Enhancing Skin Lesion Classification Generalization with Active Domain Adaptation

TL;DR

The paper addresses the generalization gap in skin lesion classification caused by cross-domain shifts. It proposes a hybrid approach that first leverages DINO-based self-supervised learning, followed by SSL retraining on in-domain data, source-domain fine-tuning, and finally active domain adaptation on target domains. Results indicate that SSL retraining substantially improves cross-domain performance over a baseline BSP-UDA method on 9 of 10 targets, and that ADA methods—especially AADA-DANN—provide additional gains with some dataset-dependent variance. The work demonstrates robust cross-domain skin lesion classification across ten target datasets and offers a practical pathway for clinical deployment under limited labeling budgets; future work could explore vision transformers as backbone architectures.

Abstract

We propose a method to improve the generalization of skin lesion classification models by combining self-supervised learning (SSL) and active domain adaptation (ADA). The main steps of the approach include selection of an SSL pre-trained model on natural image datasets, subsequent SSL retraining on all available skin-lesion datasets, fine-tuning of the model on source domain data with labels, and application of ADA methods on target domain data. The efficacy of the proposed approach is assessed in ten skin lesion datasets with five different ADA methods, demonstrating its potential to improve generalization in settings with different amounts of domain shifts.

Figures (5)

• Figure 1: Proposed workflow.
• Figure 2: DINO SSL architecture.
• Figure 3: Active domain adaptation architecture.
• Figure 4: Comparing ADA methods with AUPRC value.
• Figure 5: Comparing ADA methods.