This group project explores the power of neural networks in classifying Alzheimer’s disease using MRI scans. Our team of five — Me, Timothy Ryall, Cooper Janke, Torstein Korten, and Paal Markus Bjoernstad — sought to develop a model that not only identifies the presence of Alzheimer’s but also assesses its severity.

Leveraging a dataset of 6,400 brain MRIs, we conducted extensive experiments with machine learning models, particularly focusing on convolutional neural networks. Key methods included data preprocessing, exploratory analysis, and model experimentation using both non-neural and neural network-based approaches. We also introduced our custom model, AlzheimersNet, which achieved an impressive 99% accuracy, significantly enhancing the potential for early, accurate diagnoses of Alzheimer’s.

In addition to developing a high-accuracy model, we utilized tools like Grad-CAM to visualize crucial brain regions for Alzheimer’s diagnosis. The results offer promising implications for the medical field, demonstrating how deep learning models can augment traditional diagnostic methods.

For a full version of our report, please see the complete article here

alzheimersnet-report.pdf