Back

DermAssist: A Hybrid Vision Transformer System for Skin Lesion Diagnosis with Automated Alerting and Dual-Sided Portals

nasir, s.

2025-07-16 epidemiology
10.1101/2025.07.15.25331570 medRxiv
Show abstract

Skin cancer, one of the most prevalent forms of cancer globally, demands early and accurate diagnosis to improve patient outcomes. In this paper, we present DermAssist, a hybrid deep learning-based dermatology assistant system that integrates Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) for automated multi-class skin lesion classification. Our model was trained and evaluated on widely-used datasets including DermNet, ISIC, and HAM10000, employing a robust preprocessing pipeline to enhance lesion visibility and diversity. DermAssist combines segmentation (U-Net), feature extraction (ResNet), and classification (BEiT Transformer), and introduces a dual-portal architecture for clinicians and patients using Streamlit and Flask interfaces. We incorporate Grad-CAM and SHAP for interpretability, Twilio-based SMS alerting for high-risk cases (>90% confidence), and secure AWS S3 storage to ensure HIPAA compliance. Experimental results demonstrate an accuracy of 90.2% with strong ROC-AUC and precision scores. DermAssist is positioned as a deployable, intelligent diagnostic aid capable of enhancing dermatological workflows in real-time environments.

Matching journals

The top 6 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.