Continual Self-supervised Learning Considering Medical Domain Knowledge in Chest CT Images

TL;DR

This work proposes a novel continual self-supervised learning method (CSSL) considering medical domain knowledge in chest CT images by incorporating an enhanced dark experience replay into CSSL and maintaining both diversity and representativeness within the rehearsal buffer of DER, enabling the model to learn more richer and robust feature representations.

Abstract

We propose a novel continual self-supervised learning method (CSSL) considering medical domain knowledge in chest CT images. Our approach addresses the challenge of sequential learning by effectively capturing the relationship between previously learned knowledge and new information at different stages. By incorporating an enhanced DER into CSSL and maintaining both diversity and representativeness within the rehearsal buffer of DER, the risk of data interference during pretraining is reduced, enabling the model to learn more richer and robust feature representations. In addition, we incorporate a mixup strategy and feature distillation to further enhance the model's ability to learn meaningful representations. We validate our method using chest CT images obtained under two different imaging conditions, demonstrating superior performance compared to state-of-the-art methods.

Figures (3)

• Figure 1: Overview of the proposed CSSL method.
• Figure 2: Overview illustration of stage 2.
• Figure 3: Classification accuracy across different epochs during fine-tuning. All methods use ViT-B model.