Related Experiment Videos
SCC-Net: A lightweight attention-enhanced deep learning model for automated squamous cell carcinoma detection
Yi Wu1, Xianwei Li2, Muxin Zhao1
1The Second Hospital of Dalian Medical University, Dalian, China.
Digital Health
|July 6, 2026
Summary
A new AI model, SCC-Net, accurately detects cutaneous squamous cell carcinoma (cSCC) from dermoscopy images. This lightweight model shows promise for early cancer detection in clinical settings with limited resources.
Area of Science:
- Dermatology
- Artificial Intelligence
- Medical Imaging
Background:
- Cutaneous squamous cell carcinoma (cSCC) is a common skin cancer.
- Accurate and early detection is crucial for effective treatment.
- Resource-constrained settings face challenges in dermatological diagnostics.
Purpose of the Study:
- To develop and validate SCC-Net, a lightweight AI model for cSCC detection.
- To address limitations of current diagnostic tools in resource-limited environments.
- To enable early and accurate identification of cSCC from dermoscopy images.
Main Methods:
- Developed SCC-Net, a GhostNet-CBAM model, for binary classification (cSCC vs. non-cSCC).
- Trained on 5150 public dermoscopy images (628 cSCC).
- Validated using five-fold cross-validation, external testing, and Grad-CAM for interpretability.
Main Results:
- SCC-Net achieved high internal performance: 96.6% accuracy, 88.7% recall, 0.986 AUC.
- External testing demonstrated strong results: 94.3% accuracy, 93.8% recall, 0.986 AUC.
- The model significantly outperformed five CNN baselines with only 1.59M parameters.
Conclusions:
- SCC-Net provides accurate, explainable, and cost-effective cSCC detection.
- The model is suitable for teledermatology applications.
- Offers a valuable tool for improving cSCC diagnosis in underserved areas.