Estimating uncertainty of geographic atrophy segmentations with Bayesian deep learning

Abstract

PurposeTo apply methods for quantifying uncertainty of deep learning segmentation of geographic atrophy (GA).Study DesignRetrospective analysis of optical coherence tomography (OCT) images and model comparison.Participants126 eyes from 87 participants with GA in the SWAGGER cohort of the Non-exudative Age-Related Macular Degeneration Imaged with Swept-Source Optical Coherence Tomography (SS-OCT) study.MethodsThe manual segmentations of GA lesions were conducted on structural sub-RPE en face images from the SS-OCT images. Models were developed for two approximate Bayesian deep learning techniques, Monte Carlo dropout and ensemble, to assess the uncertainty of GA semantic segmentation and compared to a traditional deep learning model.Main Outcome MeasuresModel performance (Dice score) was compared. Uncertainty was calculated using the formula for Shannon Entropy.ResultsThe output of both Bayesian technique models showed a greater number of pixels with high entropy than the standard model. Dice scores for the Monte Carlo dropout method (0.90, 95% confidence interval [CI] 0.87-0.93) and the ensemble method (0.88, 95% CI 0.85-0.91) were significantly higher (p<.001) than for the traditional model (0.82, 95% CI 0.78-0.86).ConclusionsQuantifying the uncertainty in a prediction of GA may improve trustworthiness of the models and aid clinicians in decision making. The Bayesian deep learning techniques generated pixel-wise estimates of model uncertainty for segmentation, while also improving model performance compared with traditionally trained deep learning models.