x p e r t i z e  would  also  like  patience thank  advice  and time  as w e l l  members  this  of  thesis.  the B . C . I  would  D r . B o n n i e M a s s i n g a n d D r . Kamer T e z c a n f o r  this  Beddoes, f o r as t h e i r  t o my s a t i s f a c t i o n .  I  than  I  Finally,  I  would  like  a n d my c o - s u p e r v i s o r s , D r . B r a n k o their  patience  generosity  i n sharing  in allowing  am p a r t i c u l a r l y  f o r the use o f h i s f a c i l i t i e s  blood c e l l s .  1000  t o J o y c e Mak f o r h e r t i m e a n d  documentation.  D r . Rabab Ward,  and Dr. Michael  Palcic  a s s i s t e d me i n  i n c l a s s i f y i n g more  i n h e l p i n g me w i t h  my s u p e r v i s o r ,  work  and student  t o e x p r e s s my g r a t i t u d e  Palcic  this  the s t a f f  i n debt  me t o  their  complete  to D r . Branko  a t the B . C . Cancer Research Centre  i n w h i c h I p e r f o r m my e x p e r i m e n t s a n d r e s e a r c h .  Chapter 1 Introduction  One  of  t h e most  useful  indicators  today's medical practice is  that  the  (erythrocytes, to  stress,  1979).  The e r y t h r o c y t e s , is  cells  as w e l l  responsible  are responsible  body's  immune  agents  to  as  the  the  (Zucker-Franklin is  detecting layer  a  et  study  of  the  patient.  of  blood  on  of the body.  the  i l l  and other  of cell  the morphology  conditions  which  leads the taken  to  of  under  the  various from  cells  sensitive and  other  synthesize and  and i n a c t i v a t i n g  surface of  oxygen  bleeding the  wound.  the c e l l s they  development  patients'.  o f the blood c e l l s  of  by  i n the blood  which  types  carbon  of  blood It  is  infective  agents  a l . , 1 9 8 8 ; Begemann a n d R a s t e t t e r ,  blood  blood  The l e u k o c y t e s o r w h i t e  prevent  o f each type in  in  The r e a s o n  radiation  transporting  The p l a t e l e t s  composition of the mixture of  for  antibodies  and segmenting  deposits  health  a l . , 1 9 8 8 ; Begemann a n d R a s t e t t e r ,  producing  as the i r r e g u l a r i t y of  types  ionizing  destroying  i n t h e number  indications  of  i n the body i s h i g h l y  for  system.  clot  abnormality  for  state  commonly known a s r e d b l o o d c e l l s ,  to and from v a r i o u s p a r t s  cells  person's  various  poisons,  (Zucker-Franklin et  which  blood  thesis  the  diseases,  stimuli  well  of  leukocytes and p l a t e l e t s )  noxious  dioxide  a  i s the examination of blood c e l l s .  production  injuries,  hemoglobin  of  for  the  acting  as  Thus  the  sample as  are  important  were  produced  1979).  This  algorithms cells  thought  is linked with  in  single  that the  for  the  health  2 A common blood  technique  smear  through  spreading a thin such  that  as  platelets  the  within  et  al. ,  The  dark  and  al.,  of  1981)  Although  there  expense  of  types  such  is  general  this  classification the  by  in  which  c a n be u s e d f o r  a large  types  these  number  cells  a semi o r  cells  is  as  generally  blue.  In  been  group  attempts  (Bennett  criteria  for  c a n be  the  prescribed. scheme,  inconsistencies as  objective  well  the in as  evaluation quantitative  automated  measured,  et  sub-classification  c a n be a c h i e v e d by fully  The  human e r r o r .  have  the  and  are  institutions,  some f o r m o f  This  of  to  there  the  nucleus  smears  classification  and  diagnosis. using  these  treatment the  stain,  respectively.  of  as w e l l  different  same i n s t i t u t i o n w a r r a n t  of  blue  both  blood  Rastetter,  a n d when u s i n g W r i g h t ' s  appropriate  among  for  Begemann a n d  standardize  of  Wright's  stain  and are s t a i n e d l i g h t  analysis  observers  visual  standard  (FAB) c o o p e r a t i v e  acceptance  manual  those  measurements  that  for  the  light  by  stained  microscope.  have  a  then  a  1988;  in  produced  of  leukemia,  to  is  cells  enhanced  and s u s c e p t i b l e  the m y e l o i d and lymphoid  these  is  classification  time-consuming  et  The s m e a r  leukocytes  cytoplasm material  and  the  are  field  French-American-British  of  cell  h a s become  colour.  examine  a glass s l i d e which the  and c l a s s i f i c a t i o n  Miller  1).  h a v e no n u c l e u s  stained  tedious,  the  division  Since  1800s,  visually  (Figure  magnified  interpretation  diagnosis  1976;  in  are  have o n l y  subjective,  made b y  red  to  of  (Zucker-Franklin  and  visual  al.,  in  light  cytoplasm  the  the  The e r y t h r o c y t e s  appear  is  blood over  features  under  analysis  1979).  The  f i l m of  developed  smear  analysis  a microscope  various  interpretation stain,  of  a fully  procedure. automated  3  Figure 1  TYPICAL WRIGHT'S STAINED BLOOD SMEAR  A t y p i c a l blood  smear c o n t a i n s a) r e d b l o o d c e l l s ,  c)  and d)  platelets,  debris.  microscope m a g n i f i c a t i o n .  This  b) white b l o o d  photograph  was  taken  cells, a t 40x  4 system  must  practically  The m a j o r is  their  These  eventually  deficiency of inability  years,  eighties.  segmentation,  to  differentiate systems  lack  the  In a d d i t i o n , feature  difficult  the  its  a  (nucleus has  to  pixels machine defined  to  to  be  types of  designed of  the  over  abnormal  cells.  the  last  ten  resort  to  can  transducers  and  optical  and  which involve  cell  the  classification  of  different  is  that of  i n an automated  blood  segmenting the  image  correctly  the  nucleus  cytoplasm of  the  non-nucleated c e l l s ,  easily of  the  quantitatively  regions, means.  a  task  misinterpretation  of  cell  image,  into  describe which  the  and  in  processing  object the  determine  the  Any e r r o r s  extraction,  the  detection,  These are  digital  w h i c h make up can  to  undergoing research.  the  and cytoplasm)  feature  is  and the  cytoplasm and  Because o f n e u r o l o g i c a l p r o c e s s i n g which o c c u r s i n the  observer  qualitative  approach  and a l s o the most c r u c i a l s t e p  nucleated c e l l s ,  human  were  the a l g o r i t h m s  extraction  main components.  background.  the v a r i o u s  quality  smear c l a s s i f i c a t i o n a l g o r i t h m into  this  a s t h e p r o c e s s i n g p o w e r w h i c h became a v a i l a b l e i n  types of c e l l s are s t i l l  The m o s t  if  the c o m m e r c i a l l y a v a i l a b l e c e l l a n a l y s i s systems  imaging  and  components as w e l l late  developed  implemented.  commercial  twenty  be  the  outline image.  techniques distinct the  can  various  A machine, to  relate  regions.  features  human  the  cells  in  the  scene.  eye, parts  set  Nonetheless, from  estimate  introduced i n segmentation w i l l c l a s s i f i c a t i o n and w i l l  the  however,  the  extracted only  of  a  these using  propagate  p o s s i b l y l e a d to Therefore,  of  a  correct  5 segmentation techniques trivial  is  of  for  accurate  thesis  for  recognition,  nucleated Analyzer  is  blood  segmentation, cells  using  cytometry  d i s c u s s i o n of This  in  new  extraction,  image  is  of  not  not  supercede the accuracy of  the  The  this  body  characteristics evaluates  of  the  of  systems  a  discussion  thesis  outlined  in  Chapter  some  three.  the  cells  in  the  the  algorithms  nucleated  blood  cells,  in  segmenting  information using to  the the  enhancing  performance  presented in  of  object  object  these  methods  Chapter four.  such  as  nucleated  chain  boundary,  average  of  filter,  these algorithms  code  the  are  in  used f o r  information  are  gives  the this  i s examined i n Chapter  an  segment into  analyzing  the  separation the  the  blood  incorporated  of  and  for  analyzing  r e s e a r c h e r s to  in  less  segmentation  section  f i l t e r i n g of  introduced  of  Chapter  a n a l y z e the  existing  manipulation the  Cell  with  New a p p r o a c h e s f o r  cells,  and the  the  of  properties  systems  This  the d i f f e r e n t methods u s e d by o t h e r Parts  analysis  begins  i s used to  of  algorithms  those developed  currently  system which  The  image.  the  image c y t o m e t r y  thesis.  is  the  o f new  are c o m p u t a t i o n a l l y  of  edge  cells  system,  measure  indication  tangent  development  and  cytometry  These a l g o r i t h m s  and d e s c r i b e s the  techniques  cells  of  development  feature  a  images.  systems.  section  blood c e l l s cells  the  segmentation  w h i c h was d e s i g n e d a n d a s s e m b l e d t o  e x p e n s i v e and match i f  two.  consistent  However,  concerned w i t h the software  b a s e d on t h e i r  other  and  importance.  task.  This  cells  a paramount  of  spectral touching  angle  of  image u s i n g chapter.  five.  the an The  6  Chapter 2 Automated Blood Analyzer  2.1  Image C y t o m e t r y  The  necessity  vision  and  for  of  the  1952).  a l . ,  analysis All  cell  then,  of  computer This  scene  interest  et  control  level  success  a l . ,  into  from  decades  Kline  these  the  Beckman,  early  (Young  data  to  slide  to  have  by  superior  the  performance  1982) and t h e 1976).  from  the  view,  entire  c a n be  over  in  1970; Megla,  Works,  process.  and  humans  seen by  the  the  companies last  two  1973; Norgren,  Hematrak diff3 in  and a  programs.  different  Japan)  the  and a  reproducibility,  (eg.  the  for bringing  sensor's  systems  analyzers  Glass  other  1985; Preston,  analyzers by f i v e  a n d two  Corning  1951;  (eg. Bengtsson  of the screening  1 9 8 7 ; Imgram a n d P r e s t o n ,  LARC  the  automate  a l . ,  some r o b o t i c s  cytometry  These  with  of a sensor f o r transforming  to the performance  1981).  et  accuracy, uniformity,  States  machine  and R o b e r t s ,  and s u p e r v i s i n g  of leukocyte  of  1950s  b e e n made  and Lundsteen,  image,  automated  United  and Graham,  Electronics)  the  of quality  (Preston,  Kulkarni  the  have  some t y p e  a digital  has brought  commercial production (three  application  1 9 7 9 ; Shoemaker  on the microscope  of  in  the  analyzers  (eg. P h i l i p  analyzing  technology  to  such as the s c r e e n i n g o f c e r v i c a l c e l l s  o f chromosomes  for  led  attempts  these systems i n c o r p o r a t e  area  has  microscopy  blood  1979; Tucker  microscopic  The  to  Since  areas of medicine et  automation  robotics  introduction Walton,  Systems  by  by  Smith  Coulter  classifying  the  7 slides  as  normal  or  abnormal  differentiating  the  deficiency  economical  States  and  have  industrial analysis  different  types  their  production  interest,  there  is  of  abnormal  blood  still  cells  (1986)  a n d Haussraann a n d L i e d t k e  (1983)  i n the  and  made i n t o  of  cells  Pressman,  using 1987)  orifice  at  microseconds.  Laser l i g h t  and the are  processed. as  many  types  of  the  reagents,  which  for  started  is  transmittance The  in in  screening  in  Despite  such  in  as  the the  United lack  of  automating  Palcic  and  the Jaggi  i n the U n i t e d S t a t e s , Aus e t  al  i n Germany, and Landeweerd e t  al  white  blood  development tag With  the  the  1960s.  (5000  shone a t  view  data  cells new  is are  is  are passed through  an  in  as  transferred capable  good  image  immunologically classes of the  blood  flow  and r e c o g n i z i n g  A  cell  can  approximately  droplets,  not  1965;  suspension  cells/s).  of  the  (Fulwyler,  Fluid  and f l u o r e s c e n c e a t  new m a r k e r s ,  quickly  technology  droplets  a field  different  these  samples  of  the  speeds  These systems p r e s e n t l y  problem.  groups  this  companies  research  fluid-flow  passes through  measured.  However,  this  the  high  it  light  by  Because of  cells.  1986.  active  (1984)  and c e l l s  examined w h i l e  of  by  in  Systems  tiny droplets  cell,  three  capabilities  Netherlands.  interrogation  the  the  abnormal  Bacus and G r a c e (1987)  Flow Cytometry  The a n a l y s i s  lack  the  stopped  i n Canada,  Tyrer  of  reasons,  (1989)  2.2  but  w h i c h may various to of  a  wavelengths computer  and  differentiating  cytometry based cells,  rare  four  contain  systems.  biochemical may  overcome  systems c o u l d be the  be  abnormal  ideal cells  8 accurately. detected, for  Since  do  not  visual  observation  of  the  cells  detected  purposes.  G e n e r a l Imaging System Design  A l t h o u g h many i m a g e c y t o m e t r y similar  in  design  illumination camera,  source,  (Figure  moving o b j e c t s Z direction from the The  and  digitizing  processors  in  is  computer  operation.  circuitry, 2).  for  is  through  employed  difference and  dimensional  the  The where  such  it  Tucker,  1979).  stage  scans the  The  detectors  sample to a  digital  the  an  stage,  monitor, capable  field is  and  of  of  view.  transmitted  camera  detector.  digital  image  by  image  is  stored  the  monitor  is  the  in  the the  and/or  computer.  used  in  imaging  systems  scanning.  those  image  is  s l i d e by moving the  V e r y few s y s t e m s use a l i n e a r  Light  very  of  mechanical  samples i s  c a n be d i s p l a y e d o n  these  as  consists  display  the  the  into  found  dimensional array charge coupled device 1988;  to  the  resulting  among  method  detector  motorized  memory,  transformed  p r o c e s s e d and a n a l y z e d by the  major  image  system  focussing purposes.  source  circuitry. from  basic  optics,  the X and Y d i r e c t i o n  image  memory  The  The s t a g e w h i c h h o l d s  provided  detected  systems have been d e v e l o p e d , they are  microscope  illumination  interfacing  The  allow  a b l o o d s m e a r m u s t b e made when a n a b n o r m a l s a m p l e i s  verification  2.3  they  Most in  tube  systems cameras  use  a  two  or  a  two  (CCD) c a m e r a s ( e g . J a g g i e t  c a p t u r e d by  the  detector  s a m p l e f r o m one f r a m e  detectors  transducer  to  while the  s u c h a s d i o d e o r CCD a r r a y s  al., the next. (eg.  9  A/D Interface  Camera  I Image Memory  Microscope  I Stage Controller  Illumination Source  Host Computer  Figure 2 BLOCK DIAGRAM OF A T Y P I C A L IMAGE CYTOMETRY SYSTEM  A  typical  system  motorized stage, processors.  consists  digitizing  of  an  circuitry,  illumination image memory,  source,  microscope,  d i s p l a y monitor and  10 Jaggi In  and P a l c i c ,  these  sensor  1985;  systems,  a n image  and p i e c i n g  scanners column  is  of  field  the  of  stored  that  requires  of  the the  to  laser to  spot  s c a n the  the  two  of  optical the of  the  also  system i s  image.  distortions,  video a  or  input  are  the of  (video)  These  linear  array  a  a  wider  digitized  and  format  which  The  major  required for  high  (eg.  Graham and  a r e moved i n t o  high a one  resolution element  is  used  to  the  the  image  detector  Ingram and P r e s t o n ,  i n one d i m e n s i o n w h i l e  1970;  deflect  stage  is  the moved  dimension.  in  objective  aberrations,  for  polygon  but  imaging also  on  is the  The m o s t  lens  of  the  dependent optical  important  are  not  not  only  perfect  and s h a d i n g e f f e c t s .  and  the  on  components component  microscope which  sample and hence governs  lenses  linear  a row o r  form.  detectors  rotating  circuitry.  the  the  There are systems which use a  (eg.  image  used  of  provide  (analogue)  photodiode  A  1987).  sample a c r o s s  hence  digital  images.  al.,  advantage  sensor  systems which use  object  (digitizing)  magnification the  a  to  and m a t r i x  1982).  transducers  electronic  the  et  than e i t h e r  and  dimensional detector  are  scan the  of  in  convert  image i n t h e o t h e r  The q u a l i t y type  not  the  The  of  Objects d e t e c t e d by the  al.,  to  Tucker  a precise mechanical scanning is  linear  There  et  lines.  counterpart  elements  and  such as a p h o t o m u l t i p l i e r Shoemaker  image  two-dimensional  view of  acquisition.  the  the  data,  that  1980).  1979;  o b t a i n e d by moving  dimensional  Also,  is  combination of  field  two  re-sampling  r e s o l u t i o n of  Norgen,  together  digital  disadvantage  is  al.,  t h e y h a v e more s e n s o r e l e m e n t s  view.  as  Bengtsson et  Most cameras  and the  determines  sampling  generally  in  the  density  introduce today  are  11 built the  for  the  t e l e v i s i o n b r o a d c a s t c o m m u n i t y w h e r e t h e d e t e c t e d image  transducer  digitized  for  is  converted  not the  image i n t o d i g i t a l  conform  to  a  fixed  dependancy  Distortions circuitry  2.4  on  standard  a  by  for  the  properties Nordin,  as  data  optics, before  elsewhere  dimensional  of  The  description  and  Palcic,  the  for  CCD c a m e r a ,  stained  cells  experiments,  this  upgrades.  and  digitizing  image c a n be a n a l y z e d .  et  and  unstained  and/or of  algorithms  thesis.  al.,  Jaggi,  initial  The  study  (Palcic, design  Poon and  Cell  their  Jaggi  is  Palcic,  shown i n F i g u r e and  to  and  described 1986).  3 and 4 .  A  A  two-  image p r o c e s s i n g b o a r d ,  and  the o r i g i n a l  system f o r  1988).  addition  currently feature  cells  treatments  this  grabbing  was a d d e d t o  system i s  segmentation  live,  system i s  frame  (Jaggi  future  transducer  the  in  the B r i t i s h Columbia Cancer Research  1985;  current  d i s p l a y monitor  automatic  designed at  time  block diagram of  to the  results  S y s t e m was u s e d t o d e v e l o p a n d t e s t  function  (Jaggi  due  System  of  1987) .  later  digitize  transmission which  manufacturer  the  measurement  a  is  information  s e g m e n t i n g the b l o o d smears d e s c r i b e d i n t h i s  A n a l y z e r was o r i g i n a l l y  colour  of  must be compensated f o r  used for  i n a loss of  signal  m a c h i n e a n a l y s i s a r e e x p e n s i v e a n d do  particular  introduced  This  Cameras w h i c h d i r e c t l y  data for  The C e l l A n a l y z e r I m a g i n g  of  errors.  C e l l A n a l y z e r Imaging  Centre  an analog s i g n a l .  machine a n a l y s i s r e s u l t i n g  r e s a m p l i n g and q u a n t i z a t i o n sensor  to  on  In  being used f o r extraction  the  measurements to  live  development  algorithms  of  cell of  stained  12  Microscope  RGB Display •  Camera  Computer Monitor  e nrm  PC/AT  iiiiifflmiiifl  Computer  Power Supply and Controllers  Keyboard  Figure 3  LAYOUT OF CELL ANALYZER IMAGE SYSTEM  The l a y o u t consists computer,  of  of  a  the major  microscope,  computer  monitor,  supply f o r the l i g h t x,y  components  and z p o s i t i o n s .  camera,  of the C e l l camera  keyboard,  A n a l y z e r Imaging  control  and a rack  source and the c o n t r o l l e r s  unit,  System  RGB m o n i t o r ,  containing  the  power  t o move t h e s t a g e i n t h e  13  RGB Display  RGB CCD Camera  I  T  Frame Grabber and Image Processor  Microscope  Stage Controller  Computer Monitor  I  Stabilized Light Source  P C / A T Computer  Figure 4  BLOCK DIAGRAM OF THE CELL ANALYZER IMAGING SYSTEM  The m a j o r  components  illuminating field  into  the microscope o p t i c s  the images,  digital  o n a n RGB  the system are the s t a b i l i z e d l i g h t  t h e s a m p l e , a s t a g e t o move t h e o b j e c t  of view,  acquire  of  format  monitor  digitizing  to magnify  circuitry  to  w h i c h c a n be s t o r e d i n t o  into  for  the microscopes  the image,  transform  source  the camera to  the video  signal  i m a g e memory a n d d i s p l a y e d  or manipulated by a computer.  14 blood  and  cervical  cells  tissues which require  The C e l l Since  stained  CCD c a m e r a  and  well  as  an i n t e r a c t i v e  slides  use of  colours of board  and d i s p l a y the each spectrum, buffers.  the  hardware  histogram processor,  employed to  MVP-AT)  is  processing  p r o c e s s i n g and a n a l y s i s  PC/AT  computer.  F o r b l o o d smear a n a l y s i s ,  an o b j e c t i v e  0.95  air)  condenser  lens  0.33 microns  aperture,  a n d a TV r e l a y  is lens  used (lx)  i n b o t h x and y d i r e c t i o n s .  camera are u s e d , levels.  one o f  images  are  in  and  work. 3-chip  the  three  A frame  grabbing  store,  process  i n p u t c h a n n e l s , one the  four  of  this  performed  lens  a  in  512x512x8 board  a n d 3x3 m o r p h o l o g i c a l  functions  cells  developmental  digitize,  features  3x3 c o n v o l u t i o n s ,  other  information,  capture  used to  e a c h image i n  Other  numerical  for  This board accepts three  and s t o r e s  The  useful  s p e c t r u m s : r e d , g r e e n and b l u e .  (Matrox  image.  of  an imaging s y s t e m .  contain multi-spectral  (Sony DXC-3000A) i s  imaging  256 g r e y  experiments  A n a l y z e r system i s p a r t i c u l a r l y  many  primary  as  by  for  frame  include  operators.  the  host  IBM  ( P l a n Apochromat 40x w i t h conjunction  giving All  a spatial  three  each of which has a photometric  with  colour  a  matched  resolution images o f  r e s o l u t i o n of  a  8 bits  of the or  15  Chapter 3 Segmentation Techniques  3.1  Overview  Many  segmentation  decades  (eg.  Fu  and  categorized  into  thresholding  or  extraction. scene  in  A  Mui,  three  a combination  Detailed  Thresholding or  Thresholding  is  scene.  process  The  properties  of  a  the  as  the  gradient  or  range of  values of  image  ii)  edge  of  is  used  several  methods  and  iii)  segment  a  processes is  the  different  region  particular often  classes  following  be  feature  i n some a p p l i c a t i o n s b u t  of  can  used.  may of  fail known  sections.  Clustering  technique  assigns  image. gray  in  labels  A property  may b e  levels  or  the  gray  the  used  distinct  a given property  to  cannot  segmentation  past  characteristic  detection,  are d i s c u s s e d i n the  common  to  i)  generally  the  These  1975).  classes:  descriptions  Laplacian of  image w h i c h b e l o n g property  different  algorithm  segmentation algorithms  such  Davis,  1981;  these processes perform w e l l  others.  3.2  have been d e v e l o p e d over  clustering, single  and hence  Generally  techniques  may b e  same r e g i o n .  determine  the  to a  of  levels. is  segmenting  In  Often,  based  nature  all  cases,  a  define  the  pixels  for  each  a  some  feature  such  a histogram  in  on  characteristic  local  used to  thresholds  areas  regions  as  the  specified  of  region.  an  in  the  image These  16 histograms however,  are  to  generally  a v o i d smoothing out  A thresholding is  the  smoothed to  technique  mode m e t h o d .  c a n be a p p l i e d  type  of  histogram  to  each  intensity  peak  (mode)  level.  of  A boundary  separate  the  regions.  minimize  the  probability  number  of  pixels  small,  misplacement  relatively  cells  A  the  Liedke  different  on the  (1984)  technique  these  a given  histograms gray  differences  used  but  intensity  fails pixels  generated  at  the  threshold  level.  with  of  the  others.  of  for  the  image  (Figure  5).  areas  of  such  points  region. peaks  from  the  exact  is  is  to to  Since  the  relatively  location  For example, to  similar  between peaks  the  image.  histograms of  to  technique  the  thresholding sum o f is  image.  the  segment  the  gradient magnitude  placed at  signifies  has  Wermser, the  blood  the  location  histograms. of  highest of  gradients point'  the  images  objects  where their and  there  are  many  sum may o v e r m a s k hence  generate  in  largest  T h i s method works w e l l w i t h  gradient,  rare  each  level  image.  For  a small  edge  this  boundary  point  edges) in  the  used  the  This  (the  resulting  valley  choosing  compared  represent  level,  histogram.  images  is  for  threshold  from the background of  Since at  and  effect  the  i n the  the  misclassifying  valley  of  taken,  an i n d i c a t i o n  represents  placed at  rationale  of  the  histogram  is  The  at  little  Haussmann  the  gray  gives  number o f p i x e l s w h i c h h a v e t h e same g r a y l e v e l Thus,  C a r e m u s t be  s m a l l b u t v a l i d minima o r maxima.  which  This  remove n o i s e .  some  similar the a  sum  wrong  17  Figure 5  SELECTING A THRESHOLD IN A BI-MODAL HISTOGRAM  The  threshold  error  in  boundary  misplacing  o t h e r p o i n t s (X2).  is  placed at  the boundary  is  a valley less  at  between  the peaks.  the v a l l e y  (x-^)  than  The at  18 Clustering  extends  the  dimensional  space.  This  exists  employing  technique technique  a single  using histograms of  feature  images  features,  seen through etc.,  thresholding  is  but  used  when  distinct  two o r more c h a r a c t e r i s t i c  feature which i s useful for of  of  c a n be u s e d .  spectral  poor  features  filters,  (1984)  cells  for  l o c a t i n g the d e c i s i o n boundary between r e g i o n s  amount o f  computations  required  and  some  aggregate  very  dependent  Although  the  smoothing information  clustering  properties on the  type  segmented  to is  eliminate used  in  analysis  1988).  a n a l y s i s , the  the r e g i o n s  techniques  is  are  s m a l l e s t number  selection  some  These  the  of  use  features  are  images  boundaries. of  the  which  r e g i o n s which are segmented i n  noisy  features.  operators  of  the  multi-  employed.  global  closed,  been  To r e d u c e  different  are  have in a  of  regions  the  cluster  1 9 8 8 ; Umesh,  i n the  features which can d i s c r i m i n a t e  Thresholding  for  the  texture  nucleated  blood  d i m e n s i o n a l s p a c e (Amadasun and K i n g ,  separate  levels  green  red  Algorithms  gray  Any  use the  from  available  to  detected  gradients,  image components cells.  image  multi-  (Figure 6).  and b l u e the  the  c a n be  s u c h as the  Haussmann and L i e d t k e  of  the  discrimination  regions  segmenting a r e g i o n ,  different  to  Since  threshold,  the  image.  may  require  no  spatial  the  resulting  r e g i o n s may n o t b e c o n t i g u o u s .  3.3  Edge D e t e c t i o n  Edge  detection  determine  the  algorithms boundary  use  between  the  information  objects.  The  edge  of  edge  points  points are  to  located  Figure 6 USE OF CLUSTERING I N OBJECT D I S C R I M I N A T I O N  Objects  in  g r o u p A c a n be e a s i l y s e p a r a t e d f r o m  two-dimensional the  (x,y)  feature  one-dimensional feature  plot.  plots.  This  is  those  not  the  i n group case  in  B in either  the of  20 where  there  is  technique, first  an abrupt  the  elements  extraction  the  change  methods,  of  in  edge  gray  gray  are  such  as  purpose.  convolutions each  the  of  extracted  of  candidates  the by  the  weights  The changes  of  the  the  resulting around  thresholding  zero-crossings  in  belong  its  filter  a space constant  and  convolution gives  image.  filter.  correspond to than  the  (or  In  a lower  are  because  the  transition  from  edge  one  detection  region  of  sometimes o c c u r s over s e v e r a l p i x e l s and i s contours than  produced from t h r e s h o l d i n g  one  pixel  processing Another enough  using  problem to  be  wide  and  thinning is  that  thresholded  not and the as  contour-closing  texture edge  of  points  some  of  the  Marr  (1982)  has  this  method,  the  the  structures  spatial  to  abrupt  frequency  are  in  the  is  other The  generally  more  Hence,  regions  This  enough.  algorithm  resulting  different  then  image  then not  closed.  image  are  techniques.  the  for  process.  edge i n f o r m a t i o n  necessarily  is  points  edges o f  There  with  operators  indication  edge  to  Various  Prewitt  an  processed edge  are  corresponds  kernel  a s e l e c t e d v a l u e used i n the Gaussian b l u r r i n g  problems  an edge  implemented as a s e r i e s o f  than)  several  this  have been developed  The  greater  In  surrounding.  Kirsch,  value  a Gaussian  to  which  each p i x e l .  the  image.  boundary.  with  Sobel,  c a n be  the  a measure  pixel  gradient,  developed a L a p l a c i a n of  which have  the  in  to  form the  requires  These o p e r a t o r s where  levels  1982; Young and F u , 1 9 8 6 ) ,  filter.  strength  pixels  value  ( R o s e n f e l d and Kak,  for  which  in  e x t r a c t e d and t h e n combined to  The  this  change  are  some is  post-  required. significant  erroneous  image  21 segmentation.  Nevertheless,  the  results  from  t e c h n i q u e s c a n be u s e d i n c o n j u n c t i o n w i t h o t h e r particular  3.4  edge  methods  in  detection determining  regions.  Region Extraction  Another  approach  properties, regions.  to  segmentation  s u c h as g r a y  region  merging,  r e g i o n merging and  many  small  pixels. object are  Various  adjacent one.  regions  This  properties  regions  group color  pixels  with  information,  similar  etc.,  into  t e c h n i q u e s c a n be s e p a r a t e d i n t o splitting,  its  techniques,  such  properties  as that  and  a  three  combination  a  the  of  regions.  If  and merging  adjacent  regions  have  neighbourhood  segmentation  significantly  the  of  each  region  properties  different  the the is  object  of  the into  membership  regions  which  have  when  all  completed  properties  of the  a r e combined o r merged  each enlarged r e g i o n The  divided  characteristics  recomputing  characteristics.  initially  small  by  similar  is  The c h a r a c t e r i s t i c s o f  these regions  iterated  image  or  reflect  neighbouring  is  the  a pixel  each r e g i o n .  are s i m i l a r ,  process for  to  texture,  region  growing  a r e computed f o r  compared w i t h  is  splitting.  r e g i o n merging or  into  levels,  These r e g i o n e x t r a c t i o n  categories:  In  the  such  that  no  merge c a n f u r t h e r b e m a d e .  The r e g i o n instead  of  splitting many  or  dividing  small  regions.  techniques begin with A  predicate  the  describing  entire the  image  various  22 properties  of  the  example  to  determine  if  all  differ  a  certain  which  is  do  not  satisfied, for  the  each of  object  by  region is  the  is  all  a n d merge  technique  splitting  to  regions  adjacent  predicate  regions  describing  have the  used t h i s  in his  segmentation method.  Region e x t r a c t i o n  the  the  entire  region If  the  similar  is  of  properties.  properties  they  utilize  Although  produce  drawback i s  that these algorithms  and the  An levels  is  not  predicate  The p r o c e s s c o n t i n u e s  not  and  until  r e g i o n merging Regions are are  satisfied.  t e c h n i q u e on b l o o d c e l l s to e x t r a c t  directly.  gray  predicate  smaller regions  a combination  similar  techniques  have  region.  satisfied.  property  (1987)  in  amount.  uses  of  from  recomputed.  regions are  The s p l i t  obtain  pixels  divided into  sub-regions is  the p r e d i c a t e s f o r  when  evaluated  split  merged  when  Liedtke  et  the p r i m i t i v e s  the  local  properties  of  closed  and  contiguous  regions,  are computationally  the  intensive.  and  the al. used  image the  23  Chapter 4 B l o o d C e l l Analysis Algorithms  4.1  Overview  A program are  for  to  of  in  the  of  features,  first  of  two  required  the  of cells  cells  are  which  isolated cells  into  defined  are  just  is  two  cells  are in  the  and  then  is  the  images w i l l the  all  of  information scene.  All  the  single  touching  required calculated  scene are  is  in  to  nucleated blood are  step  and c y t o p l a s m . fine-tune on  the  classified  cells  the  7).  are the  are  The  critical amount  for  step  the  and  also  boundary  segment  the  of  defined  of  each  region  b a s e d on the  to from cell any  each  isolated  Post-processing of mask  of The  selected Once  the  to  cells  extraction  following  cells.  is  iii)  analysis.  the  and  the  vi)  images,  accounted  separated  based  then  used  steps:  images,  (Figure  the  The  seven  of  and reduce of  The n e x t  nucleus  the  stages  cells  are  regions,  which  cells.  acquired  object  simplify  red  blood  segmentation  segmented  later  non-nucleated  regions:  the  scene, iv)  algorithms  following  the  and p r e - p r o c e s s i n g  determined.  regions  Features  features.  in  found,  cell  in  spectral  background  locations  input  the  pre-processing  i n the  the  nucleated  involves  classification  acquiring  quality  manipulation detect  vii)  incorporate  segment  post-processing  and  to  cells  ii)  possible cells  steps,  processing  these  images,  scene, v)  since high  the  automatically  analyzing  acquisition  detection  The  d e s i g n e d and w r i t t e n  developed  approach i)  was  the  region.  boundaries.  values  of  the  24  IMAGE ACQUISITION  1 IMAGE PRE-PROCESSING  OBJECT DETECTION  I IMAGE SEGMENTATION  I  IMAGE POST-PROCESSING  i  • ~  F E A T U R E EXTRACTION  i OBJECT CLASSIFICATION  Figure 7  BLOCK DIAGRAM OF THE PROCEDURES TO ANALYZE BLOOD CELLS  Algorithms  a r e employed  to  that  ensure  Errors limit are  images  i n the detection their  propagation  extracted  object  the  i n the a c q u i s i t i o n used  in  the  and segmentation to  the l a t e r  from t h e segmented  i s made b a s e d o n t h e f e a t u r e  analysis  sections  parts  regions  and p r e - p r o c e s s i n g  of  are  corrected  for.  s h o u l d be m i n i m i z e d  the a n a l y s i s .  and the c l a s s i f i c a t i o n  values.  sections  to  Features of the  25 The m o s t the  difficult  cell.  will  Unlike  suffice,  geometrical, the  scene  feature  of  as  segmentation  well  as  the  analysis  extraction  morphological,  Since object from the  portion  also  of  Image  Since  each r e g i o n  the  the c e l l s  the  regions  for  segmentation  is  of  algorithms  knowledge  properties  approach  i n the  of  the  of  the  cells  analyzing  the  problem.  crucial  for  in  derived  the  correct  scene.  Acquisition  of  the  blood  multi-spectral  a n a l y s i s i s used,  colour  must be  filters  cell  is  ensure  correct  colour  observed at  (red,  the  a different  colour  and  different  obtained.  a c c o m p l i s h e d by a d j u s t i n g component  stained  two o r more i m a g e s t a k e n w i t h  Camera c a l i b r a t i o n must be p e r f o r m e d a t  is  prior  and t o p o l o g i c a l  segmented r e g i o n s ,  4.2  colour  define  where o n l y m a t h e m a t i c a l requires  a heuristic  to  c l a s s i f i c a t i o n i s b a s e d on f e a t u r e v a l u e s w h i c h a r e  interpretation  to  is  the b e g i n n i n g of  registration  of  the g a i n and o f f s e t  g r e e n and b l u e )  output of  the of  each experiment  image.  This  the a m p l i f i e r s  such t h a t  a similar  for  light  is each  level  e a c h c h a n n e l when o n l y t h e b a c k g r o u n d  light  i s measured.  Light  s o u r c e and frame  experiment the  begins.  digitizer  light  source  (8 and  In bits the  grabber order or  256  gain  calibration  to  utilize  grey and  is  the  levels) ,  offset  of  also full the the  executed before photometric  voltage analogue  range  level to  of  the of the  digital  26 converter level set  of to  of the  the  background,  approximately  saturation. approximately  In  the  are  number o f the  source, be  The five  analysis,  spectrum  has  frame  grabber  board  which  is  below  darkest  stained  nucleus  images  from  each of  Since  the  reducing  the  and d i g i t i z e r . as  a  sum  of  -  S(x,y)  (x.y)  If  the  be  shown  factor  noise that  is  of  maximum  in  the  scene  Since a  red, is  the to  green,  stationary  part  image set  and b l u e an  to  colour  average This  introduced  each detected  is  is  level.  taken beforehand. noise  avoid  image  static,  image,  by  of  averaging the  light  image I ^ ( x , y ) ,  S(x,y),  a  and  a  can part  i.e.  (1)  the  form  M  Z i-1  {S(x,y)  assumed to  the  part  The  + Nj/x.y)  the average o f M images i s o f  1 - J  the  random  c o n t a i n i n g random n o i s e N ^ ( x , y ) ,  I^x.y)  the  accordingly.  p e r c e n t a b o v e t h e minimum d e t e c t a b l e  of  represented  brightest  percent  images from each spectrum i s  detector  I  the  adjusted  five  acquired.  effect  are  power  + N^x.y)}.  (2)  be u n c o r r e l a t e d w i t h  spectrum  of  o f M (Castleman, 1979; P r a t t ,  the  noise  1979),  the  image,  (^NN^ '  *"  s  then r e  it  d u c e d by  can a  i.e.  (3)  27  where If  (Sjr^)  the  is  a v  noise  the  is  assumed  standard deviation of  t h e number o f  a  <7  where Due  is  av  to  image  jk  av "  the is  Achieving required should at  power  a, o f  to  the  have  a  the n o i s e Gaussian  random n o i s e  frames averaged,  is  in  the  averaged  distribution,  reduced by  the  image.  then  the  square  root  i.e.  <*>  i  a  the  standard deviation  quantization  chosen this for  spectrum of  to  limit,  be  at  average each  sample  l e a s t be f o u r  the  most  noise  a  the  noise  standard half  level  (i.e.  times  of  the  of  the  deviation a  implies  grey that  number  the v a r i a n c e  in  of  of  of  level the  frames  the  averaged the (ff  number to  be  image.  averaged <  a v  of  0.5). frames  averaged)  d e t e c t e d random n o i s e ,  i.e.  M > 4 o^  (5)  2  The  standard  acquired frames  4.3  deviation  spectral  averaged for  Image  images  the  ranges  grey  level  of  from 2 to  3.  e a c h image a n a l y z e d ,  is  the  random  Hence,  M,  noise the  in  the  number  c h o s e n t o b e 36 ( F i g u r e  of 8).  Calibration  Although  random n o i s e  must  corrected  be  of  by  can be using  reduced by other  averaging,  means.  These  fixed noise  pattern patterns  noise are  28  H  32 Pixels  Figure 8  NOISE REDUCTION DUE TO IMAGE AVERAGING  The  variation  histogram adjusted was  in  (top)  background can  to h i g h l i g h t  captured  variations.  36  readily  intensity be  seen.  in  the The  original color  s m a l l changes i n i n t e n s i t y  times  and  averaged  (bottom)  mapping  levels. to  image  and  its  table  is  T h e same image  reduce  the  noise  29 produced  by  sensitivity level  various  between elements  values  among  Some  camera  with  slightly  caused by  pixels  control  the  resulting  components of  in the  that  optics  generate  a l s o produce  system.  results  received  intensities.  in a brighter  imaging  detector  have  circuitries  varying  the  the  evenly  Shading  in  different  same  aberration  uneven i l l u m i n a t i o n the  gray  illumination.  spaced v e r t i c a l  and  spot near the c e n t r e o f  Unequal  at  bars  effects  the  detector  image w h i c h f a d e s  to  the edges.  Decalibration  is  one m e t h o d w h i c h  noise.  Each p i x e l  C(x,y),  u s i n g an e q u a t i o n  a dark  image D ( x , y ) ,  nt  This is  corrected  scale  of  0 to  dark  intensity  the  Hx.y)  for  k.  by  the  used to  correct  image I ( x , y )  is  involving  bright  and a c o n s t a n t  represents  and d a r k  to d i s t o r t the  i  \  formula  bright  in  is  the  the  transformed  fixed  into  a new  b a c k g r o u n d image  scaling factor  pattern image  B(x,y),  k:  - D(x.v')  the  transmittance  values  images i s Because  the d i s t r i b u t i o n  .,.  of  the  used to  220.  u s i n g the d e c a l i b r a t e d i n s t e a d of  image where e a c h  map t h e  used,  gray l e v e l s .  image u s u a l l y has a g r a y r a n g i n g f r o m 200 t o  is  the  calibration  linearly  division of  of  l e v e l of Hence,  images.  A range  detected  image  truncation In b r i g h t  error field  0 and the b r i g h t up t o  the raw image.  pixel  is  to  of a  likely  microscopy, image h a s  10% i m p r o v e m e n t  is  a  seen  30 Background  subtraction  of  optical  densities  can  be  used  instead  achieve a s i m i l a r r e s u l t .  T h i s method i s based on the c o n v e r s i o n  transmittance  density  values;  to  optical  by  taking  the dark image i s assumed to be zero  logarithm  in this  of  of  the  constant Taking  value  of  K  logarithms  distribution  of  is  on  the  a  added  case.  offset  discrete  optical  (7)  from the o r i g i n a l o p t i c a l  to  density  image  the  intensity  will  profile  produce  which  l a t e r i n the a n a l y s i s o f the  images.  A third  images i s to s u b t r a c t  then add  approach to  correct  an o f f s e t e q u a l to an average v a l u e  C(x,y) = I ( x , y ) - B(x,y) +  adjusts  b r i g h t background  Most o f the  adjusted  for.  is  introduced  The  in  the  complications  the b r i g h t  pixel  by  cells  image  adding  an  i n t h i s work  and  (Figure  o f f s e t such  image appears to have e q u a l gray l e v e l  pixels.  gaps  (8)  9).  each  a  distribution.  K.  used to a n a l y z e white b l o o d  method  adds  image and  o f the b r i g h t image, i . e .  T h i s approach was The  the  pixel  l o g C(x,y) - l o g I ( x , y ) - l o g B(x,y) + K  where the background i s s u b t r a c t e d  to  value  that  a  for a l l  image c o n s i s t i n g o f background p i x e l s i s c o r r e c t l y  advantage o f t h i s method i s t h a t no  since only  s u b t r a c t i o n of  the approach i s not  as good as the f i r s t  good a p p r o x i m a t i o n  for  image  integers  correction.  two  truncation  i s involved.  Although  methods, i t does serve The  error  i n the  error  as  a  correction  50K +  40K •  C 30K-I u  3  cr g 20K--  h 10K--  +  0 4—V  230  240 250  Intensity  256 Pixels  +\ Figure 9  NOISE REDUCTION USING BACKGROUND SUBRACTION  The  aberration  histogram  (top)  and shading can readily  adjusted to h i g h l i g h t subtraction to  t h e image  effects be  in  seen.  the  original  The c o l o r  s m a l l changes i n t h e i n t e n s i t y  mapping levels.  ( t h e t h i r d method as d e s c r i b e d i n t h e t e x t ) (bottom)  t o remove t h i s  shading  effect.  image  and i t s table  is  Background  was t h e n  applied  32 does i n c r e a s e i n the d a r k e r change  in  the  segmentation  4.4  error  is  r e g i o n s where the  gradual  first  step  nucleated  cells  visual  with  occurrence  of  of  of  Figure of  in  nucleated To  10. the  the  It  colour  cytoplasm of  rest  of  Haussmann and L i e d k e  (1984)  instead  the  time  In  from a l i n e a r  this  two  more  = a G(x,y)  + b  green,  and  two  the  colour  the  B(x,y)  for  spectral histograms  was  of  generated c a n be  separate regions  red blood  the  of  the  cells,  histograms.  p r o p o s e d a method  the green G ( x , y ) ,  used  threshold  However,  stained  for  analysis  peripheral  multidimensional  feature  any  frequency  axis  to  consuming  are  blue  histograms  Pappenheim  is  and the  third  two c o l o u r  approach, c h a r a c t e r i s t i c  combination of  colour  axis,  on the  there  Since  point,  image.  i n the  on  X(x,y)  affecting  if  n u c l e a t e d c e l l s , , the  histogram  analysis.  this its  the  dimensional of  red,  colour  the  and the b a c k g r o u n d form c l u s t e r s  smears  image.  pair  individual  nucleus  one  determine  c a n be s e e n t h a t no s i n g l e  from the  a  to  cells,  on e a c h o f  nucleated c e l l s and the  is  illustrate  a particular  any  Wermser,  in  the  Cell  analysis  interest  information  as shown i n in  the  was u s e d .  colour  used  in  detection  information  insignificant  but  algorithms.  Recognition of Nucleated  The  and  stained cells l i e ,  X(x,y),  is  and b l u e B ( x , y ) ,  of  blood cluster  generated images:  (9)  33  CLUSTER PLOTS OF SPECTRAL IMAGES OF BLOOD CELLS  A photograph  of  (left):  green,  each  red,  of  (from top  the  two  the  intensity and b l u e  colours  to bottom),  red blood c e l l s ,  (C)  are  variations (from top  (right): shown.  cytoplasm,  of  each of  to b o t t o m ) ,  red-blue,  the  and c l u s t e r  red-green,  Clustering of a n d (D) n u c l e u s  the  spectral  (A)  and  images  plots  green-blue  background,  a r e a s c a n be  of  seen.  (B)  34 where  a and b a r e c o n s t a n t s  histogram valley  o f the new image i s generated  of  the h i s t o g r a m  of this  s e p a r a t e the r e d b l o o d c e l l s  A similar in  this  using values  o f a=0.390 and b-0.546.  and a s i m p l e  characteristic  t h r e s h o l d a t the  feature  i s used  to  from the cytoplasm o f n u c l e a t e d c e l l s .  approach c a n be employed w i t h the Wright's work.  A  The c h a r a c t e r i s t i c  f e a t u r e uses  stained cells  the r e d , R ( x , y ) ,  used and  b l u e , B ( x , y ) , images i n a formula:  X(x,y) = R(x,y) - B(x,y) + 1 2 8  This  method  point The  a r i t h m e t i c i s i n v o l v e d which would  introduce  No  floating  truncation errors.  the o r i g i n such  that values  -128 t o 127 a r e mapped t o the v a l u e s 0 t o 255.  extract  histogram  the n u c l e a t e d  blood c e l l s .  averaged  i n the image,  thresholding  histogram  of  t h r e s h o l d T,  consists  The s u b t r a c t e d  o f c o n v o l u t i n g the image w i t h a 3x3  e q u a l to l ' s and d i v i d i n g the r e s u l t by a f a c t o r o f 9. this  near  filtered  the v a l l e y  image i s found  i s produced  are represented  (Figure  by s e l e c t i n g  o f the peak where p i x e l s b e l o n g i n g  background  on the  t o remove s p e c k l e d n o i s e a t the edges o f the r e d  This f i l t e r  k e r n e l o f weights  the l e f t  cells  o f the smoothed s u b t r a c t e d image i s performed.  image i s f i r s t  A  s u b t r a c t i o n o f the images.  c o n s t a n t v a l u e o f 128 i s added t o s h i f t  from  To  i n v o l v e s simple  (10)  11).  the f i r s t  The  point to  t o the r e d b l o o d c e l l s and  i n the h i s t o g r a m  H ( T ) , and which  g r a d i e n t l e s s than 2% o f the maximum peak v a l u e P  m a x  , i.e.  has a  35  Intensity (Grey Levels) F i g u r e 11  SEGMENTATION OF NUCLEATED CELLS  The p h o t o g r a p h at the  the top.  of the blue Because  image h i s t o g r a m  (bottom).  i m a g e s u b t r a c t e d f r o m t h e r e d i m a g e i s shown  of the noise  i n the image,  i s used to determine  a filtered  the l o c a t i o n  of  version  the  of  threshold  36  H(T)  All  and a l l  other  but  higher  also  cells  value  is  the  mask  pixels  will  be  is  holes  0.  The 2% P  belonging selected in  to  cell the  A 3x3 m e d i a n in  the  the  a  x  white  than  mask.  blood  is  image.  Because a b i n a r y  implemented as a l o c a l result A  is  thresholded  single  dilation  applied  to  this  contour  of  the  neighbours  neighbours  removed  the  cells  is  spatial  u s e d as the  128 l e v e l by  image  nucleated  adjacent is  at  binary  a pixel  is  high  3x3 a v e r a g e o p e r a t i o n  followed  process,  pixel  image  any  cell  included belong in  from  to  the  does n o t b e l o n g to  in the  two  the  to  the  mask  larger  the  mask.  if  to  the  a  holes In  mask, i f  any  the of  gradient be  lost  the  cell  remove  the acts  from  filter  the  c a n be  and then  3x3  a red  This f i l t e r  12).  In  more  will on  If  t h e new b i n a r y  using  (Figure  ),  frequencies  generate  c e l l mask. mask  m a x  on e a c h p i x e l  new c e l l  cell  cells.  lower  and  input,  erosions  fill  the  applied  filter  removing  nucleated  a  as  pass  only  If  i n the r e d b l o o d c e l l s and b a c k g r o u n d .  low  threshold  nucleated c e l l s filter  cell  gradient  2% P  unwanted p o i n t s a  255 ( t h e  red blood  (greater  the  m  encompass n o t  cytoplasm of  region.  the  to a gray l e v e l of  to o p t i m a l l y  included  some o f  background f i l l  (11)  x  are set to  threshold  chosen,  to  a  several pixels  gradient  blood  m  threshold are set  selected experimentally  cells  to  < 2% P  p i x e l s below t h i s  mask) is  - H(T-l)  the  mask.  window  are  smooth  the  and  the  any o f  dilation its  eight  erosion process, its  eight  a  adjacent  37  Figure  12  SMOOTHING THE NUCLEATED CELL MASK  The p r o c e s s o f shown.  the  operations  The o r i g i n a l  low pass f i l t e r dilation  (bottom  is  nucleated  first  left)  cell  a p p l i e d to and  two  are smoothed, h o l e s are f i l l e d , processes.  used to  smooth t h e mask i s  nucleated c e l l  shown i n  t h e mask ( t o p  erosions  (bottom  the  right), right).  and s m a l l f r a g m e n t s  top  mask left.  f o l l o w e d by  is A a  Jagged  edges  a r e removed by  these  38 4.5  Boundary D e t e c t i o n of  Some o f  the  further  nucleated  analysis.  segmentation,  certain  overlapping  together  that  this  of  point  nucleated  is  13)  one o f  the  complexity  arrangements of  cells,  a cell  that  Hence,  the  from  mask may b e u s e f u l  and the h i g h e r r o r  c e l l s can not  cells  or  cells  would have  for  rate  be a n a l y z e d .  is  then  which  are  too  close  a hard  time  segmenting  encountered.  cell  c h a i n code  each boundary p o i n t eight  a consecutive  mask  A boundary  The  with  the  is  of  these  points  searched u n t i l  chain is  code  defines  starting  a boundary  direction  p o s s i b l e n e i g h b o u r s as the  list  or  analysis.  nucleated  generated.  labels its  the  further  in  These  Hence, only nucleated c e l l s which are standing alone  these  scheme of  extracted  t o u c h i n g were u s e d f o r  To e x t r a c t  Cells  e v e n a human o b s e r v e r  them p r o p e r l y .  boundary  cells  Because of  include  are j u s t  Single  at  numbering  code  (Figure  next boundary the  the  boundary  point. of  the  object.  3  2  1  4  centre  0  5  6  7  Figure  13  BOUNDARY DIRECTION CODES. This point  coding  scheme  relative  to  is  used  the centre  to  label  the  boundary  points  of  the  l o c a t i o n i n the boundary c h a i n code.  next  39  An a l g o r i t h m has been d e v e l o p e d to boundary  information.  calculated find  from  the  the angle of  The l e a s t  The  location  boundary  the  square f i t  tangent  of  the  chain code. line  i s used to  b a s e d on the n i n e p o i n t s  s e p a r a t e t o u c h i n g c e l l s based on boundary  This  information  to each p o i n t  determine  centered at  is  on the o b j e c t  the s l o p e of  the p o i n t  pixels  of  the  interest,  this  can  be  used  to  boundary.  tangent  line  i.e.  8 dx(i)  = S  [x(j+i-4)  - x(i-4)]  (12)  [y(j+i-4)  - y(i-4)]  (13)  j-o 8 dy(i)  -  S  j-o  where d x ( i ) the  x  and  relative  Q /  and d y ( i ) y  to  • \  9(i)  By t a k i n g  directions  the  total  variations  respectively.  the x - a x i s i s  The  then c a l c u l a t e d ,  over  angle  the of  nine this  points line  the d i f f e r e n c e  -  away,  9(i-2)  in  9(i),  i.e.  dv(i) arctan  -  boundary p o i n t s  d9(i)  are  .... (14)  of  angles d 6 ( i ) ,  between p o i n t s which are  four  i.e.  9(i+2)  (15)  40 a n o t i c e a b l e peak i s cells  touch  difference  A line  to  (Figure angle to  A value of  thus  separation of  the  this  separating the  the  line two  touching  touching  cells  to  added o r  to  separate the  touching  t h e b a c k g r o u n d mask l e v e l  A smoothing  operation  is  4.6  represent a  i n t r o d u c e d by  the  (cells)  are  cells.  are  separated,  all  objects  further analysis.  which  This  eliminates  step  regions of  cell.  in  the  cell  is  AND f u n c t i o n  masked image more  operation  segmentation process i s  a nucleated c e l l :  single  logical  has  large  N u c l e u s and C y t o p l a s m Segmentation  The n e x t  each  of  applied  the unnecessary need to analyze o b j e c t s which are too s m a l l or too truly  the  and «.  -n  smooth any c o r n e r s  w i t h i n a p r e s c r i b e d s i z e are used f o r  to  two  s u b t r a c t e d from  is within  the peaks to  are set cells.  separated object the  is  2n  ensure t h a t the r e s u l t  The v a l u e s o f  each of  After  14).  the c o r r e s p o n d i n g l o c a t i o n o f where the  segment i s p r o d u c e d by j o i n i n g  cells. zero,  seen at  is  than  the  generated. two  utilizing  the  green the  spectral  image.  This histogram  is  peaks  15).  edge  e a c h r e g i o n more r e a d i l y a n d t o each r e g i o n .  on  mask w i t h  distinct the  determine  the  the n u c l e u s and the c y t o p l a s m .  overlaid of  to  (Figure  information smooth the  is  different  The mask  image  using  A histogram  generally Hence,  performed  intensity  very an to  of the  of  this  noisy  and  additional help  define  level variations  in  41  T  1  1  1  0  20  40  60  1  1  1  80  100  120  i  1  140  160  »-l  180  Boundary Point Number F i g u r e 14  SEPARATION OF TOUCHING CELLS  The  outline  plots of two  of  of  the d i f f e r e n c e  the s i n g l e peaks  a single  in  cell the  a n d two t o u c h i n g  cells  of the angles of the tangent  a n d t h e two t o u c h i n g c e l l s angle  (top)  difference  plot  f o u n d a t t h e b o u n d a r y o f t h e two t o u c h i n g  The  along the boundary  (bottom)  correspond to cells.  a r e shown.  a r e shown. the  The  indentations  42  0  40  80  120  160  200  Intensity  F i g u r e 15 HISTOGRAM OF THE SEGMENTED NUCLEATED CELLS  The o r i g i n a l cell  mask  image the  image  in  (top l e f t )  (bottom r i g h t )  the green a r e shown.  t o show o n l y  g r e e n image i s g e n e r a t e d  spectrum  (top l e f t )  The mask i s  overlaid  the nucleated c e l l .  (bottom  right).  and the on the  nucleated original  The h i s t o g r a m  of  43 This  additional  filter,  the  operation  sample  gray l e v e l s  mean M ( P ) ,  in a local  If  the  value is  variance is  = J  is  1 S =-1  S i-1  Otherwise,  is is  The h i s t o g r a m  of  the  sample  P(x+i,  If  replaced  the by  variance P(x,y),  In  Var(P), are  this  of  the  calculated.  (16)  y+j)  - M(P(x,y))}  "conditional"  Otherwise,  value  sum o f  the  filter.  y+j)  (P(x+i,  the  mean  each p i x e l  mean v a l u e .  r e p l a c e d by  it  and  a pre-defined  examined and a d j u s t e d . it  conditional  S j - 1  below  r e p l a c e d by the  mean t h e n  the  3x3 w i n d o w a t  1 S i—1  M(P(x,y))  Var(P(x,y))  is  is the  the  greater  (17)  2  limit, value  than  or  of  of  the  pixel  the  pixel  equal  mean a n d s t a n d a r d  difference  the  to  the  deviation.  mean a n d  standard  deviation.  the  resulting  between the n u c l e u s and the  image  cytoplasm  is  smoothed by a 9x1 median f i l t e r  to  the  of  t h e maximum p e a k P  right  H(T)  of  the  first m a x  ,  - H(T+1) < 5% P  dark is  is  used to  (Figure  16).  t o remove n o i s y  peak where  c h o s e n as the  the  determine  the  boundary  This histogram spikes.  gradient  threshold  is  level,  H(T),  The p o i n t less  T,  t h a n 5%  i.e.  (18)  F i g u r e 16  PROCESS TO SEGMENT NUCLEUS AND CYTOPLASM  A filter generated the  i s a p p l i e d t o t h e m a s k e d image (top r i g h t ) .  resulting  A threshold  n u c l e u s mask i s  smoothed (bottom  right).  formed  (top l e f t )  i s determined  and i t s h i s t o g r a m  is  on the h i s t o g r a m and  (bottom l e f t ) .  This  mask i s  then  45 The  gradient  optimal  threshold  f o r generating  greater,  be  c l a s s i f i e d as cytoplasm. stained  cytoplasm  mask.  performed  on  i s experimentally  m a x  the cytoplasm  is  (darker  the l e s s  o f 5% P  dense  parts)  chosen  and n u c l e a r mask.  (lighter  stained parts)  I f the t h r e s h o l d  o f the n u c l e u s  I f the t h r e s h o l d i s lower,  of  the cytoplasm  will  to be the  be  will  the more dense  included  i n the  A median f i l t e r and d i l a t i o n and e r o s i o n o p e r a t i o n s a r e  the r e s u l t i n g  mask  to  fill  the h o l e s  and  smooth  the  boundary o f the n u c l e a r mask.  4.7  Simple Feature  Features known.  Extraction  can be e x t r a c t e d  from  the r e g i o n s  once  their  S e v e r a l f e a t u r e s are implemented t o h e l p v e r i f y  a l g o r i t h m and to e v a l u a t e  the f e a s i b i l i t y  of  features  blood  cells.  These  measurements as w e l l as r a t i o s  include  size,  intensity,  is.  These measurements i n c l u d e the a r e a and p e r i m e t e r  determining  types  and  shape  o f these measurements.  features  and  the segmentation  of c l a s s i f y i n g different  Size  cytoplasm,  give  boundaries are  an i n d i c a t i o n  the n u c l e u s .  the number o f p o i n t s  o f how  The  area  l a r g e each r e g i o n  o f the c e l l , the  measurement  i n the s p e c i f i c  region  or cytoplasm) as d e f i n e d by i t s image mask M(x,y),  o f the c e l l  i s obtained (cell,  i n a 64 p i x e l  by  nucleus by 64  p i x e l matrix, i . e .  Area =  63 63 Z 2 M(x,y) x=0 y=0  (19)  46  where M ( x , y )  is  the c e l l  mask a n d h a s a v a l u e  of  1 i n the object  and 0  elsewhere.  The  perimeter  chain  measurement  code.  The c h a i n  even and c o r n e r these  elements  code,  elements is  is  used  to  compensates  The f o r m u l a  determine  indication  Intensity  of  of circular  areas  of  features  associated with  Chapter  o  d  d  4 . 5 , contains  This  i n each of the three  Imean = "  "  x=0 y=0 63 Z  + 0.980 N  e  to  v  63 S  x=0 y=0  colour  I ( X  e  n  - 0.091 N  c  the cytoplasm  o f the d i f f e r e n t  »  of  regions  t h e amount  spectrums  (red, green,  will since grid.  i s optimized  for  M ( X | Y )  - I ] : Area m e a n  o  r  n  e  r  + 272  (20)  i s used to give i n the  of  an  cell.  stain  that  is  of  intensity  and b l u e ) ,  I  v a r  i.e.  (21)  - IA r e a  [I(x,y)  image  of  The m e a s u r e m e n t s u s e d a r e t h e  and the sample v a r i a n c e  }  odd,  method  shaped o b j e c t s  i s used since i t  an i n d i c a t i o n  m e a n  the  sum o f t h e number  for circularly  the nucleus of the c e l l . I  from  i.e.  the nucleus  give  s a m p l e mean i n t e n s i t y  Q T  value  objects,  of the proportion  I„ . var  in  A weighted  information  f o r e d g e s w h i c h do n o t a l i g n w i t h t h e s q u a r e  P e r i m e t e r = 1.406 N  The r a t i o  the  the perimeter.  d e v e l o p e d b y Young (1988)  the p e r i m e t e r  using  described  o f the edge.  g i v e a more a c c u r a t e p e r i m e t e r it  obtained  2  M(x,y) (22)  47  Shape f e a t u r e s g i v e an i n d i c a t i o n o f how l i k e . One such measurement  „. , .^ Circularity-  the n u c l e u s  o f the c e l l  looks  is circularity, i.e.  o  Perimeter ^ ^  (23)  48  Chapter 5 Discussion and Results  5.1  In  Data  Collection  order to t e s t  cells  from  the b l o o d c e l l  10 s l i d e s  analysis  of blood  smears  algorithm,  were  a p p r o x i m a t e l y 1000  used.  These  slides  o b t a i n e d from the C h i l d r e n ' s H o s p i t a l o f B r i t i s h Columbia and  contained t y p i c a l  preparations.  variations  which  The c l a s s i f i c a t i o n  can be expected  o f these s l i d e s  were  i n Vancouver i n b l o o d smear  was known s i n c e  they  had been p r e p a r e d a t l e a s t one year ago and the h i s t o r y and p r o g r e s s o f each p a t i e n t a r e known t o - d a t e .  To c o l l e c t t h e c e l l s , the  microscope  device.  randomly  field  Each  chosen areas on the s l i d e were brought t o  o f view,  field  was  using  manually  the m o t o r i z e d focussed  c o n t r a s t i n the image as seen on the monitor. acquired  and  nucleated  the system  cells  and  to  was  programmed  perform  the  b o u n d a r i e s were o v e r l a i d on the c e l l for  visual  pixels), on  inspection.  obtain  the  each  pathologist  were of  into  the  f o r later  detected  19 groups  greatest  to  automatically  segmentation.  The  find  the  resulting  images and d i s p l a y e d on the monitor  The s p e c t r a l  stored  o f the  S p e c t r a l images were then  images  (each  64  pixels  the n u c l e u s and cytoplasm masks, and. the l o c a t i o n  the s l i d e  search,  to  x,y stage  cells  observation. was  (Appendix A ) .  manually The c e l l  by 64  o f the c e l l  A t the end o f the classified  by  classification  a  was  49 used  in  conjunction  with  determine  if  a correlation  Slides  3,  4,  1,  leukemia  (ALL)  5,  6,  the  values  the  a n d 8 came f r o m p a t i e n t s  and c o n t a i n  the c l a s s i f i c a t i o n of  calculated  a large  LI  number  All  of  with acute  type w i t h the e x c e p t i o n o f  and L2 s u b - c l a s s i f i c a t i o n t y p e s .  with  acute myelogenous  l e u k e m i a (AML) a n d t h e y  population of  myeloblast c e l l s .  S l i d e s 9 a n d 10 w e r e  different  classes  of  ALL and hence normal  amongst  they  nucleated  cells  s l i d e 5 which has  patients  for  cells  these s l i d e s have lymphoblast  LI  treated  to  lymphoblastic  lymphoblast  both the  had been  features  exists.  the n u c l e a t e d c e l l p o p u l a t i o n . of  of  S l i d e s 2 and 7 were  contain  cells  as  from  contain a  large  from p a t i e n t s  a mixture well  as  of  some  the  who 15  abnormal  cells.  5.2  An  Detection Accuracies  important  correctly other  slide  detect  debris.  those of  aspect  a specific  is  normal from  abnormality  in  detecting  blood  cell  nucleated c e l l s  analyzers in  the  class,  where the  in  cells  (false  manually examined to v e r i f y  their  ability  to  and e l i m i n a t e  all  undetected,  especially  may g e n e r a t e r e s u l t s w h i c h i n d i c a t e fact  analysis  slide.  is  a given f i e l d  Any n u c l e a t e d c e l l s w h i c h a r e l e f t  eliminated  not  all  of  it may  Although  is  not.  Debris  produce  results  this  negatives), its  is  not  these  normality.  which that  as s e r i o u s slides  will  that  the  are  not  indicate  an e r r o r  as  have  be  to  50 All  nucleated  Most d e b r i s easily  there  red blood  were  have equal  eliminated  Although  small  cells  by  c o r r e c t l y detected  i n t e n s i t y i n a l l colour  the s u b t r a c t i o n  a r e many r e g i o n s  cells,  i n the s e l e c t e d  platelets,  spectra  o f the b l u e  clumps  of cells)  were  object.  also  Objects  eliminated  the b o r d e r s  the r e d .  to parts  and the s i z e  by  the s i z e  touching  data  t h e r e was not enough i n f o r m a t i o n  of  very  criteria  which a r e too l a r g e  which were since  image from  d e b r i s , these were g e n e r a l l y  and were e l i m i n a t e d by the e r o s i o n p r o c e s s  imposed on each i s o l a t e d  and hence were  i n the image which b e l o n g  and o t h e r  frames.  (such  criterion.  as  Cells  o f the image were not i n c l u d e d  i n the  to c l a s s i f y a f r a c t i o n o f a  cell.  Of and Of  the 1078 d e t e c t e d 297 were these  cells,  had  experienced misplaced  two o r t h r e e  touching  18  nucleated  cells,  minor  major  cells 271  errors  errors  cells,  t h a t were j u s t  were  where  touching  correctly divided  i n the p o s i t i o n the  of  location  smoothing o p e r a t i o n  each  into  of  the  cells  another.  individual  the boundary,  and  boundary  8  was  c e l l s i s l a r g e l y due t o the boundary  a p p l i e d to the n u c l e a t e d  i s performed  to c o l l e c t  c e l l mask.  the l o c a t i o n where two c e l l s touch i s b l u r r e d . separating algorithm actual  boundary  This  smoothing  and merge any s c a t t e r e d p i e c e s  b e l o n g s to the c y t o p l a s m a f t e r the t h r e s h o l d i n g o p e r a t i o n .  the  single  ( F i g u r e 17).  Minor e r r o r s i n s e p a r a t i n g t o u c h i n g  operation  781 were i n d i v i d u a l ,  which  As a r e s u l t ,  Hence, the t o u c h i n g  cell  can choose p o i n t s which are upto 3 p i x e l s away from position.  This  error  can  be  corrected  by  51  F i g u r e 17  MINOR AND MAJOR ERRORS IN SEPARATING TOUCHING CELLS  The minor is  (top) and major  shown.  The s e p a r a t i n g  (bottom) e r r o r s line  i n separating  produced by the a l g o r i t h m  b l a c k and the a c t u a l boundary l o c a t i o n i s shown i n w h i t e . slightly  s h i f t e d i n the case o f the minor e r r o r s .  those which a r e n o t s e p a r a t e d  touching  or i n c o r r e c t l y  i s shown i n This  Major e r r o r s  separated.  cells  line i s includes  52 introducing coarsely  an a l g o r i t h m  chosen  image and o t h e r  Major errors which  are  cell  separation  criteria,  a r e due t o near  will  either  fragments  which  do n o t  these  to  errors  or  in  defining  segmentation defined Tables where the  is  I  to  small  is  the  certain  The m a j o r  the  cell  cell  other  intensities  of  the  for  by  of  the  rest  of  the  the  pieces  cell. the  the  cell  the  not  large  s m o o t h i n g p r o c e s s w h i c h smooths cells.  out  is  those  region. of  the  the in  nucleus  concavities  in  cell Minor  cytoplasm  of  minor  of  errors  another  in  sharp  the  included  included  number  of  tabulated  not  cytoplasmic  of  features  b a s e d on  are  are  where p a r t s  types of white blood  is  analyzed are 18)  mask.  correctness  analysis  are those e r r o r s  the  include  images  The  (Figure  are  The  or  the  cell  19)  nucleus  cells  the n u c l e a t e d c e l l  blood  defined  or  cytoplasm.  a r e a s , such as c y t o p l a s m o f  in  reason for  cells  analyzing  cells  errors  point.  of  into  into  cytoplasm of  included  parts  of  cytoplasm  of  (Figure or  the  the  elliptic)  pieces  analyzing  regions  results  t o o much o f are  the  in  since  Minor  fragments  nucleus errors  region.  The  V.  region or  included  the  crucial  regions.  and background,  are  criterion  of  the  shaped (not  c a n be c o r r e c t e d  Segmentation Accuracies  neighbourhood  the exact d i v i d i n g  single  belong to  5.3  accuracy  the  the  using  fragmented  split  objects.  important  point,  irregularly  these  Another  search in  locate  platelets  algorithm  Some o f  which w i l l  the  nuclear errors  present  in  53  Major  cytoplasm  shaped  errors  (Figure  o r when the c y t o p l a s m  very s i m i l a r the c e l l uneven  when  cells  are  o f some n u c l e a t e d c e l l s  irregularly  possess  a colour  t o t h a t o f the r e d b l o o d c e l l s and thus i s e l i m i n a t e d from  mask. stain  incorporates  20) a r i s e  Major n u c l e u s e r r o r s uptake  in  too much  region.  Some  nuclei  algorithm  assigns  ( F i g u r e 21) a r e c o n t r i b u t e d to the  the c e l l s .  stain have  The  resulting very  the l i g h t e r  dark  parts  in a  cytoplasm larger  stained  of  some  segmented  regions  o f the n u c l e u s  and  cells  nuclear thus  the  t o the cytoplasm  area.  The  a c c u r a c y o f the segmentation  focussed. chosen  The f o c u s s i n g  microscope  intensity  levels  enough and may cytoplasm background  range  setup.  a l s o depends on how w e l l the c e l l s a r e i s i n the o r d e r o f f i v e  As the focus  a t the n u c l e u s result  i n errors  i s changed,  and cytoplasm i n defining  o f the n u c l e a t e d c e l l s  can a l s o  c a u s i n g c y t o p l a s m segmentation  microns  the t r a n s i t i o n o f  boundary  i s n o t abrupt  the n u c l e u s  blend  a t the  into  region.  the r e s t  The  o f the  errors.  As shown i n the T a b l e s I t o V, the e r r o r r a t e s i n segmentation v a r y from slide  to s l i d e .  T h i s i s due t o the type o f c e l l s  as the way the s l i d e was p r e p a r e d . mixture are  o f the c e l l  present  because  types  In s l i d e s  on the s l i d e  9 and 10 which c o n t a i n a  i n the b l o o d , more cytoplasm  there  is a  as w e l l  greater p r o b a b i l i t y  related  of finding  errors cells  which have a cytoplasm c o l o u r s i m i l a r to the r e d b l o o d c e l l s compared t o the o t h e r e i g h t  slides.  The percentage  of correct  segmentation  ranges  54  F i g u r e 18 MINOR CYTOPLASM ERRORS  Examples include  o f minor small  cytoplasm  fragments  errors  are shown.  o f the background  The a l g o r i t h m s  or other  cytoplasm or e x c l u d e p a r t s o f i t s own cytoplasm.  cells  either  into  the  55  Examples  o f minor  nucleus  p a r t s o f the c y o t p l a s m  errors  are  i n t o the nucleus  shown. region.  The  algorithms  include  Examples o f major cytoplasm e r r o r s are shown.  The a l g o r i t h m s e x c l u d e a  major p o r t i o n the cytoplasm o f the c e l l from the c y t o p l a s m  region.  57  F i g u r e 21 MAJOR NUCLEUS ERRORS  Examples o f major n u c l e u s  errors  major p o r t i o n o f the n u c l e u s .  a r e shown.  The a l g o r i t h m s  exclude  a  58  Table  I  SEGMENTATION ERRORS I N NON-TOUCHING NUCLEATED C E L L S .  Slide Number  Correct Seg.  Minor Nucleus  Minor Cyto.  Major Nucleus  Major Cyto.  Major N & C  Total  1 2 3 4 5 6 7 8 9 10  82 93 63 72 68 67 58 59 56 58  7 10 2 6 2 2 7 2 1 0  2 0 2 0 3 1 1 3 1 4  5 0 0 0 11 0 8 2 2 2  2 0 2 0 0 0 0 0 5 7  2 0 0 0 0 0 0 0 1 0  100 103 69 78 84 70 74 66 66 71  otal  676  39  17  30  16  3  781  Table  II  PERCENTAGE ERRORS I N NON-TOUCHING NUCLEATED C E L L S . (Table  I  r e p r e s e n t e d as  percentages)  Slide Number  Correct Seg.  Minor Nucleus  Minor Cyto.  Correct & Minor  Maj o r Nucleus  Major Cyto.  Maj o r N & C  1 2 3 4 5 6 7 8 9 10  82.0 90.3 91.3 92.3 81.0 95.7 78.4 89.4 84.8 81.7  7.0 9.7 2.9 7.7 2.4 2.8 9.4 3.0 1.5 0  2.0 0 2.9 0 3.6 1.4 1.4 4.6 1.5 5.6  91.0 100.0 97.1 100.0 87.0 100.0 89.2 97.0 87.8 87.3  5.0 0 0 0 13.1 0 10.8 3.0 3.0 2.8  2.0 0 2.9 0 0 0 0 0 7.6 10.0  2.0 0 0 0 0 0 0 0 1.5 0  Total  86.4  5.0  2.2  93.6  3.8  2.0  0.4  59 Table SEGMENTATION  III  ERRORS I N TOUCHING NUCLEATED  CELLS.  Slide Number  Correct Seg.  Minor Nucleus  Minor Cyto.  Major Nucleus  Major Cyto.  Maj o r N & C  Total  1 2 3 4 5 6 7 8 9 10  8 3 35 23 24 29 30 35 31 29  0 0 0 0 1 2 5 8 0 0  1 0 0 2 0 2 0 3 5 5  1 0 1 1 3 0 0 1 1 0  0 0 3 0 0 0 0 0 2 3  0 0 0 0 0 0 0 0 0 0  10 3 39 26 28 33 35 47 39 37  Total  247  16  18  8  8  0  297  T a b l e: IV PERCENTAGE ERRORS (Table  III  I N TOUCHING NUCLEATED  represented  CELLS.  as p e r c e n t a g e s )  Slide Number  Correct Seg.  Minor Nucleus  Minor Cyto.  Correct & Minor  Major Nucleus  Maj o r Cyto.  Major N 6c C  1 2 3 4 5 6 7 8 9 10  80.0 100.0 89.7 88.5 85.7 87.9 85.7 74.5 79.5 78.4  0 0 0 0 3.6 6.1 14.3 17.0 0 0  10.0 0 0 7.7 0 6.1 0 6.4 12.8 13.5  90.0 100.0 89.7 96.2 89.3 100.0 100..0 87.9 92.3 91.9  10.0 0 2.6 3.8 10.7 0 0 2.1 2.6 0  0 0 7.7 0 0 0 0 0 5.1 8.1  0 0 0 0 0 0 0 0 0 0  Total  83.2  5.4  6.1  94.6  2.7  2.7  0  60 Table V PERCENTAGE SEGMENTATION ERRORS IN NUCLEATED CELLS. (Tables  I and I I I combined and r e p r e s e n t e d as p e r c e n t a g e s )  Slide Number  Correct Seg.  Minor Nucleus  Minor Cyto.  Correct & Minor  Major Nucleus  Major Cyto.  Maj or N & C  1 2 3 4 5 6 7 8 9 10  81.8 90.6 90.7 91.3 82.1 93.2 80.7 83.2 82.9 80.6  6.4 9.4 1.9 5.8 2.7 3.9 11.0 8.8 0.9 0  2.7 0 1.9 1.9 2.7 2.9 1.0 5.3 5.7 8.3  90.9 100.0 94.5 99.0 87.5 100.0 92.7 97.3 89.5 88.9  5.5 0 0.9 1.0 12.5 0 7.3 2.7 2.9 1.9  1.8 0 4.6 0 0 0 0 0 6.4 9.2  1.8 0 0 0 0 0 0 0 0.9 0  Total  85.4  5.1  3.2  94.0  3.5  2.2  0.3  61 from 80.6% nucleus major  to 93.2%.  i s 0%  errors  cytoplasm  5.4  are  process.  and  11.0%, minor  i n the  i s 0%  range o f percentages  to  nucleus  e r r o r i n the  i s 0%  to  12.5%,  calculated  from  the  errors introduced errors  circle  satisfies  i n the  errors,  a  image i s generated the e q u a t i o n  i s 0%  major  to  errors  8.3%, in  the  d e f i n e d by  the  segmentation  segmentation,  there  image to a s q u a r e - p i x e l  method used  to  test  consisting  by  the  image  calculate  a  of  will grid,  feature.  To  a  is  circle  the a r e a f e a t u r e i s c a l c u l a t e d .  a s s i g n i n g a l l p o i n t s i n the g r i d which  to a v a l u e  o f 1 and  a l l other p i x e l s  to a  value  zero, i . e .  (x - x )  2  c  where x  c  and  2 + (y - y )  < square  c  y  c  are  the  area  image.  feature  radius of c i r c l e  coordinates of  and y are the l o c a t i o n i n the  The  and  e r r o r s i n the  in tessellating  introduced these  regions  t e s s e l l a t e d a t d i f f e r e n t r e s o l u t i o n s and  of  cytoplasm  1.8%.  Even i f t h e r e were no  illustrate  The  f o r minor e r r o r s i n the  Feature C a l c u l a t i o n Accuracies  Features  be  to  The  is calculated  circle  is  o f the  circle  and  x  grid.  by  counting  S i n c e the p o s i t i o n o f a c e l l the  the c e n t r e  (24)  position  of  allowed  spacing.  A p l o t o f the percentage  the  number  of  can  l i e anywhere i n the  to  randomly  vary  l's in  the  image,  the  within  e r r o r from the a c t u a l v a l u e  a  pixel  i s shown  10  100  Diameter of Circle, D (pixels) F i g u r e 22 FEATURE CALCULATION ACCURACIES  The  percentage  different seen that circle.  number  errors  in  of p i x e l s  the e r r o r  calculating to represent  the  area  the c i r c l e  d e c r e a s e s a s more p i x e l s  of  a  circle  i s shown.  are used to  It  using c a n be  represent  the  63 in Figure  22.  It  c a n be s e e n t h a t  more t h a n  ten p i x e l s  in width,  if  the  an e r r o r  diameter  of  of  the  circle  spans  l e s s t h a n two p e r c e n t  in  the  long  as  a r e a c a l c u l a t i o n c a n be o b t a i n e d .  Although  there  are  their  distribution  other  classes,  errors for  in  the  a certain  the e r r o r w i l l  calculation class  of  of  cell  features, does n o t  n o t be s i g n i f i c a n t  as  conflict  with  i n c l a s s i f y i n g the  cell  type.  5.5  Cell  Classification  The n u c l e a t e d c e l l s , w h i c h w e r e d e t e c t e d a n d s e g m e n t e d , w e r e by  an  experienced  features  were  of  certain  based  green  using  lymphoblast nucleus It  is  that  evident clusters  for  approximately  (Figure  ratio  of  of  segment  c l a s s i f i c a t i o n of  and  23).  of  nucleus the  the types  the  are  regions are  some t y p e s  of  to  cell  of  and the  sufficient  in  cells,  nucleated c e l l .  the  This  can  be  the  red  and  such  as  the  of  the  (Figure  (Figures  23 a n d  verifies  calculations of in  . as  perimeter  area features  present.  are  such  cytoplasm  determining  the  which  normoblast,  calculated features  classes  classes  of  Simple  determine  types  information,  basophilic  Other  groups.  Some c e l l  c a n be s e p a r a t e d u s i n g the  the  certain  these  colour  20  and are u s e d to  cells.  mean i n t e n s i t i e s  from p l o t s  methods u s e d t o values  their  and m y e l o b l a s t ,  and the  of  normoblast  the  spectrums  types  on  polychromatophilic separated  into  c a l c u l a t e d from these c e l l s  classification identified  pathologist  classified  the  that  24). 24) the  feature  the  correct  A l t h o u g h more  features  64 250  Polychromatophilic Normoblast  200"  E *i—-»  Basophilic Normoblast  #150CQ  >, 100"  c a 50--  0+0  —t-  1  50  100  200  150  250  Mean Intensity in the Green Spectrum  F i g u r e 23 CLUSTER PLOT OF THE MEAN I N T E N S I T I E S  A  cluster  plot  spectrums  of  displaying the  the mean  images  of  intensities two  i n the green  classes  of  p o l y c h r o m a t o p h i l i c normoblast and b a s o p h i l i c normoblast.  blood  and b l u e cells:  65  100-  Lymphoblast  75-  <u  Myeloblast  U o-> •* 3 503  z o _o  J 25-  50  100  200  150  Perimeter of the Nucleus  F i g u r e 24 CLUSTER PLOT OF THE PERIMETER AND RATIO OF AREAS  A cluster  plot  the n u c l e u s cells:  to  displaying cell  lymphoblast  area  the perimeter  o f the nucleus  of  two c l a s s e s  images  and m y e l o b l a s t .  of  of  and the r a t i o malignant  of  blood  66 and  multi-dimensional  separate additional regions.  the  different  algorithms  cluster classes to  fine  analysis of tune  algorithms  cells, the  i t is  boundaries  are  required  unnecessary of  the  to  to add  segmented  67  Chapter 6 Conclusion and Future Suggestions  6.1  Overview  The  algorithms  segment  to i ) capture  the c e l l s  segmented  regions,  and c a l i b r a t e  i n the image,  the image,  i i i ) generate  and i v ) c l a s s i f y  the c e l l s  i i ) d e t e c t and  features  based  from the  on the f e a t u r e  v a l u e s , have been shown t o be u s e f u l i n the a n a l y s i s o f n u c l e a t e d cells.  Techniques  first  employed  method  t o improve  of subtracting  generating  bi-modal  smoothing  and  employed The  o f image  level  threshold  detection  filter more  from  t h e chosen  has produced  readily  techniques  cells  touching  areas  from  found t o a d e q u a t e l y  fill  boundaries.  Once,  the r e g i o n s  calculated.  These  feature  erosion  the h o l e s  along  The new useful i n  image where  subsequentially  the r e s t  o f the image. the boundary o f  and cytoplasm  i n the r e g i o n s  a r e then  to be  The c o n d i t i o n a l mean  and d i l a t i o n  are defined,  values  t o be v e r y  are  o f the s l i d e .  Finally,  image.  c e l l s has a l l o w e d more c e l l s  images where the nucleus  defined.  s u b t r a c t i o n are  o f the r e s u l t i n g  the d i f f e r e n c e i n angles  the b i n a r y image t o s e p a r a t e analyzed  i s shown  histogram  the n u c l e a t e d  new method o f u s i n g  and background  the q u a l i t y o f the i n p u t  s p e c t r a l images  grey  to separate  averaging  blood  boundary i s  operations  were  and smooth the r e g i o n  various compared  features  c a n be  t o determine i f  t h e r e a r e any groupings amongst the d i f f e r e n t c l a s s e s o f c e l l s .  68 It  is  evident  from  the  a n a l y s i s i s important have d i f f i c u l t y Analyzer  of  cytometry  i n medical p r a c t i c e  in classifying  Imaging  algorithms  survey  System was  discussed  in  abnormal b l o o d  used  this  as  algorithm w i l l  different  algorithms  work on may  the  to  The  survey  of  The  systems  the  segmentation  t h a t no  algorithms  Cell  test  particular  However, combining  the  cell  the  and  which  those which are t a i l o r e d to a p a r t i c u l a r type o f images one  6.2  blood  Hence,  develop  confirmed  a l l images. well;  cells.  tool  thesis.  perform  that  and t h a t a l l o f these  a l g o r i t h m s employed to segment b l o o d c e l l s single  systems  several work  are  obtains.  System Performance  algorithms  smear.  perform  Of the 1078  w e l l i n a n a l y z i n g the  cells  p r o p e r l y s e p a r a t e d due  types  chosen, 3% o f the 297  to the odd c e l l shape.  6% o f these t o u c h i n g c e l l s  were  incorrectly  Although  i s s l i g h t l y misplaced,  segmented  and  7%  extracted  indicates  that  different  blood  the boundary f o r  these c e l l s were 6% o f the 781  have s l i g h t  t h e r e are minor e r r o r s i n the segmentation,  in a  t o u c h i n g c e l l s were not  p r o p e r l y c l a s s i f i e d based on t h e i r f e a t u r e v a l u e s . cells  of c e l l s  errors.  still single  Although  the d i s t r i b u t i o n o f f e a t u r e s  classes  of  blood  cells  can  be  distinguished.  6.3  The  Future  Plans  a l g o r i t h m s developed  types  of  different  i n t h i s t h e s i s permit c l a s s i f i c a t i o n of s e v e r a l  white  blood  cells.  To  d i s c r i m i n a t e between  more  69 classes  and  segmented  subclasses  areas,  algorithm. at  of  would  blood  have  to  cells, be  F e a t u r e s s u c h as the  more  developed  ratios  based  on  the  incorporated  in  the  of  intensity  may b e u s e d .  w h i c h g i v e a n i n d i c a t i o n o f how i n t e n s i t y to  and  mean a n d v a r i a n c e  d i f f e r e n t w a v e l e n g t h s and t h e i r  a r e a l s o good f e a t u r e s  features,  Texture  l e v e l s are v a r y i n g  a criterion  the  must be d e v e l o p e d t o  linear  the  cells  stepwise  feature  to  different  classes  features,  which  be  not  of  to  interpret  slide  (Poon,  function the  cells.  focus,  the  which  its  effects  contrast  an i m p o r t a n t f e a t u r e 1989).  An  autofocussing  l e v e l of  determined  criteria,  whereas  vary  image  image)  involved. gradients  A in  to  the  image  a  system  such  as  be  sum  c a n be u s e d t o  give  A the  separate  the  removal  of  those  analysis  should errors. are  values.  are  dependent autofocus  produce  an  on  the  (Poon et  objective  and  t h e n e x t b a s e d o n some p r e level  chosen  the  it  on  and d i s c a r d c e l l s t h a t  will  subjective  would  1987).  classification  images  f o c u s f r o m one i m a g e t o  criterion, the  of  best  from the  on the  cell,  v a l u e s and use  the  t h a t must be d e v e l o p e d i s  consistent  from  in  expensive,  s h o u l d a l s o be d e v e l o p e d t o d e t e c t  and  would  Experimentation  determine  each  a n a l y s i s s h o u l d be p e r f o r m  features  computationally  sharpness  feature  J a g g i and P a l c i c ,  segmented p r o p e r l y b a s e d on t h e s e f e a t u r e  Since  al.,  select  are  investigated  Algorithms  on the  discriminant  values  region  incorporate.  are c a l c u l a t e d for  classify  features  in a  Once a c o l l e c t i o n o f more t h a n 20 f e a t u r e s  to  levels  if of  of  focus  human the  (which  might  intervention  absolute  an i n d i c a t i o n  of  is  intensity the  focus  70 level.  A high  variations of  b a s e d on  instruct  detection  Using  slide.  (details)  Thus,  the  feature  stepping  additional  system The  slide. the  system  developed include  An  automatic  slide  on the  slide  motorized  being  abnormal  allow  little  hospitals  human  Cell  Analyzer  cells,  under  author  of  the  this  1986;  focus  fully  to  classify  a  motorized  to  focus  before  automated the  as  levels, the  cell  cells  cell on  the the  x,y  stage  to  scan  automatically  focus  the  cells  on  incorporated  to  should  also  stage  and to  place  a  focus  blood  blood  loader  in  optimum  initiated.  a  a different  large well  be  those  place  slides  for  number as  of  slides  generating  which  future  manual  This  system  quickly  a  with  standard  in  Contributions  Imaging  System  supervision  of  has  Branko  t h e s i s has p l a y e d a major  including  Palcic,  are  the  different  the  large  slides.  Summary o f A u t h o r ' s  system,  adjust  is  characteristic  the machine c l a s s i f i c a t i o n .  scan  interaction,  c l a s s i f y i n g abnormal  6.4  to  of  at  there  is  determine  obtained to  that  image w h i c h  system w i l l  motors  should  indicate  the  algorithms,  c a n be  suspected of  would  and  in  values  o b s e r v a t i o n and v e r i f i c a t i o n  The  sum w i l l  and a mechanism and a l g o r i t h m  place are  the  and s e g m e n t a t i o n a l g o r i t h m s  analyzing  the  the  the  these  slide  for  intensity  a f o c u s s e d image.  level and  in  value  hardware, Poon,  device  Jaggi  and  been  built  Palcic role  drivers, Palcic,  to  and  i n the  analyze  Bruno  Jaggi.  development  and  software  1987;  Palcic  et  stained The  of  this  (Jaggi,  Poon  al.,  1988;  J a g g i e t a l . , 1988; Poon e t a l . , 1989; P o u l i n e t a l . , 1989; Poon and P a l c i c ,  1989a;  and Spadinger,  development o f the system was d e s i g n e d presented  Using  i n this  this  1989b).  The  t o a l l o w work t o be performed as  thesis.  system,  segmentation  Poon and P a l c i c ,  Spadinger,  a  procedure  containing  algorithms  o f n u c l e a t e d b l o o d c e l l s was developed  for  automatic  by the author.  This  i n c l u d e d the methods o f image a c q u i s i t i o n , c o r r e c t i o n s , and s u b t r a c t i o n , and b i n a r y mask p r o c e s s i n g which were m o d i f i e d and adapted i n t o the c e l l analysis  procedure.  New  separate  touching  intensity  variations  cells  methods,  and  while  using  the  difference  the c o n d i t i o n a l mean  preserving  edges,  were  angles  filter  to  introduced  to  reduce  and a l s o  i n c o r p o r a t e d i n t o the procedure.  A t e s t data s e t o f over 1000 c e l l s was  used t o e v a l u a t e the segmentation  algorithms.  The  main  contribution  introducing algorithms field  several into  new  t o new  knowledge  algorithms,  i n this  i s t o combine  field,  the hardware  a working system which a l l o w s s c i e n t i s t s  r e l a t e d t o leukemia  apart  from and  i n the m e d i c a l  d i s e a s e s t o study t h i s d i s e a s e i n much g r e a t e r  d e t a i l than was p o s s i b l e b e f o r e .  72  Chapter 7 Bibliography Abraayr W . , M a n n w e i l e r E . , O e s t e r l e D . , a n d Demi E . ( 1 9 8 7 ) of scenes in tissue sections. Analytical and C y t o l o g y and H i s t o l o g y , 9:190-196  Segmentation Quantitative  Amadasun M. , and K i n g R.A. (1988) Low-level segmentation of multispectral images v i a a g g l o m e r a t i v e clustering of uniform neighbourhoods. Pattern Recognition, 21(4):261-268. A r c e l l i C . , C o r d e l i a L . P . , and L e v i a l d i S . (1981) F r o m l o c a l maxima t o connected skeletons. P a t t e r n A n a l y s i s and Machine I n t e l l i g e n c e , 3(2):134-143 A r c e l l i C , and D i B a j i G . S . (1978) medial line transformation. 8(2):139-144 Aus  On t h e Systems,  sequential Man, and  approach to Cybernetics,  H . M . , Harms H . , H a u c k e M. , B e r i t o v a J . , T e r M e u l e n V . , G u n z e r U . , Baumann I., a n d A b m a y r W. ( 1 9 8 6 ) L e u k e m i a - r e l a t e d m o r p h o l o g i c a l features i n blast c e l l s . Cytometry, 7:365-370  A u s H . M . , R u t e r A . , T e r M e u l e n V . , G u n z e r U . , a n d N u r n b e r g e r R. ( 1 9 7 7 ) Bone m a r r o w cell scene segmentation by computer-aided color cytophotometry. The J o u r n a l o f H i s t o c h e m i s t r y a n d C y t o c h e m i s t r y , 25(7):662-667 Bacus  J . W . and Grace L . J . (1987) s t a n d a r d i z e d c e l l measurements 26(16):3280-3293  Optical microscope system for and a n a l y s e s . Applied Optics,  Bacus  J . W . , Gose E . E . (1972) Leukocyte Man, a n d C y b e r n e t i c s , 2 ( 4 ) : 5 1 3 - 5 3 6  Pattern  Begemann H . , a n d R a s t e t t e r J . ( 1 9 7 9 ) Atlas Third Edition, Springer-Verlay, Berlin.  of  Reognition.  Clinical  Systems,  Hematology.  B e n g t s s o n E. , E r i k s s o n 0 . , H o l m q u i s t J . , J a r k r a n s T. , N o r d i n B. , and S t e n k v i s t B. (1981) Segmentation of c e r v i c a l c e l l s : d e t e c t i o n of overlapping c e l l nuclei. C o m p u t e r G r a p h i c s a n d Image P r o c e s s i n g , 16:382-394 Bengtsson E. , E r i k s s o n 0 . , H o l m q u i s t J . , N o r d i n B. , and S t e n k v i s t B. (1979) High r e s o l u t i o n segmentation of c e r v i c a l c e l l s . The J o u r n a l of H i s t o c h e m i s t r y and Cytochemistry, 27(1):621-628 Bengtsson E . , E r i k s s o n 0 . , Holmquist J . , and S t e n k v i s t B. (1979) Implementation and E v a l u a t i o n o f a Diode A r r a y Scanner for D i g i t i z i n g M i c r o s c o p i c Images. The A u t o m a t i o n o f C a n c e r C y t o l o g y  and C e l l 289  Image  Analysis.  Eds. Pressmann N . J .  and Wied S . L . ,  269-  B e n n e t t J . M . , C a t o v s k y D. , D a n i e l M . T . , F l a n d r i n G . , G a l t o n D . A . G . , Gralnick H.R., and Sultan C. (1976) Proposal for the C l a s s i f i c a t i o n of the Acute Leukemias: French-American-British (FAB) c o o p e r a t i v e g r o u p . B r i t i s h J o u r n a l of Hematology, 33:451458 Brenner J . F . , Necheles T . F . , Bonacossa I.A., F r i s t e n s k y R. , W e i n t r a u b B . A . , a n d N e u r a t h P.W. ( 1 9 7 7 ) Scene s e g m e n t a t i o n t e c h n i q u e s for the analysis of routine bone marrow smears from acute lymphoblastic leukemia p a t i e n t s . The J o u r n a l o f Histochemistry and C y t o c h e m i s t r y , 25(7):601-613 Castleman K.R. (1987) Spatial and calibration requirements for cell A p p l i e d O p t i c s , 26(10):3338-3342 Castleman K.R. York  (1979)  Digital  Image  photometric resolution and image a n a l y s i s instruments.  Processing.  Prentice  Cheng G . C . (1974) Color information in blood c e l l s . H i s t o c h e m i s t r y and C y t o c h e m i s t r y , 22(7):517-521  The  Hall,  Journal  New  of  C l a r k M, B o v i k A . C . , a n d G e i s l e r W . S . ( 1 9 8 7 ) Texture segmentaion using Gabor m o d u l a t i o n /demodulation. Pattern Recognition Letters, 6:261-267 Davis  Dill  L . S . (1975) A s u r v e y o f edge d e t e c t i o n G r a p h i c s a n d Image P r o c e s s i n g , 4 : 2 4 8 - 2 7 0  techniques.  Computer  A . R . , L e v i n e M . D . , and N o b l e P . B . (1987) Multiple skeletons. P a t t e r n A n a l y s i s and Machine I n t e l l i g e n c e , 503  resolution 9(4):495-  D y t c h H . E . , B i b b o M. , B a r t e l s P . H . , P u i s J . H . , a n d W i e d G . L . ( 1 9 8 7 ) A PC-based system f o r the o b j e c t i v e a n a l y s i s of h i s t o l o g i c specimens through quantitative contextual karyometry. Applied Optics, 26(16):3270-3279 Fu K . S . , and M u i J . K . (1981) Recognition, 13:3-16 Fulwyler M.F. Volume, Garbay  A survey of  (1965) Electronic S c i e n c e , 150:910  image  Separation  of  segmentation.  Biological  Pattern  Cells  by  C , C h a s s e r y J . M . , and B r u g a l G. (1986) An i t e r a t i v e regiong r o w i n g p r o c e s s f o r c e l l image s e g m e n t a t i o n b a s e d o n l o c a l c o l o r s i m i l a r i t y and g l o b a l shape c r i t e r i a . A n a l y t i c a l and Q u a n t i t a t i v e C y t o l o g y and H i s t o l o g y , 8:25-34  Graham  74 M.D., and Norgren P.E. (1980) The Diff3 Analyzer: A P a r a l l e l / S e r i a l G o l a y Image P r o c e s s o r , R e a l - T i m e M e d i c a l Image P r o c e s s i n g . Onoe M . , P r e s t o n K. J r . , a n d R o s e n f e l d A . E d s . P l e n u m , New Y o r k .  Hadon J . F . (1988) G e n e r a l i s e d t h r e s h o l d Pattern Recognition, 21(3):195-203  selection  for  H a l l T . L . , K e y a n i D . H . L . , and R o s e n t h a l D . L . (1987) image p r o c e s s i n g w o r k s t a t i o n s for cytology. 26(16):3266-3269  edge  detection.  Microcomputer-based Applied Optics,  H a r a l i c k R . M . , S t e r n b e r g S . R . , and Zhuang X . (1987) Image a n a l y s i s using mathematical morphology. P a t t e r n A n a l y s i s and Machine Intelligence, 9(4):532-550 Harms  H . , A u s H . M . , H a u c k e M. , a n d G u n z e r U . ( 1 9 8 6 ) Segmentation of s t a i n e d b l o o d c e l l images measured a t h i g h s c a n n i n g d e n s i t y w i t h h i g h m a g n i f i c a t i o n and h i g h n u m e r i c a l a p e r t u r e o p t i c s . Cytometry, 7:522-531  Haussmann G. and L i e d t k e C . E . (1984) A r e g i o n e x t r a c t i o n approach to b l o o d smear s e g m e n t a t i o n . C o m p u t e r V i s i o n , G r a p h i c s , a n d Image P r o c e s s i n g , 25:133-150 Ingram M . , and P r e s t o n K . J r . (1970) S c i e n t i f i c America, 223:72-82 Inoue S.  (1986)  Video Microscopy.  Automatic  A n a l y s i s of  P l e n u m , New Y o r k ,  Blood  Cells.  1986.  J a g g i B . , Poon S . S . S . , M a c A u l a y C , and P a l c i c B. (1988) Imaging System f o r Morphometric Assessment o f C o n v e n t i o n a l l y and F l u o r e s c e n t l y Stained Cells. Cytometry, 9:6 J a g g i B . , and P a l c i c B. (1985) The D e s i g n a n d D e v e l o p m e n t o f a n O p t i c a l Scanner for Cell Biology. IEEE P r o c e e d i n g s , E n g i n e e r i n g in M e d i c i n e and B i o l o g y , 2:980-985 Jaggi  B. , Poon S . S . S . , and P a l c i c B. (1986) Implementation and Evaluation of the Dmips Cell Analyzer. IEEE Proceedings, E n g i n e e r i n g i n M e d i c i n e and B i o l o g y , 3:906-911  K a s d a n H . L . , L a n f o r d K . C . , L i b e r t y J . , Z a c h a r i a s h M. , a n d D i e n d o e r f e r F.H. (1987) High performance pathology w o r k s t a t i o n using an automated m u l t i s p e c t r a l microscope. Applied O p t i c s , 26(16):32943300 L a c r o i x V . (1988) A t h r e e - m o d e l s t r a t e g y f o r edge d e t e c t i o n . A n a l y s i s and Machine I n t e l l i g e n c e , 10(6):803-810  Pattern  L a n d e w e e r d G . H . , Timmers T . , and Gelsema E . S . (1983) C l a s s i f i c a t i o n of normal and abnormal samples o f p e r i p h e r a l b l o o d by l i n e a r mapping of the f e a t u r e space. Pattern Recognition, 16(3):319-326  75 Landeweerd G . H . , Gelsema E . S . , B r e n n e r J . F . , S e l l e s W . D . , and Z a h n i s e r D.J. (1983) P a t t e r n r e c o g n i t i o n of n u c l e a t e d c e l l s from the peripheral blood. Pattern Recognition, 16(2):131-140 Lee J . S . (1983) D i g i t a l Image S m o o t h i n g a n d t h e S i g m a F i l t e r . V i s i o n , G r a p h i c s , a n d Image P r o c e s s i n g , 2 4 : 2 5 5 - 2 6 9  Computer  Lemkin  P. , and L i p k i n L. (1979) Use o f the positive difference transform for RBC e l i m i n a t i o n in bone marrow smear images. A n a l y t i c a l and Q u a n t i t a t i v e C y t o l o g y , 1:67-76  Lesty  C . , R a p h a e l M. , N o n n e n m a c h e r L . , L e b l o n d - M i s s e n a r d V . , Delcourt A., Homond A . , and Binet J.L. (1986) An a p p l i c a t i o n of mathematical morphology to a n a l y s i s of the s i z e and shape of n u c l e i i n t i s s u e s e c t i o n s o f n o n - H o d g k i n ' s lymphoma. Cytometry, 7:117-131  L i e d t k e C . E . , Gahn T . , K a p p e i F . , and A e i k e n s B. (1987) Segmentation of M i c r o s c o p i c c e l l s c e n e s . A n a l y t i c a l and Q u a n t i t a t i v e C y t o l o g y and Histology, 9:197-211 L i n Y . K . , and Fu K . S . (1981) S e g m e n t a t i o n o f P a p a n i c o l a o u smear i m a g e s . A n a l y t i c a l and Q u a n t i t a t i v e C y t o l o g y , 3:201-206 M a c A u l a y C , a n d P a l c i c B. ( 1 9 8 8 ) A c o m p a r i s o n o f some q u i c k a n d s i m p l e t h r e s h o l d s e l e c t i o n methods f o r s t a i n e d c e l l s . A n a l y t i c a l and Q u a n t i t a t i v e Cytology, 10:134-138 Mantas  J. (1987) Methodologies in pattern recognition and analysis - a b r i e f survey. Pattern Recognition, 20(1):1-6  Marr,  D.  (1982)  image  V i s i o n . W . H . F r e e m a n a n d Company, New Y o r k  M a s t i n G . A . (1985) A d a p t i v e f i l t e r s f o r d i g i t a l image n o i s e s m o o t h i n g : an e v a l u a t i o n . C o m p u t e r V i s i o n , G r a p h i c s , a n d Image P r o c e s s i n g , 31:103-121 Miller  D . R . , L e i k i n S . , A l b o V . , S a t h e r H . , K a r o n M . , Hammond D. ( 1 9 8 1 ) Prognostic Importance of Morphology (FAB C l a s s i f i c a t i o n ) in Childhood Acute Lymphoblastic Leukemia (ALL). B r i t i s h Journal of Hematology, 48:199-206  Megla G . K . (1973) Cytologica,  The LARC a u t o m a t i c 17(1):3-14  white blood c e l l  analyzer:  Acta  Mui J . K . , F u K . S . , and Bacus J . W . (1977) Automated c l a s s i f i c a t i o n of blood cell neutrophils. The J o u r n a l of Histochemistry and Cytochemistry, 25(7):633-640 N o r g r e n P . E . , K u l k a r n i A . V . , a n d Graham M . D . ( 1 9 8 1 ) L e u k o c y t e image a n a l y s i s i n the d i f f 3 system. Pattern Recognition, 13(4):299-314  O h l a n d e r R . , P r i c e K. , a n d R e d d y R. ( 1 9 7 8 ) r e c u r s i v e r e g i o n s p l i t t i n g method. P r o c e s s i n g , 8:313-333 Pal  76 P i c t u r e segmentation using a C o m p u t e r G r a p h i c s a n d Image  S . K . , and P a l N.R. (1987) S e g m e n t a t i o n b a s e d on measures contrast, homogeneity, and r e g i o n s i z e . IEEE T r a n s a n c t i o n s Systems, Man, and C y b e r n e t i c s , 1 7 ( 5 ) : 8 5 7 - 8 6 8  of on  Palcic  B . a n d J a g g i B. ( 1 9 8 9 ) Image C y t o m e t r y S y s t e m f o r Morphometric Measurements of L i v e C e l l s . B i o - i n s t r u m e n t a t i o n : Developments and A p p l i c a t i o n s . E d . W i s e D . L . , B u t t e r w o r t h P u b l i s h e r s , S t o n e h a m , MA.  Palcic  B . , Poon S . S . S . , T h u r s t o n G. and J a g g i B. (1988) Time l a p s e r e c o r d s o f c e l l s i n v i t r o u s i n g o p t i c a l memory d i s k a n d C e l l Analyzer. J o u r n a l of Tissue Culture Methods, ll(l):19-22.  Palcic  B . , J a g g i B. a n d N o r d i n J . A . ( 1 9 8 7 ) Processing Scanner. United States Patent  Dynamic M i c r o s c o p e #4,700,298.  Image  P h i l i p J . and L u n d s t e e n C. (1985) S e m i a u t o m a t e d Chromosome A n a l y s i s : C l i n i c a l Test, C l i n i c a l G e n e t i c s , 27:140-146  A  Poon S . S . S . , J a g g i B . , and P a l c i c B. (1987) C e l l Recognition Algorithm f o r the C e l l A n a l y z e r . IEEE P r o c e e d i n g s , E n g i n e e r i n g i n M e d i c i n e and B i o l o g y , 3:1455-1456 Poon  S . S . S . , J a g g i B . , S p a d i n g e r I., and P a l c i c B. (1989) methods u s e d i n the C e l l A n a l y z e r . IEEE P r o c e e d i n g s , i n M e d i c i n e and B i o l o g y , ( i n p r e s s )  Poulin  N. , Poon S . S . S . , K a n d o l a G. and d e t e c t i o n o f metaphase chromosomes. i n M e d i c i n e and B i o l o g y ( i n p r e s s ) .  Pratt  W.K. York  (1978)  Digital  Image  P a l c i c B. (1989) IEEE P r o c e e d i n g s ,  Processing.  John Wiley  P r e s t o n K . J r . (1987) High-resolution leukocyte analyzers: and p r o s p e c t i v e . A p p l i e d Optics 26(16):3258-3265 Preston K.Jr. (1980) Automation of the analysis A n a l y t i c a l and Q u a n t i t a t i v e C y t o l o g y , 2 : 1 - 1 4 Preston K.Jr. Picture 209-293  (1976) Digital Picture Analysis in A n a l y s i s . Ed. Rosenfeld A. , Springer  R o s e n f e l d A . , and Kak A . C . (1982) Digital Picture E d i t i o n , V o l u m e 2 . A c a d e m i c P r e s s , New Y o r k  of  Focussing Engineering  Automatic Engineering  & Sons,  New  retrospective  cell  images.  Cytology, Digital V e r l a y , New Y o r k ,  Processing.  Second  S c h w a r t z M. , a n d Shaw L . ( 1 9 7 5 ) Signal Processing: Discrete Spectral A n a l y s i s . D e t e c t i o n , and E s t i m a t i o n . M c G r a w - H i l l , New y o r k  77 Shoemaker R . L . , B a r t e l s P . H . , H i l l m a n D.W., J o n e s J . , K e s s l e r D . , Shack R.V., a n d V u k o b r a t o v i c h D. (1982) An U l t r a f a s t L a s e r Scanner Microscope for Digital Image A n a l y s i s , IEEE T r a n s a c t i o n s on Biomedical E n g i n e e r i n g , BME29(2):82-91 S p a d i n g e r I., Poon S . S . S . and P a l c i c B. (1989a) Automated d e t e c t i o n and r e c o g n i t i o n o f l i v e c e l l s i n t i s s u e c u l t u r e u s i n g image c y t o m e t r y . Cytometry ( i n p r e s s ) . S p a d i n g e r I., Poon S . S . S . and P a l c i c B. (1989b) E f f e c t o f f o c u s on c e l l detection and r e c o g n i t i o n by the Cell Analyzer. Cytometry (submitted). T a n a k a N . , I k e d a H . , Ueno T . , Mukawa A . , W a t a n a b e S . , Okamoto K. H o s o i S . , and Tsunekawa S . (1987) Automated c y t o l o g i c s c r e e n i n g system (CYBEST m o d e l 4 ) : a n i n t e g r a t e d image c y t o m e t r y s y s t e m . Applied Optics, 26(16):3301-3307 Tucker  J . H . , H u s a i n O . A . N . , W a t t s K. , F a r r o w S . , B a y l e y R. , a n d S t a r k M.H. (1987) Automated densitometry of c e l l populations in a continuous-motion imaging cell scanner. Applied Optics, 26(10):3315-3324  Tucker  J . H . (1979) A n Image A n a l y s i s S y s t e m f o r Automation Using Nuclear DNA Content, H i s t o c h e m i s t r y and C y t o c h e m i s t r y 27(1) :613-620  C e r v i c a l Cytology The Journal of  T y r e r H.W., and Pressman N . J . (1987) S t r a t e g i e s f o r automated placement of c e l l s for microscopy. A p p l i e d O p t i c s , 26(10):3308-3314 Umesh  R.M. (1988) A technique Recognition, 21(4):393-400  W a l t o n W.H. 169  (1952)  Automatic  for  Counting of  cluster  formation.  Microscope P a r t i c l e s ,  Pattern  Nature,  Wermser D . , H a u s s m a n n G . , a n d L i e d k e C . E . ( 1 9 8 4 ) Segmentation of blood smears by H i e r a r c h i a l t h r e s h o l d i n g . Computer V i s i o n , G r a p h i c s , a n d Image P r o c e s s i n g , 2 5 : 1 5 1 - 1 6 8 Young  I.T. (1972) The c l a s s i f i c a t i o n o f w h i t e b l o o d c e l l s . T r a n s a c t i o n s on B i o m e d i c a l E n g i n e e r i n g , 1 9 ( 4 ) : 2 9 1 - 2 9 8  Young  I.T., and structures. 35-40  Young  IEEE  Paskowitz I.L. (1975) Localiztion of cellular IEEE T r a n s a c t i o n s on B i o m e d i c a l E n g i n e e r i n g , 2 2 ( 1 ) :  I.T. (1988) Sampling Density and Q u a n t i t a t i v e Microscopy, A n a l y t i c a l and Q u a n t i t a t i v e C y t o l o g y and H i s t o l o g y , 4 : 2 6 9 - 2 7 5 Young J . Z . and R o b e r t s F. (1951) A F l y i n g Spot M i c r o s c o p e , Nature, 167:231  Young T.Y., and Fu K.S., Eds. (1986) Handbook o f P a t t e r n and Image P r o c e s s i n g . Academic P r e s s , London Z a h n i s e r D.J., Brenner J . F . , and S e l l e s W.D. i n automated l e u k o c y t e c l a s s i f i c a t i o n .  78 Recognition  (1986) S p e c t r a l bandwidth Cytometry, 7:518-521  Z u c k e r - F r a n k l i n D. , Greaves M.F., G r o s s i C.E., and MArmont A.M. (1988) A t l a s o f B l o o d C e l l s : F u n c t i o n and Pathology. Second E d i t i o n , E d i . Ermes, M i l a n .  79  Appendix A  The  nucleated  blood  f o l l o w i n g codes.  cells  are c l a s s i f i e d  into  and o b j e c t s w i t h codes  to  cells.  the m a l i g n a n t b l o o d  + 0 1 2 3 4 5 6 7 8 9 a b c d e f g h  with  the  O b j e c t s w i t h codes '0' t o '9' and 'a' t o 'c' b e l o n g t o  the normal b l o o d c e l l s  Code  the groups  Cell  ' f , 'g', and 'h' b e l o n g  Classification  dead c e l l ignore (too d i f f i c u l t to c l a s s i f y ) neutophil band metamyelocyte myelocyte promyelocyte blast o r t h o c h r o n i c normoblast p o l y c h r o m a t o p h i l i c normoblast b a s o p h i l i c normoblast pronormoblast lymphocyte monocyte plasma c e l l megakaryocyte macrophage lymphoblast L I lymphoblast L2 myeloblast  R e f e r t o Z u c k e r - F r a n k l i n e t a l . (1988) and Begemann and R a s t e t t e r (1979) for  examples and a d e s c r i p t i o n o f these d i f f e r e n t c e l l  types.  Poon,  S.S.S.  PUBLICATIONS: 1.  J a g g i , B . , P o o n , S . a n d P a l c i c , B. : I m p l e m e n t a t i o n and e v a l u a t i o n o f t h e DMIPS C e l l A n a l y s e r . IEEE P r o c e e d i n g s , E n g i n e e r i n g i n M e d i c i n e and B i o l o g y 3: 906-911, 1986.  2.  J a g g i , B. , Poon, S . and P a l c i c , B . : O p t i c a l Memory D i s k s i n Image D a t a b a s e Management f o r C y t o m e t r y . A p p l i e d O p t i c s p u b l . by Optical S o c i e t y of America 26: 3325-3329, 1987.  3.  P o o n , S . S . S . , J a g g i , B. and P a l c i c , B . : C e l l recognition algorithms the C e l l A n a l y s e r . IEEE P r o c . Eng. B i o l . 3: 1454-1456, 1987.  4.  Palcic, B . , Poon, S.S.S., Thurston, G. and J a g g i , B . : Time lapse r e c o r d s o f c e l l s i n v i t r o u s i n g o p t i c a l memory d i s k a n d C e l l A n a l y z e r . J . T i s s u e C u l t u r e Methods 1 1 ( 1 ) : 1 9 - 2 2 . 1988.  5.  J a g g i , B . , Poon, S . S . S . , MacAulay, C. and P a l c i c , B . : Imaging system f o r morphometric assessment o f c o n v e n t i o n a l l y and f l u o r e s c e n t l y stained cells. Cytometry 9 : 5 6 6 - 5 7 2 , 1988.  6.  S p a d i n g e r , I., P o o n , S . S . S . and P a l c i c , B. : recognition of live cells in tissue culture Cytometry. (In p r e s s , 1989).  7.  Poon, S.S.S., J a g g i , B. , S p a d i n g e r , I. and P a l c i c , methods u s e d i n the c e l l a n a l y z e r . P r o c . 11th IEEE S o c , S e a t t l e , WA. , N o v . 8 - 1 2 , 1989 ( A c c e p t e d ) .  for  Automated d e t e c t i o n and u s i n g image cytometry.  B. : Eng.  Focussing Med. B i o l .  SUBMITTED P A P E R S : 1.  Poon, S . S . S . , used in the (submitted).  2.  P o u l i n , N . , Poon, S . S . S . , o f metaphase chromosomes.  3.  S p a d i n g e r , I., detection and (submitted).  REPORTS  J a g g i , B. , S p a d i n g e r , I., Cell Analyzer. IEEE  P a l c i c , B. : Proc. Eng.  F o c u s s i n g methods Med. Biol., 1989  Kandola, G . , P a l c i c , B.: Automatic detection I E E E P r o c . E n g . M e d . B i o l . , 1989 (submitted).  Poon, S . S . S . recognition  and P a l c i c , B. : Effect by the cell analyzer.  of  f o c u s on Cytometry  cell 1989  AND A B S T R A C T S :  1.  J a g g i , B . , P o o n , S . a n d P a l c i c , B. : Implementation and e v a l u a t i o n of t h e DMIPS C e l l A n a l y s e r . IEEE Soc. C o n f e r e n c e , D a l l a s - F o r t Worth, T e x a s , November 7 - 1 0 , 1 9 8 6 .  2.  P a l c i c , B . , Poon, S . S . S . and J a g g i , B. : The d e v e l o p m e n t o f t h e C e l l A n a l y z e r f o r use i n a n a l y t i c a l c y t o l o g y . 1st I n t ' l . C o n f e r e n c e on A r t i f i c i a l I n t e l l i g e n c e Systems, Los A n g e l e s , F e b r u a r y 1-3, 1987.  Poon,  S.S.S.  3.  Poon, S . S . S . , J a g g i , B . , A n d e r s o n , G. and P a l c i c , B. : Image d a t a b a s e system used for computer-aided teaching i n cytology. 1st Int'l. C o n f e r e n c e o n A r t i f i c i a l I n t e l l i g e n c e S y s t e m s , L o s A n g e l e s , F e b r u a r y 13, 1987.  4.  J a g g i , B. , M a s s i n g , B. , MacAulay, C , Poon, S . S . S . and P a l c i c , B. : Quantitative morphometric assessment of leukemic c e l l s . 1st Int'l. C o n f e r e n c e o n A r t i f i c i a l I n t e l l i g e n c e S y s t e m s , L o s A n g e l e s , F e b r u a r y 13, 1987.  5.  J a g g i , B. , Poon, S . S . S . , P o n t i f e x , B . D . , Deen, M . J . and P a l c i c , B. : D e s i g n a n d d e v e l o p m e n t o f a s o l i d s t a t e m i c r o s c o p e f o r image cytometry. 4th I n t ' l . Cong, o f C e l l B i o l . , M o n t r e a l , Aug. 14-19, 1988.  6.  Poon, S . S . S . , J a g g i , B. and P a l c i c , B . : Colour segmentation of blood smears. 4 t h I n t ' l . Cong, of C e l l B i o l . , M o n t r e a l , Aug. 14-19, 1988.  7.  P a l c i c , B . , Poon S . S . S . and J a g g i , B . : R e c o g n i t i o n and a n a l y s i s of l i v e unstained c e l l s . D i g i t a l Imaging T e c h n o l o g y . f o r Oncology, T e r r y Fox Workshop, Vancouver, B . C . , October 19-22, 1988.  8.  Poon, S . S . S . , Ward, R . K . , Beddoes, M . P . and P a l c i c , B . : Detection of leukocytes i n Wright's stained c e l l s . D i g i t a l Imaging Technology f o r Oncology, T e r r y Fox Workshop, Vancouver, B . C . , October 19-22, 1988.  9.  J a g g i , B . , Poon, S . S . S . , P o n t i f e x , B . , F e n g l e r , J . P . and P a l c i c , B . : The d e v e l o p m e n t o f a q u a n t i t a t i v e m i c r o s c o p e f o r i m a g e c y t o m e t r y . 1st M t g . E u r . S o c . f o r A n a l y t . C e l l . P a t h . , S c h l o s s E l m a u , F R G , November 1 2 17, 1989.  10.  Poon, S . S . S . , Ward, R . K . , Beddoes, M . P . and P a l c i c , B . : D e t e c t i o n and segmentation of n u c l e a t e d c e l l s i n blood smears. 1st Mtg. Eur. Soc. for A n a l y t . C e l l . P a t h . , S c h l o s s E l m a u , F R G , November 1 2 - 1 7 , 1 9 8 9 .  SCHOLARSHIPS AND AWARDS: 1985 1984 1983 1980  University Scholarship, U.B.C. University Scholarship, U.B.C. M a c k e n z i e Swan M e m o r i a l S c h o l a r s h i p B . C . Government S c h o l a r s h i p G o l d L i o n Award (honour s t u d e n t i n grades 10-12) Book P r o f i c i e n c y Awards ( t o p s t u d e n t i n E l e c t r o n i c s and S c i e n c e )  

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.831.1-0064935/manifest

Comment

Related Items