Alzheimer's Stage Prediction

This project focuses on predicting Alzheimer's disease stages using deep learning models applied to brain MRI data. Specifically, I have utilised the ConvNeXt architecture on ADNI and OASIS datasets, and MedViT on the ADNI dataset.

---

Table of Contents

• Introduction
• Datasets
• Model Architectures
• Methodology
• Results
• Dependencies
• How to Run
• Acknowledgements

---

Introduction

Alzheimer's Disease (AD) is a neurodegenerative disorder characterized by progressive cognitive decline. Early and accurate stage prediction can significantly help in timely interventions. This project explores deep learning-based MRI analysis for multi-class classification of Alzheimer's stages.

---

Datasets

ADNI (Alzheimer's Disease Neuroimaging Initiative)

• Multi-class MRI dataset categorized into: Alzheimer's Disease (AD), Cognitive Impairment (CI), and Cognitively Normal (CN).
• Preprocessed axial slices were used for training.

OASIS (Open Access Series of Imaging Studies)

• Contains MRI scans categorized similarly into stages of cognitive decline.
• Used for additional training and validation with ConvNeXt.

---

Model Architectures

ConvNeXt

• Applied on both ADNI and OASIS datasets.
• Modern convolutional neural network architecture inspired by vision transformers.
• Used pretrained weights for transfer learning.

MedViT

• Applied on ADNI dataset.
• Medical Vision Transformer architecture designed specifically for medical imaging tasks.
• Used pretrained weights to leverage learned features from large-scale medical image data.

---

Methodology

1. Data Preparation:

   • Loaded and preprocessed MRI slices.
   • Resized images according to the requirement of tghe given models.

2. Model Training:

   • Fine-tuned pretrained ConvNeXt and MedViT models.
   • Hyperparameter tuning for optimal performance.
   • Training progress, losses, and accuracies logged at each epoch.

3. Evaluation:

   • Evaluated using metrics like accuracy, precision, recall, and confusion matrix.
   • Separate train/validation/test splits maintained.

---

Results

Model	Dataset	Accuracy
ConvNeXt	ADNI	99.33%
ConvNeXt	OASIS	99.45%
MedViT	ADNI	96.51%

---

Dependencies

• Python 3.x
• PyTorch
• scikit-learn
• numpy
• pandas
• matplotlib
• torchvision
• timm (for ConvNeXt pretrained models)
• MedViT package (or custom MedViT implementation)
• tqdm

---

How to Run

1. Clone this repository:

git clone <your-repo-url>

2. Install the dependencies:

pip install -r requirements.txt

3. Prepare the datasets (place ADNI and OASIS datasets in the data/ directory following the expected folder structure).

4. Train the models:

python train_convnext.py  # For ConvNeXt training
python train_medvit.py    # For MedViT training

5. Evaluate the models:

python evaluate.py

---

Acknowledgements

• ADNI: Alzheimer's Disease Neuroimaging Initiative
• OASIS: Open Access Series of Imaging Studies
• MedViT: Medical Vision Transformer authors for open-source pretrained models
• ConvNeXt: ConvNeXt authors for open-source pretrained models

---

Note: This project was conducted purely for academic and research purposes.