Search papers, labs, and topics across Lattice.
The paper addresses the challenge of domain shift in skin lesion classification by proposing a contrastive meta-domain adaptation strategy to transfer visual representations from dermoscopic datasets to clinical image domains. They hypothesize that visual artifacts and domain shifts degrade deep learning performance and aim to improve generalization robustness by leveraging visual meta-domains. The experiments demonstrate consistent improvements in classification performance across multiple dermatology datasets, effectively reducing the performance gap between dermoscopic and clinical images.
Bridging the gap between dermoscopic and clinical skin lesion images can be achieved through contrastive meta-domain adaptation, enhancing the robustness of deep learning models in real-world clinical settings.
Deep learning models for dermatological image analysis remain sensitive to acquisition variability and domain-specific visual characteristics, leading to performance degradation when deployed in clinical settings. We investigate how visual artifacts and domain shifts affect deep learning-based skin lesion classification. We propose an adaptation strategy, grounded in the idea of visual meta-domains, that transfers visual representations from larger dermoscopic datasets into clinical image domains, thereby improving generalization robustness. Experiments across multiple dermatology datasets show consistent gains in classification performance and reduced gaps between dermoscopic and clinical images. These results emphasize the importance of domain-aware training for deployable systems.
