Prediction of visual function from automatically quantified optical coherence tomography biomarkers in patients with geographic atrophy using machine learning

Abstract

Geographic atrophy (GA) is a vision-threatening manifestation of age-related macular degeneration (AMD), one of the leading causes of blindness globally. Objective, rapid, reliable, and scalable quantification of GA from optical coherence tomography (OCT) retinal scans is necessary for disease monitoring, prognostic research, and clinical endpoints for therapy development. Such automatically quantified biomarkers on OCT are likely to further elucidate structure-function correlation in GA and thus the pathophysiological mechanisms of disease development and progression. In this work, we aimed to predict visual function with machine-learning applied to automatically acquired quantitative imaging biomarkers in GA. A post-hoc analysis of data from a clinical trial and routine clinical care was conducted. A deep-learning automated segmentation model was applied on OCT scans from 476 eyes (325 patients) with GA. A separate machine learning prediction model (Random Forest) used the resultant quantitative OCT (qOCT) biomarkers to predict cross-sectional visual acuity under standard (VA) and low luminance (LLVA). The primary outcome was regression coefficient (r²) and mean absolute error (MAE) for cross-sectional VA and LLVA in Early Treatment Diabetic Retinopathy Study (ETDRS) letters. OCT parameters were predictive of VA (r² 0.40 MAE 11.7 ETDRS letters) and LLVA (r² 0.25 MAE 12.1). Normalised random forest feature importance, as a measure of the predictive value of the three constituent features of GA; retinal pigment epithelium (RPE)-loss, photoreceptor degeneration (PDR), hypertransmission and their locations, was reported both on voxel-level heatmaps and ETDRS-grid subfields. The foveal region (46.5%) and RPE-loss (31.1%) had greatest predictive importance for VA. For LLVA, however, non-foveal regions (74.5%) and PDR (38.9%) were most important. In conclusion, automated qOCT biomarkers demonstrate predictive significance for VA and LLVA in GA. LLVA is itself predictive of GA progression, implying that the predictive qOCT biomarkers provided by our model are also prognostic.

Figure 1

Image analysis workflow. (a) For each OCT volume, all b-scans were segmented for RPE-loss (orange), photoreceptor degeneration (blue), hypertransmission (red), and RPE and outer retinal atrophy (RORA; green). RORA is taken to be overlapping regions of the three former features i.e. co-occurrence as per a-scan. Exemplar segmentation of a single b-scan and its axis along en face fundus photograph. (b) Resultant feature probability maps from total volume segmentations collectively presented by projection onto en face fundus photograph. Colour legends represent target feature probability. Manual central foveal point annotation permitted interpolation of a given voxel’s localisation in relation to the fovea. (c) ETDRS regions were also considered wherein the macula is considered as a 6 mm diameter circle divided into 9 areas: central foveal area (1 mm diameter); 4 parafoveal (collectively span 3 mm diameter); and 4 perifoveal areas. (d) Here, the mean feature probability within each region is displayed.

Figure 2

Heatmap of relative feature predictive value. Normalised random forest feature importance as a measure of the predictive value of the four considered features of GA; RPE-loss, photoreceptor degeneration, and hypertransmission and RORA and their locations relative to the fovea to the predicted value for (a) standard visual acuity (Overall cohort) and (b) low luminance visual acuity (FILLY cohort). Feature importance values were averaged across 100 bootstraps of the dataset.