CauDR: A Causality-inspired Domain Generalization Framework for Fundus-based Diabetic Retinopathy Grading

TL;DR

This paper introduces CauDR, a causality-inspired framework for diabetic retinopathy grading that reduces spurious correlations and improves model generalization across different fundus imaging domains.

Contribution

It proposes a novel causal model and a new benchmark dataset to enhance the robustness and generalizability of DR grading algorithms.

Findings

CauDR achieves state-of-the-art performance on the 4DR benchmark.

Incorporating causality reduces domain shift effects.

The framework improves diagnostic accuracy across diverse datasets.

Abstract

Diabetic retinopathy (DR) is the most common diabetic complication, which usually leads to retinal damage, vision loss, and even blindness. A computer-aided DR grading system has a significant impact on helping ophthalmologists with rapid screening and diagnosis. Recent advances in fundus photography have precipitated the development of novel retinal imaging cameras and their subsequent implementation in clinical practice. However, most deep learning-based algorithms for DR grading demonstrate limited generalization across domains. This inferior performance stems from variance in imaging protocols and devices inducing domain shifts. We posit that declining model performance between domains arises from learning spurious correlations in the data. Incorporating do-operations from causality analysis into model architectures may mitigate this issue and improve generalizability. Specifically, a novel universal structural causal model (SCM) was proposed to analyze spurious correlations in fundus imaging. Building on this, a causality-inspired diabetic retinopathy grading framework named CauDR was developed to eliminate spurious correlations and achieve more generalizable DR diagnostics. Furthermore, existing datasets were reorganized into 4DR benchmark for DG scenario. Results demonstrate the effectiveness and the state-of-the-art (SOTA) performance of CauDR.