Improving Medical Image Classification in Noisy Labels Using Only Self-supervised Pretraining

TL;DR

This paper investigates how different self-supervised pretraining methods, including contrastive and pretext task-based approaches, improve medical image classification performance in noisy label settings, especially for small datasets with subtle differences.

Contribution

It is the first to compare contrastive and pretext task-based self-supervised pretraining for medical images with noisy labels, demonstrating their effectiveness in enhancing robustness and feature learning.

Findings

Self-supervised pretraining improves robustness to noisy labels in medical images.

Contrastive and pretext task-based methods both enhance feature quality.

Pretrained models outperform randomly initialized models in noisy label scenarios.

Abstract

Noisy labels hurt deep learning-based supervised image classification performance as the models may overfit the noise and learn corrupted feature extractors. For natural image classification training with noisy labeled data, model initialization with contrastive self-supervised pretrained weights has shown to reduce feature corruption and improve classification performance. However, no works have explored: i) how other self-supervised approaches, such as pretext task-based pretraining, impact the learning with noisy label, and ii) any self-supervised pretraining methods alone for medical images in noisy label settings. Medical images often feature smaller datasets and subtle inter class variations, requiring human expertise to ensure correct classification. Thus, it is not clear if the methods improving learning with noisy labels in natural image datasets such as CIFAR would also help with medical images. In this work, we explore contrastive and pretext task-based self-supervised pretraining to initialize the weights of a deep learning classification model for two medical datasets with self-induced noisy labels -- NCT-CRC-HE-100K tissue histological images and COVID-QU-Ex chest X-ray images. Our results show that models initialized with pretrained weights obtained from self-supervised learning can effectively learn better features and improve robustness against noisy labels.