Skin disease ML(senior project)
Developed a deep learning-based web application for skin disease classification using medical image datasets. Built the frontend and backend, integrated trained machine learning models, and implemented image upload with real-time prediction.
Screenshots
A visual tour of the application
Features
Key features and functionalities
AI-powered skin disease classification
Lesion segmentation using U-Net
Deep learning prediction with EfficientNet-B0 Ensemble
Image upload with real-time diagnosis prediction
Architecture
System architecture and design
Challenge
The main problem I solved
Building an accurate skin disease classification model from medical image datasets while addressing image variability, class imbalance, and overfitting to achieve reliable prediction performance.
Solution
How I addressed the challenge
Developed a deep learning pipeline using U-Net for lesion segmentation and an EfficientNet-B0 ensemble model for classification. Trained and evaluated the models on the ISIC 2018 and HAM10000 datasets, then integrated the trained model into a web application for real-time image prediction.
Lessons Learned
Key takeaways from this project
Gained hands-on experience in computer vision, deep learning, medical image preprocessing, model training and evaluation, and deploying AI models within a full-stack web application.