| license: mit | |
| tags: | |
| - medical | |
| - image-classification | |
| - tensorflowjs | |
| - keras | |
| - brain-tumor | |
| - edge-inference | |
| # BrainScan // DenseNet Brain Tumor Classification Model (TF.js format) | |
| This repository contains the weights and topology of the **BrainScan** classification model converted to TensorFlow.js layers format for in-browser client-side execution. | |
| ## Model Files | |
| * `model.json`: Patched TensorFlow.js Functional Model topology (Keras 3 compatible). | |
| * `group1-shard*of7.bin`: Binary weights shards. | |
| * `model/brain_tumor_model.h5`: The raw, pre-converted Keras H5 model. | |
| * `model/BrainScan_Model_Training.ipynb`: Jupyter notebook containing the DenseNet model definition and training history. | |
| ## Classes | |
| The model classifies T1/T2 brain MRI scans into four distinct categories: | |
| 1. `glioma` | |
| 2. `meningioma` | |
| 3. `notumor` (Normal) | |
| 4. `pituitary` | |
| ## Usage in TensorFlow.js | |
| ```javascript | |
| const model = await tf.loadLayersModel('https://huggingface.co/grvsnh/brainscan-tfjs/resolve/main/model.json'); | |
| const prediction = model.predict(imageTensor); | |
| ``` | |