---
tags:
- image-classification
- tensorflow
- medical
- mri
- brain-tumor
datasets:
- the-shoaib2/Brain_Tumor_MRI
library_name: tensorflow
pipeline_tag: image-classification
license: mit
---

# MRI Brain Tumor Classification Model V3

This model is a fine-tuned Xception network for classifying brain MRI scans into 4 categories:
`glioma`, `meningioma`, `notumor`, `pituitary`.

## Model Details
- **Architecture**: Xception (Pre-trained on ImageNet)
- **Input Size**: (299, 299)
- **Framework**: TensorFlow / Keras
- **Classes**: 4 (glioma, meningioma, notumor, pituitary)

## Training Configuration
- **Epochs**: 10
- **Batch Size**: 32
- **Learning Rate**: 0.001
- **Optimizer**: Adamax
- **Loss**: Categorical Crossentropy
- **Metrics**: Accuracy, Precision, Recall

## Performance Metrics

```text

Final Training Metrics (Epoch 10):
- Training Accuracy: 0.9979
- Training Loss: 0.0076
- Validation Accuracy: 0.9786
- Validation Loss: 0.1827
- Precision: 0.9979
- Recall: 0.9977

```

## Training History

### Metrics Plot
![Training History](training_history.png)

### Final Metrics
![Training Metrics Final](training_metrics_final.png)

## Confusion Matrix
![Confusion Matrix](conf_matrix.png)

## Data Distribution
![Training Data Distribution](count_train.png)
![Test Data Distribution](count_test.png)

## Sample Predictions
![Data Samples](data_samples.png)

## Usage

```python
import tensorflow as tf
from huggingface_hub import from_pretrained_keras
import numpy as np
from PIL import Image

# Load model
model = from_pretrained_keras("the-shoaib2/Brain_Tumor_MRI_Classification")

# Load and preprocess image
img = Image.open("path/to/mri_scan.jpg")
img = img.resize((299, 299))
img_array = np.array(img)
img_array = np.expand_dims(img_array, axis=0)
img_array = img_array / 255.0

# Predict
predictions = model.predict(img_array)
class_names = ['glioma', 'meningioma', 'notumor', 'pituitary']
predicted_class = class_names[np.argmax(predictions[0])]
confidence = np.max(predictions[0])

print(f"Predicted: {predicted_class} ({confidence:.2%} confidence)")
```

## Model Architecture

The model uses transfer learning with Xception as the base:
- Xception base (pre-trained on ImageNet)
- Global Max Pooling
- Flatten layer
- Dropout (0.3)
- Dense layer (128 units, ReLU activation)
- Dropout (0.25)
- Output layer (4 units, Softmax activation)

## Dataset

This model was trained on the [Brain Tumor MRI Dataset](https://huggingface.co/datasets/the-shoaib2/Brain_Tumor_MRI).

## Citation

If you use this model, please cite:

```bibtex
@misc{brain_tumor_mri_v3,
  author = {Shoaib},
  title = {MRI Brain Tumor Classification Model V3},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/the-shoaib2/Brain_Tumor_MRI_Classification}}
}
```