Brain Tumor Classification using VGG16 (Grayscale MRI)
This repository contains a VGG16 transfer learning model trained on enhanced grayscale MRI images for automated brain tumor classification.
🧠 Tumor Classes
- Glioma
- Meningioma
- Pituitary
📊 Model Performance
- Test Accuracy: 93.52%
- Framework: PyTorch
- Architecture: VGG16 (Transfer Learning)
- Pre-trained on: ImageNet
- Input Size: 224×224 RGB (grayscale replicated to 3 channels)
- Number of Classes: 3
🌑 Grayscale Strategy
MRI images were processed in grayscale format and enhanced using CLAHE contrast enhancement to preserve structural and intensity-based features critical for medical diagnosis.
This approach demonstrated superior performance compared to colorized representations.
🏆 Best Model Checkpoint
The uploaded file represents the best-performing checkpoint, saved at the highest validation accuracy during training.
🔬 Training Highlights
- Transfer learning with frozen convolution layers
- Fine-tuned classifier head
- CLAHE-based preprocessing
- Data augmentation
- Stratified train/validation/test split (70/15/15)
- Early stopping and learning rate scheduling
⚠️ Disclaimer
This model is intended strictly for research and educational purposes and must not be used for clinical diagnosis or treatment planning.
👤 Author
Prashant Parwani