Distributional Shifts in Automated Diabetic Retinopathy Screening

TL;DR

This paper introduces a Dirichlet Prior Network framework to improve the robustness of diabetic retinopathy screening models by effectively detecting out-of-distribution images and reducing false positives from non-retina inputs.

Contribution

It proposes a novel framework combining OOD detection with DR classification to enhance generalizability and reliability of automated screening systems.

Findings

Effectively detects out-of-distribution images

Reduces false positives from non-retina images

Improves model robustness in real-world datasets

Abstract

Deep learning-based models are developed to automatically detect if a retina image is `referable' in diabetic retinopathy (DR) screening. However, their classification accuracy degrades as the input images distributionally shift from their training distribution. Further, even if the input is not a retina image, a standard DR classifier produces a high confident prediction that the image is `referable'. Our paper presents a Dirichlet Prior Network-based framework to address this issue. It utilizes an out-of-distribution (OOD) detector model and a DR classification model to improve generalizability by identifying OOD images. Experiments on real-world datasets indicate that the proposed framework can eliminate the unknown non-retina images and identify the distributionally shifted retina images for human intervention.