SSP-RACL: Classification of Noisy Fundus Images with Self-Supervised Pretraining and Robust Adaptive Credal Loss

TL;DR

This paper introduces SSP-RACL, a novel framework combining self-supervised pretraining and adaptive loss to improve fundus image classification accuracy under noisy labels, outperforming existing methods.

Contribution

The paper proposes a new robust framework that integrates Masked Autoencoders and a confidence-based adaptive loss to effectively handle label noise in fundus image datasets.

Findings

Outperforms existing noise-handling methods in fundus image classification.

Effectively suppresses memorization of noisy labels.

Demonstrates robustness with clinical knowledge-based noise simulation.

Abstract

Fundus image classification is crucial in the computer aided diagnosis tasks, but label noise significantly impairs the performance of deep neural networks. To address this challenge, we propose a robust framework, Self-Supervised Pre-training with Robust Adaptive Credal Loss (SSP-RACL), for handling label noise in fundus image datasets. First, we use Masked Autoencoders (MAE) for pre-training to extract features, unaffected by label noise. Subsequently, RACL employ a superset learning framework, setting confidence thresholds and adaptive label relaxation parameter to construct possibility distributions and provide more reliable ground-truth estimates, thus effectively suppressing the memorization effect. Additionally, we introduce clinical knowledge-based asymmetric noise generation to simulate real-world noisy fundus image datasets. Experimental results demonstrate that our proposed method outperforms existing approaches in handling label noise, showing superior performance.