3D-Convolutional Neural Network for Alzhimer's Disease diagnosis

Machine and deep learning methods for healthcare have became dominant, showing their effectiveness over traditional statistical techniques. However, reproducibility and reliability of model results are often limited due to bad practices or lack of detailed information on preprocessing and training phases.  Further, the scarcity of benchmarks for testing typically makes the comparison among studies impossible. This project considers the problem of discriminating between Cognitive Normal subjects and patients with Alzheimer's Disease (AD) based on 3D brain T1w MRI. The aims of this work are i) to build a reproducible pipeline from data processing to sample classification; ii) to investigate the impact of different data augmentation strategies and model depth on 3D Convolutional Neural Network (CNN) performance.