Markov Probability Model Assessment of Conversion Rates from Intermediate Pathologies to Geographic Atrophy.

Abstract

Purpose : To evaluate the feasibility of assessing early pathology indicators for geographic atrophy (GA) expansion in dry age-related macular degeneration (AMD).

Methods : This was an IRB-approved retrospective image analysis study. Using a machine learning-enhanced segmentation platform, the outer nuclear layer (ONL), ellipsoid zone (EZ), retinal pigment epithelium, and bruchs membrane were all segmentated with certified reader B-scan level validation. The first intermediate pathology evaluated was quantitative iRORA, defined as EZ loss, ONL attenuation with an ONL-RPE thickness of < 50 microns, and RPE disruption. The second intermediate pathology utilized a machine learning model to identifiy “at-risk” EZ. At-Risk EZ was then evaluated in 3 categories: total at-risk EZ, at-risk drusen (i.e., areas of concurrent at-risk EZ and RPE deflections consistent with drusen) and thin at-risk EZ. The B-scan based intermediate pathologies and segmented GA were then assembled into 2D enface maps. Time points were aligned in the X and Y en face axis using the foveal centers. Pixels were categorized based on their starting and ending condition into healthy, intermediate pathology, or GA conditions. The probability of conversion between each of those states was measured, allowing the interrogation of causal relationships between disease states.

Results : One hundred and nineteen patients at 0, 1, 3, and 5 years were segmented and foveal centers identified by experts. Total at-risk EZ appeared to be the most rapidly expanding pathology with the highest amounts of conversion from healthy to at-risk EZ, and conversion from at-risk EZ to GA.

Conclusions : At-risk EZ appears to be highly dynamic with newly developing intermediate pathology as well as conversion to GA seen during a 2 year period. Further research is needed to better understand the interplay between various biomarkers for progression risk to GA.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.

 

Table 1: Labels starting and ending conditions.
Table 2. Mean Conversion Percentage of the Macular Cube Area