alzheimer-classification-swin-transformer

Alzheimer Classification with Swin Transformer

Overview

This project focuses on multiclass Alzheimer MRI image classification using a custom Swin Transformer-based deep learning architecture implemented with PyTorch.

The model was trained to classify brain MRI scans into four different Alzheimer stages:

• Mild Demented
• Moderate Demented
• Non Demented
• Very Mild Demented

The project includes:

• Custom Swin Transformer implementation
• PyTorch training pipeline
• Early stopping mechanism
• Evaluation metrics and classification reports
• Confusion matrix analysis
• Visualization of training curves

---

Model Architecture

The model is based on the Swin Transformer architecture, which applies self-attention within shifted local windows to efficiently learn both local and global image representations.

Main Components

• Patch Embedding
• Window-based Multi-Head Self Attention (W-MSA)
• Shifted Window Attention (SW-MSA)
• MLP Blocks
• Layer Normalization
• Hierarchical Feature Extraction

Frameworks and Libraries

• PyTorch
• torchvision
• scikit-learn
• matplotlib
• PIL

---

Model Performance

Classification Report

Overall Metrics

• Accuracy: 96.09%
• Macro F1-Score: 96.33%
• Weighted F1-Score: 96.07%

The results demonstrate that the Swin Transformer architecture can effectively classify Alzheimer stages from MRI images with high performance.

---

Features

• Custom Swin Transformer implementation
• Transformer-based medical image classification
• Early stopping for overfitting prevention
• Detailed evaluation metrics
• GPU-supported training pipeline
• Modular PyTorch code structure

---

Dataset

The model was trained on an Alzheimer MRI image classification dataset containing four diagnostic categories.

Dataset Categories

• Mild Demented
• Moderate Demented
• Non Demented
• Very Mild Demented

---

Model Weights

You can download the pretrained model weights directly from this repository.

---

Example Inference

import torch
from PIL import Image
from torchvision import transforms

# Load model
model.load_state_dict(torch.load("alzheimer_classification_swin_transformer_model.pth"))
model.eval()

# Image transforms
transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor()
])

# Load image
image = Image.open("sample.jpg").convert("RGB")
image = transform(image).unsqueeze(0)

# Prediction
with torch.inference_mode():
    output = model(image)
    pred = torch.argmax(output, dim=1)

print(pred)

---

Project Structure

├── notebooks/
├── models/
│   └── alzheimer_classification_swin_transformer_model.pth
├── images/
├── app/
├── README.md
└── requirements.txt

---

Future Improvements

Potential future enhancements:

• Grad-CAM visualizations
• Streamlit deployment
• Larger MRI datasets
• Model quantization
• Ensemble learning
• Explainable AI techniques

---

License

This project is released under the MIT License.