Brain Tumor Detection using CNN

Manual inspection of MRI scans is slow, expensive, and depends heavily on specialist availability. NeuroScan demonstrates how a trained CNN can screen scans quickly while remaining interpretable — a point that impresses examiners and makes viva answers easy. The project ships with a working upload-and-diagnose Flask web app: drop in an MRI image, get the predicted class with a confidence score plus the Grad-CAM overlay showing the tumor region the model focused on.

1. MRI scans are preprocessed — resized to a fixed input size, normalized, and augmented (rotation, flip, zoom) to improve generalization.
2. A convolutional neural network (custom CNN, with a VGG16 transfer-learning variant) is trained on a labeled public MRI dataset to classify scans as tumor or non-tumor.
3. At inference time, the trained model predicts the class of an uploaded scan along with a confidence score.
4. Grad-CAM back-propagates the class score to the last convolutional layer, generating a heatmap of the regions that most influenced the decision — the visual explanation an examiner asks about.
5. The NeuroScan web app presents the original scan, the Grad-CAM overlay, the predicted class, and the confidence score in a clean results dashboard.

• CNN classifier for brain tumor detection on MRI scans (tumor vs. non-tumor)
• Transfer-learning option (VGG16 backbone) plus a custom CNN baseline for comparison
• Grad-CAM heatmaps that visually explain every prediction — the key differentiator
• Upload-and-diagnose NeuroScan web app: drag in a scan, get class + confidence + overlay
• Model training pipeline with data augmentation, normalization, and checkpointing
• Accuracy / precision / recall / F1 metrics and confusion-matrix evaluation
• Side-by-side original vs. heatmap view for presentations and report figures
• Clean, documented, runnable codebase with a requirements file

• Full source code (training scripts, Grad-CAM module, Flask app)
• Trained model weights for instant demo without retraining
• Complete project report PDF (with architecture diagrams and results)
• Project PPT for your final presentation
• Viva Q&A document covering CNN, transfer learning, Grad-CAM, and medical-imaging questions
• Dataset setup guide with public dataset links
• Installation and run instructions

• This is an academic demonstration project, NOT a clinical diagnostic tool — it is not validated for medical use and must never be used for real diagnosis.
• Performance depends on the public training dataset; accuracy on unseen scanners or protocols may differ.
• Binary classification (tumor vs. non-tumor) is implemented; multi-class tumor-type classification is a listed extension.
• Training a deep CNN benefits from a GPU; CPU-only machines can run the demo with the provided trained weights.