Segmentation of biological cell images for sonification

1
Segmentation of biological cell images for
sonification

• Alistair D N Edwards, Geneviève Hines, Andy Hunt
  
• University of York, UK
• http//www-users.cs.york.ac.uk/alistair/research/
  cervical/

2
Cervical screening

• Very successful in preventing cancer
• saves around 4,500 lives each year in UK
• Relies on detecting pre-cancerous changes in
  cervical cells
• Depends on visual scanning of cell samples

3
Cervical cell sample
4
Notes on samples

• Smears or liquid-based
• Processed with dyes Papanicolaou or Pap stain
• Each slide contains 10,000-20,000 cells

5
Automatic screening

• Primary screening
• Quality control
• Manual screening still required

6
Conventional screening

• Decisions are not always clear-cut
• Errors are made

7
Sonification

• If screeners hear a representation as well as
  seeing it, they may make fewer errors

8
Note

• Cannot simply generate alarm sound when a
  suspicious cell is encountered
• That would amount to fully automated screening

9
Features

• In order to sonify the image it must first be
  processed
• Nuclei are important
• Shape
• Relative size
• Texture
• Chromaticity

10
Unprocessed slide
11
Segmentation

• Initial segmentation using the mean-shift method
  as implemented in EDISON software

12
Segmentation

• Edge strength assessment

13
Segmentation

• Regions most likely to be nuclei are selected
• Region size

14
Segmentation

• Regions most likely to be nuclei are selected
• Region size
• Gray value - global

15
Segmentation

• Regions most likely to be nuclei are selected
• Region size
• Gray value - local

16
Sonification plan

• Based on nuclear textures
• Collect a set of typically normal cells
• and typically abnormal cells
• Select most interesting cell in view
• Sonify that within the space delimited by the two
  sets
• Have not yet chosen the sounds

17
Conclusions

• Sonification of cervical cells may improve
  diagnosis
• Preparing images requires considerable effort
• Sonification results due soon

18
Acknowledgements

• Sponsored by the UK Engineering and Physical
  Sciences Research Council (EPSRC) (EP/C512413/1)
• Leeds Teaching Hospitals NHS Trust