How ytashan/brain-tumor-mri-image-classification-hybrid-model works
This notebook implements a brain-MRI classification workflow: it loads and stratifies image data, prepares an input pipeline, trains a six-branch hybrid model, and evaluates it alongside individual-backbone and soft-voting comparisons. It also documents Grad-CAM, model persistence, and single-image or batch inference. The README supplies the workflow and capabilities, but no notebook source excerpts were available, so internal call details beyond those documented stages are not inferred.
Main entry points: brain-tumor-hybrid-model.ipynb Two-phase training
Components 22
Data preparation
- Load and split
brain-tumor-hybrid-model.ipynb - Explore data
brain-tumor-hybrid-model.ipynb - tf.data pipeline
brain-tumor-hybrid-model.ipynb - Data augmentation
brain-tumor-hybrid-model.ipynb
Model construction
- Custom CNN
brain-tumor-hybrid-model.ipynb - EfficientNetB3
brain-tumor-hybrid-model.ipynb - VGG16
brain-tumor-hybrid-model.ipynb - ResNet50
brain-tumor-hybrid-model.ipynb - DenseNet121
brain-tumor-hybrid-model.ipynb - MedicalNet-inspired CNN
brain-tumor-hybrid-model.ipynb - Hybrid feature fusion
brain-tumor-hybrid-model.ipynb
Training and evaluation
- Class weights and callbacks
brain-tumor-hybrid-model.ipynb - Two-phase training
brain-tumor-hybrid-model.ipynb - Test evaluation
brain-tumor-hybrid-model.ipynb - Backbone baselines
brain-tumor-hybrid-model.ipynb - Soft-voting ensemble
brain-tumor-hybrid-model.ipynb
Interpretation and inference
- Grad-CAM
brain-tumor-hybrid-model.ipynb - Save and load model
brain-tumor-hybrid-model.ipynb - Image and batch inference
brain-tumor-hybrid-model.ipynb - Tumor class predictions
brain-tumor-hybrid-model.ipynb
Other components
- MRI dataset
- User
Connections 21
- MRI dataset → Load and split: provides images
- Load and split → Explore data: explores data
- Load and split → tf.data pipeline: creates splits
- tf.data pipeline → Data augmentation: augments images
- tf.data pipeline → Two-phase training: supplies batches
- Custom CNN → Hybrid feature fusion: extracts features
- EfficientNetB3 → Hybrid feature fusion: extracts features
- VGG16 → Hybrid feature fusion: extracts features
- ResNet50 → Hybrid feature fusion: extracts features
- DenseNet121 → Hybrid feature fusion: extracts features
- MedicalNet-inspired CNN → Hybrid feature fusion: extracts features
- Class weights and callbacks → Two-phase training: configures training
- Hybrid feature fusion → Two-phase training: trains model
- Two-phase training → Test evaluation: evaluates model
- Two-phase training → Backbone baselines: compares backbones
- Backbone baselines → Soft-voting ensemble: combines predictions
- Two-phase training → Grad-CAM: explains predictions
- Two-phase training → Save and load model: saves model
- Save and load model → Image and batch inference: loads model
- User → Image and batch inference: submits images
- Image and batch inference → Tumor class predictions: predicts class