Abstract
A three-dimensional convolutional neural network was developed to classifyT1-weighted brain MRI scans as healthy or Alzheimer. The network comprises 3Dconvolution, pooling, batch normalization, dense ReLU layers, and a sigmoidoutput. Using stochastic noise injection and five-fold cross-validation, themodel achieved test set accuracy of 0.912 and area under the ROC curve of0.961, an improvement of approximately 0.027 over resizing alone. Sensitivityand specificity both exceeded 0.90. These results align with prior workreporting up to 0.10 gain via synthetic augmentation. The findings demonstratethe effectiveness of simple augmentation for 3D MRI classification and motivatefuture exploration of advanced augmentation methods and architectures such as3D U-Net and vision transformers.