Brain PET Dementia Classification using Inception-ResNet-V2
Overview
This project utilizes Inception-ResNet-V2 as the base architecture for a deep learning model trained on the OASIS-1 dataset. The model is designed to classify brain PET scans into four categories:
- Non-Demented
- Very Mild Dementia
- Mild Dementia
- Moderate Dementia
With extensive training and fine-tuning, the model has achieved an accuracy of over 99%.
Dataset: OASIS-1
The OASIS-1 (Open Access Series of Imaging Studies) dataset contains brain PET scans with labels indicating different levels of dementia. The dataset is publicly available for research and includes:
- Cross-sectional brain PET scans
- Manually assigned labels for dementia severity
- Preprocessed images with skull stripping and intensity normalization
For more information about the dataset, visit: OASIS Dataset
Model Architecture
The model is based on Inception-ResNet-V2, a powerful deep convolutional neural network (CNN) architecture that combines the advantages of Inception and ResNet modules for highly efficient feature extraction.
Key Features:
- Pretrained on ImageNet for weight initialization.
- Custom classification head added for dementia severity prediction.
- Fine-tuned on OASIS-1 PET scans
- Achieved 99%+ accuracy on validation and test sets.
Installation
Ensure you have all dependencies installed by running:
pip install -r requirements.txt
Performance
The model was trained using transfer learning and fine-tuning on the OASIS-1 dataset, achieving:
- 99%+ accuracy on validation and test sets.
- High precision & recall across all dementia categories.
Contributors
- (@Twast0)
License
This project is licensed under the MIT License. See LICENSE for details.