🩺 Skin Disease Classification using CNN & ResNet18 This project focuses on the binary classification of skin images as Infected or Not Infected using deep learning models, specifically a custom CNN and a fine-tuned ResNet18 (via ResNet50). The goal is to assist in early skin disease detection using image-based AI diagnosis tools.
✅ Custom-built Convolutional Neural Network (CNN)
✅ Transfer Learning with ResNet18
✅ Evaluation using Accuracy, Precision, Recall, F1-Score
✅ Grad-CAM visualization for explainable AI
✅ Preprocessing pipeline for image normalization and resizing
✅ Designed for lightweight deployment in resource-constrained environments
- Total Images: 1,287
- Classes: Infected, Not Infected
- Format: .JPG images
- Preprocessing: Resized to 200x200, normalized to [0, 1]
This project integrates Grad-CAM to highlight which regions of the image the model focused on during classification, enhancing trust and transparency in AI-driven diagnostics.
🛠# Future Enhancements
- Multi-class skin condition classification
- Mobile deployment with TensorFlow Lite
- Integration with real-time webcam inference
- Incorporate patient metadata for hybrid decision-making
- This project is open-source and licensed under the MIT License.