1 / 30

A PERIMETRIC RE-TEST ALGORITHM THAT IS SIGNIFICANTLY MORE ACCURATE THAN CURRENT PROCEDURES

A PERIMETRIC RE-TEST ALGORITHM THAT IS SIGNIFICANTLY MORE ACCURATE THAN CURRENT PROCEDURES. Allison McKendrick Department of Optometry and Vision Science University of Melbourne. Andrew Turpin School of Computer Science and Information Technology RMIT University, Melbourne. Darko Jankovic

Download Presentation

A PERIMETRIC RE-TEST ALGORITHM THAT IS SIGNIFICANTLY MORE ACCURATE THAN CURRENT PROCEDURES

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. A PERIMETRIC RE-TEST ALGORITHM THAT IS SIGNIFICANTLY MORE ACCURATE THAN CURRENT PROCEDURES Allison McKendrick Department of Optometry and Vision Science University of Melbourne Andrew Turpin School of Computer Science and Information Technology RMIT University, Melbourne Darko Jankovic Department of Optometry and Vision Science University of Melbourne

  2. Possible re-test algorithms • Use individual presentation information… • Use test Hill of Vision to bias re-test • Continue the previous test with “next” termination criteria • More reversals (staircase, MOBS) • Tighter PDF standard deviation (Bayesian) • Fixed number of presentaitons • Use test thresholds to seed re-test • Starting point for staircase (FT From Prior) • Initialisation of MOBS stacks • Centre a PDF around threshold (ZEST, SITA)

  3. Prior distribution (before first presentation)

  4. After 1 presentation

  5. After 2 presentations

  6. After 3 presentations

  7. After 4 presentations

  8. Gaussian with standard deviation 3dB After 5 presentations

  9. Computer Simulations 350 realpatients 486 procedures • Continued ZEST • Termination Criteria • Fixed # presentations 4,5,6 • Standard deviation 0.7, 0.8, 0.9, 1.0 • LF • Steep, steeper, steepest • Seeded ZEST • PDFs • Gaussian standard deviation 2,3,4 dB • Step function, width 4,6,8,10 dB • LF • Steep, steeper, steepest • Termination criteria • Fixed # presentations 4,5,6 • Standard deviation 0.7, 0.8, 0.9, 1.0 • MOBS • Stack initialisation 2, 3, 4 dB • Termination criteria: 2, 3 reversals 2, 3, 4 width 8 Patientmodels

  10. ITA S eeded Zest S ontinued Zest C Bengtsson et al, ACTA ‘97 Performance: No Error, No Change 2 ull Threshold F Mean absolute error (dB) est Z 1 4 5 6 7 Mean number of presentations

  11. ITA S Performance: General Height -3dB 2 ontinued Zest C ull Threshold F Mean absolute error (dB) eeded Zest S est Z 1 4 5 6 7 Mean number of presentations

  12. Problems Continue • General Height change ignored, need many presentations to get right answer if GH changes, and there is still a bias towards original test value Seed • Could adjust seed if GH change known • Estimate with “primary points” algorithm • Would be slower than Full Threshold (and SITA) Katz et al, IOVS 1632

  13. Speeding up GH-corrected Seed • Spend 2 or 3 presentations per location checking if threshold not less than last time (multi-sample supra-threshold) • If so, then do no more for that location • Otherwise, assume threshold decreased, and seed a ZEST accordingly McKendrick & Turpin, OVS 2005

  14. Automated Static Perimetry, 1999, Anderson & Patella

  15. Test Re-Test 27 29 31 30 31 31 General Height decrease of 2dB Supra-threshold decrement of 2dB So multi-sample all locations at previous less 4dB If see this 2 of 3 times, then just use previous threshold - 2dB else do a full ZEST on the location

  16. 31 29 31 30 33 31 31 General Height decrease of 2dB Supra-threshold decrement of 2dB So test all locations at previous less 4dB If see this 2 of 3 times, then just use previous threshold - 2dB else do a full ZEST on the location

  17. ITA S Performance with no error 2 ull Threshold F Mean absolute error (dB) ew eeded Zest est N S Z 1 ontinued Zest C 4 5 6 7 Mean number of presentations

  18. ITA S General height -3dB 2 ontinued Zest C ull Threshold F Mean absolute error (dB) eeded Zest S ew N est Z 1 4 5 6 7 Mean number of presentations

  19. Conclusions • Continuing previous procedure doesn’t work • Seeding a ZEST with a Gaussian pdf about previous threshold works, but is slow • Adding multi-sampling supra-threshold step gives speed and accuracy gains • The resulting re-test procedure is as fast, but more accurate, than existing test algorithms BUT does not detect an isolated increase in threshold

  20. Hill of Vision Approach • Alter eccentricity adjustments in growth pattern based on individual’s HoV • Takes into account General Height change • Very small gains, but not really worth the effort

  21. After 6 presentations

  22. After 7 presentations

  23. After 8 presentations

  24. After 9 presentations

More Related