Towards Automated Differential Diagnosis of Skin Diseases Using Deep Learning and Imbalance-Aware Strategies

TL;DR

This work tackles automated skin disease diagnosis under limited specialist availability by developing a Swin Transformer–based framework augmented with BatchFormer, Focal Loss, and ReduceLROnPlateau to address dataset imbalance on the ISIC2019 benchmark. It systematically evaluates segmentation-based preprocessing and data augmentation, finding that SAM can hurt performance while AutoAugment and Elastic Deformation improve generalization, especially for tail classes. The best configuration—Swin Transformer with BatchFormer and imbalance-aware losses—achieves 87.71% accuracy across eight skin lesion classes, demonstrating strong potential for clinical decision support and patient self-assessment. The study highlights the importance of tailored imbalance strategies and domain-oriented augmentation in medical imaging, and points to future work in domain-specific fine-tuning and more robust imbalance-handling techniques.

Abstract

As dermatological conditions become increasingly common and the availability of dermatologists remains limited, there is a growing need for intelligent tools to support both patients and clinicians in the timely and accurate diagnosis of skin diseases. In this project, we developed a deep learning based model for the classification and diagnosis of skin conditions. By leveraging pretraining on publicly available skin disease image datasets, our model effectively extracted visual features and accurately classified various dermatological cases. Throughout the project, we refined the model architecture, optimized data preprocessing workflows, and applied targeted data augmentation techniques to improve overall performance. The final model, based on the Swin Transformer, achieved a prediction accuracy of 87.71 percent across eight skin lesion classes on the ISIC2019 dataset. These results demonstrate the model's potential as a diagnostic support tool for clinicians and a self assessment aid for patients.

Figures (6)

• Figure 1: Overview of the proposed model structure based on Swin-Transformer.
• Figure 2: W-MSA to SW-MSA Attention Process in the Swin-Transformer Block.
• Figure 3: BatchFormer Block Architecture.
• Figure 4: Cross-sample Gradient Propagation in BatchFormer.
• Figure 5: Class distribution in the ISIC2019 dataset.