Scripts are for our Computer Vision course project "Convolutional Neural Networks for Alzheimers Disease Prediction Using Brain MRI Images".
We preprocessed brain MRI scans with segmentation, registration, normalization, noise addition and skull-stripping.
We first trained a 2D-AlexNet with transfer learning on scan slices and then combined it with SVM, where the ConvNet is a feature extractor. We also trained a 3D sparse auto-encoder on small patches extracted from scans to generate features, which were then inputs to a simple 3D CNN for the prediction task.
The best accuracy we got on the validation set was 86%.