Investigation of Machine Learning methods for Alzheimer and Mental Health

Abstract

Alzheimer's disease and its precursor, Mild Cognitive Impairment, are among the leading causes of dementia worldwide. Early detection of these conditions is crucial for effective intervention and management. This study explores the use of deep learning models, specifically Convolutional Neural Networks, for the classification of MRI scans into different stages of Alzheimer's disease, namely, MildDemented, ModerateDemented, NonDemented, and VeryMildDemented. The goal of this study was to test the influence of different settings, including the use of different optimizer techniques (Adam, SGD, ADAMW) and training methods (early stopping, dropout), on the model's classification accuracy for these categories. This study comprises training CNNs with different architectures, including the dense layers and dropout approach. It prevents overfitting and improves generalization. The dataset contained MRI scans, which were augmented to provide a rich training set that enhances the model's ability to learn complex features of the brain images. The ADAM optimizer proved the best for overall classification accuracy, particularly when combined with early stopping and dropout to balance training efficiency and model generalization. A total of 11 versions of the model were tested, and among them, Version 10 (trained for 50 epochs using the ADAM optimizer, 2 dense layers, and 0.5 dropout) achieved the highest accuracy of 90% with a specificity of 0.92, demonstrating the best performance. This model shows strong potential for real-world diagnostic applications in early-stage Alzheimer’s detection.