About DermaScan
Project Overview
DermaScan is an AI-powered web application designed to detect and classify skin lesions using advanced deep learning techniques. Our project aims to provide an accessible tool for preliminary skin lesion assessment, helping users identify potential concerns that may require professional medical attention.
The application uses a convolutional neural network trained on the HAM10000 dataset to analyze dermatoscopic images and provide detailed classifications across seven different types of skin lesions. This technology can help with early detection of potentially serious skin conditions, though it should always be used in conjunction with professional medical advice.
The Dataset
HAM10000 Dataset
Our model is trained on the Human Against Machine with 10000 training images (HAM10000) dataset, a large collection of dermatoscopic images of common pigmented skin lesions. This dataset includes over 10,000 high-quality images across seven different diagnostic categories:
- Actinic Keratosis (akiec)
- Basal Cell Carcinoma (bcc)
- Benign Keratosis (bkl)
- Dermatofibroma (df)
- Melanoma (mel)
- Melanocytic Nevus (nv)
- Vascular Lesion (vasc)
The dataset was collected over 20 years and contains images from different populations, captured with different devices. This diversity helps our model generalize well to various skin types and imaging conditions.
Technology Stack
Frontend
- Next.js for server-side rendering and routing
- Tailwind CSS for responsive design
- React for interactive UI components
- Shadcn/UI for accessible component library
AI Model
- EfficientNet architecture
- PyTorch for model development
- Transfer learning with ImageNet weights
- Streamlit for model deployment
The AI Model
Our model uses the EfficientNet-B3 architecture, which provides an excellent balance between accuracy and computational efficiency. The model was trained using the following approach:
- Transfer learning from ImageNet pre-trained weights
- Data augmentation to improve generalization (rotation, flipping, color jittering)
- Class balancing to handle the imbalanced nature of the HAM10000 dataset
- Fine-tuning with a learning rate scheduler for optimal convergence
The model achieves over 85% accuracy on the test set, with particularly high sensitivity for melanoma detection, which is critical for early intervention in potentially life-threatening cases.
Medical Disclaimer
DermaScan is designed as an educational and assistive tool only. It is not intended to replace professional medical diagnosis, advice, or treatment. The predictions made by this application should be verified by qualified healthcare professionals. Early detection and proper medical care are essential for skin conditions, particularly for potentially serious conditions like melanoma.