Abstract:
The early detection of Alzheimer’s disease (AD) is crucial for effective interventionand management. This study explores the application of deep learning techniques toclassify MRI images into three categories: mild demented, moderate demented, andnon-demented. Utilizing a comprehensive preprocessing pipeline, including datanormalization, resizing, histogram equalization, augmentation, dataset balancing, andbatch normalization, we prepared a balanced dataset comprising 800 training images
and 200 testing images per class. We implemented and evaluated four deep learningmodels: a modified Convolutional Neural Network (CNN), AlexNet, VGG16, andahybrid model integrating EfficientNet-b0 for feature extraction and Support Vector
Machine (SVM) for classification. The performance metrics, including accuracy, precision, recall, F1-score, and support, were determined for each model. Additionally, we analyzed the accuracy and loss curves, confusion matrices, ROC curves, andprecision-recall curves to comprehensively assess the models. The results demonstratethat the hybrid model combining EfficientNet-b0 and SVM outperformed the others
with an accuracy of 96.17%, followed by VGG16 with 95.17%, and both the modifiedCNN and AlexNet with 94.83%. These findings suggest the potential of advanceddeep learning architectures in enhancing the diagnostic accuracy for Alzheimer’s
disease, thereby supporting early intervention strategies.
