Deep Learning Image Classifier for Medical Diagnostics
Overview
A convolutional neural network system for automated detection of skin lesions in dermatological images
Why this matters?
Project Overview
This project develops a robust CNN architecture for automated skin lesion classification, achieving 94.2% accuracy on the HAM10000 dataset. The system incorporates advanced data augmentation techniques and transfer learning from pre-trained ResNet models.
Key Features
- Multi-class classification (7 types of skin lesions)
- Real-time inference capabilities
- Integration with clinical workflow systems
- Explainable AI visualizations using GradCAM
Technical Stack
- Framework: PyTorch, FastAPI
- Architecture: Modified ResNet-50 with custom classification head
- Deployment: Docker containers on AWS ECS
- Monitoring: MLflow for experiment tracking
Gallery
Content and Media Library
Browse through your visual content, images, and media files. This gallery provides an organized view of all your visual assets with easy navigation and preview capabilities.
