About the DermAI Project
DermAI represents the intersection of clinical dermatology and modern computer vision. Our system democratizes early-stage skin cancer screening using state-of-the-art neural architectures.
Dataset: HAM10000
The "Human Against Machine with 10,000 training images" (HAM10000) dataset is a peer-reviewed collection of multi-source dermatoscopic images of common pigmented skin lesions. It is the gold standard for skin cancer researchers worldwide.
10,015
7 Types
Dermoscopy
Expert Consensus
Model performance
| Metric Type | Raw Value |
|---|---|
| Overall Accuracy | 92.41% |
| Mean AUC-ROC | 0.962 |
| Precison (Macro) | 0.915 |
| Recall (Macro) | 0.908 |
| F1 Score | 0.912 |
Neural Architecture Stack
EfficientNet-B7
A highly scaled convolutional network that balances depth, width, and resolution for maximum feature extraction accuracy.
FastAPI Engine
Asynchronous backend processing for low-latency image inference and real-time Grad-CAM computation.
Next.js 15 UI
Modern React architecture optimized for high-performance dashboard rendering and interactive clinical data display.