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.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **the label noise problem in fundus image classification**. Specifically, label noise significantly affects the performance of deep neural networks, especially in computer - aided diagnosis tasks. Fundus image classification is crucial for diagnosing various ophthalmic diseases. However, due to the limited quality and quantity of labeled data and possible errors in the labeling process (such as inconsistent expert opinions, differences in image quality, etc.), label noise inevitably exists in fundus image datasets. These problems make the model prone to over - fitting noisy labels, thus severely degrading the classification and generalization performance. ### Problem Background 1. **Importance of Fundus Image Classification**: Fundus image classification plays an important role in computer - aided diagnosis of ophthalmic diseases. 2. **Impact of Label Noise**: Label noise can confuse supervised training, causing the model to over - fit noisy labels, thereby reducing classification and generalization performance. 3. **Limitations of Existing Methods**: Existing methods for handling label noise in natural images perform poorly on fundus image datasets because fundus image datasets usually contain hard samples and are class - imbalanced. ### Solution To solve the above problems, the authors propose a new framework - **Self - Supervised Pretraining and Robust Adaptive Credal Loss (SSP - RACL)**, which specifically includes the following aspects: - **Self - Supervised Pretraining (SSP)**: Use Masked Autoencoders (MAE) for pretraining to extract features that are not affected by label noise. - **Robust Adaptive Credal Loss (RACL)**: Construct a possibility distribution by introducing a confidence threshold and an adaptive label relaxation parameter to provide more reliable true - value estimates, thereby effectively suppressing the memory effect. - **Asymmetric Noise Generation Guided by Clinical Knowledge**: Simulate noisy fundus image datasets in the real world to better reflect misdiagnosis patterns in clinical practice. ### Method Advantages - **Improve Robustness**: Through self - supervised pretraining and robust adaptive credal loss, the robustness of the model to noisy labels is improved. - **Simplify the Process**: No additional evaluation set is required, simplifying the model training process. - **Enhance Generalization Ability**: Ensure the reliability and generalization ability of the model in clinical applications. ### Experimental Results The experimental results show that the SSP - RACL framework is superior to existing methods in handling label noise and performs well on multiple performance indicators on the OIA - ODIR and Kaggle DR datasets. In summary, this paper aims to solve the label noise problem in fundus image classification by proposing a new framework, thereby improving the robustness and classification performance of the model.

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

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

A Label Noise Robust Stacked Auto-Encoder Algorithm for Inaccurate Supervised Classification Problems

Learning with Noisy Labels Via Self-supervised Adversarial Noisy Masking

Label-noise-tolerant medical image classification via self-attention and self-supervised learning

Suppressing label noise in medical image classification using mixup attention and self-supervised learning

Adaptive Contrastive Learning for Learning Robust Representations under Label Noise.

Robust Medical Image Classification from Noisy Labeled Data with Global and Local Representation Guided Co-training

Self-Adaptive Transfer Learning for Multicenter Glaucoma Classification in Fundus Retina Images

How does self-supervised pretraining improve robustness against noisy labels across various medical image classification datasets?

Superpixel-guided class-level denoising for unsupervised domain adaptive fundus image segmentation without source data

Curriculum Fine-tuning of Vision Foundation Model for Medical Image Classification Under Label Noise

Learning with Noisy Labels Via Self-Reweighting from Class Centroids

Robust PCA with Lw,∗ and L2,1 Norms: A Novel Method for Low-Quality Retinal Image Enhancement

QMix: Quality-aware Learning with Mixed Noise for Robust Retinal Disease Diagnosis

Bayesian statistics guided label refurbishment mechanism: Mitigating label noise in medical image classification

Improving Medical Images Classification with Label Noise Using Dual-Uncertainty Estimation

Learning from Noisy Labels with Coarse-to-Fine Sample Credibility Modeling

Learning Image Labels On-the-fly for Training Robust Classification Models

ProCNS: Progressive Prototype Calibration and Noise Suppression for Weakly-Supervised Medical Image Segmentation

Boosting Semi-Supervised Face Recognition With Noise Robustness