DerMAE: Improving skin lesion classification through conditioned latent diffusion and MAE distillation

TL;DR

This paper introduces DerMAE, a method that enhances skin lesion classification by generating synthetic images with class-conditioned diffusion models, pretraining with MAE, and distilling knowledge into lightweight models for clinical deployment.

Contribution

It combines class-conditioned diffusion, MAE pretraining, and knowledge distillation to improve skin lesion classification and enable efficient on-device inference.

Findings

Synthetic data improves classification accuracy.

MAE pretraining enhances feature robustness.

Distilled models perform well on mobile devices.

Abstract

Skin lesion classification datasets often suffer from severe class imbalance, with malignant cases significantly underrepresented, leading to biased decision boundaries during deep learning training. We address this challenge using class-conditioned diffusion models to generate synthetic dermatological images, followed by self-supervised MAE pretraining to enable huge ViT models to learn robust, domain-relevant features. To support deployment in practical clinical settings, where lightweight models are required, we apply knowledge distillation to transfer these representations to a smaller ViT student suitable for mobile devices. Our results show that MAE pretraining on synthetic data, combined with distillation, improves classification performance while enabling efficient on-device inference for practical clinical use.