---
license: gpl-3.0
datasets:
- Kaynaaf/Brain-Tumour-MRI
metrics:
- accuracy 0.90
- precision 0.90
library_name: keras
tags:
- medical
- healthcare
---

# Model Card 

An Image Classifier that predicts the presence of certain Brain tumours from their MRI scans

## Model Details
A 134M Parameter ConvNet designed for classification of Brain tumours in MRI scans.

## Paper

Interpretable Deep Learning for Brain Tumor Diagnosis: Occlusion Sensitivity-Driven Explainability in MRI Classification
DOI: [10.21015/vtse.v13i2.2082](10.21015/vtse.v13i2.2082)

## Uses


### Direct Use

Load the model, finetune the model if needed or just go straight towards generating inferences using the model.



### Downstream Use

Finetune the model on other diagnostic scans, though the model only accepts grayscale images of size 256x256.

## How to Get Started with the Model

[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1SfK9d2In3JHDvyXH4jpwznVGEG_wXRuQ?usp=sharing)

## Training
The colab notebook used to train the model can be found below

[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1SfK9d2In3JHDvyXH4jpwznVGEG_wXRuQ?usp=sharing)

## Evaluation

### Metrics

| Class       | Precision | Recall | F1-Score | Support |
|-------------|-----------|--------|----------|---------|
| Glioma      | 0.96      | 0.87   | 0.91     | 300     |
| Meningioma  | 0.84      | 0.71   | 0.77     | 306     |
| No Tumor    | 0.88      | 1.00   | 0.93     | 405     |
| Pituitary   | 0.93      | 0.99   | 0.96     | 300     |
| **Accuracy**|           |        | **0.90** | 1311    |
| **Macro Avg** | 0.90    | 0.89   | 0.89     | 1311    |
| **Weighted Avg** | 0.90 | 0.90   | 0.90     | 1311    |

![Roc curve](https://raw.githubusercontent.com/Kaynaaf/BrainMRI-Classifier/main/results/roc_curve.png)
### Results
This model was developed for my project that can be found on github [here](https://github.com/Kaynaaf/BrainMRI-Classifier)
. This project involved generating sensitivity maps to explain the predictions of the model.
These maps assign values to areas of the image that act as feature importance markers.
![pituitary Tumour sensitivity map](https://raw.githubusercontent.com/Kaynaaf/BrainMRI-Classifier/main/results/pituitary_1.png)
![Glioma  sensitivity map](https://raw.githubusercontent.com/Kaynaaf/BrainMRI-Classifier/main/results/glioma_1.png)