Deep Dirichlet uncertainty for unsupervised out-of-distribution detection of eye fundus photographs in glaucoma screening

TL;DR

This paper introduces a Dirichlet-based model for uncertainty estimation in glaucoma screening from fundus images, improving robustness to out-of-distribution cases and achieving top performance in the AIROGS challenge.

Contribution

The paper presents a novel Dirichlet distribution approach for uncertainty estimation that does not require out-of-distribution data during training.

Findings

Achieved the highest average score in the AIROGS challenge

Demonstrated improved robustness to out-of-distribution cases

Provided reliable class-wise probability and uncertainty estimates

Abstract

The development of automatic tools for early glaucoma diagnosis with color fundus photographs can significantly reduce the impact of this disease. However, current state-of-the-art solutions are not robust to real-world scenarios, providing over-confident predictions for out-of-distribution cases. With this in mind, we propose a model based on the Dirichlet distribution that allows to obtain class-wise probabilities together with an uncertainty estimation without exposure to out-of-distribution cases. We demonstrate our approach on the AIROGS challenge. At the start of the final test phase (8 Feb. 2022), our method had the highest average score among all submissions.