shashika05/brain-tumor-hybrid-classifier

Hybrid Brain Tumor Classifier

This model is a Hybrid CNN-Transformer architecture (ConvNeXt + Swin Transformer) trained to classify brain MRI scans into four categories:

1. Glioma
2. Meningioma
3. No Tumor
4. Pituitary

Performance

• Architecture: ConvNeXt-Tiny & Swin-Tiny Fusion
• Test Accuracy: ~93.46%
• Inference Time: ~18.77 ms

Training Details

• Trained for 20 epochs using Focal Loss to handle class imbalance.
• Optimized using AdamW with Cosine Annealing learning rate schedule.