Optimizing CNN-Based Alzheimer's Detection Through Augmentation Techniques in Biomedical Image Processing

Alzheimer's Disease (AD) is a progressive neurodegenerative disorder characterized by cognitive decline and memory impairment, posing significant challenges to global healthcare. Early and accurate diagnosis is essential for effective management and intervention. Convolutional Neural Networks (CNNs) have shown promise in automating the detection and classification of AD from neuroimaging data like Magnetic Resonance Images. However, the performance of these models is often hindered by limited and imbalanced datasets. This study investigates the effect of various data augmentation techniques like Cutout, Mixup, Rotation with Flip, and a combination of blurring and sharpening applied on biomedical images like MRI for three model architectures: Baseline CNN, Depthwise Separable CNN, and Pretrained ResNet152v2. Experimental results demonstrate that augmentation significantly enhances model robustness and generalization. The Pretrained ResNet152v2 model achieved the highest accuracy of 86.5% using blurring and sharpening, while the Depthwise separable CNN balanced performance and computational efficiency with a maximum accuracy of 74.2%. This study underscores the importance of tailored data augmentation strategies in improving the performance of deep learning models for AD detection, offering valuable insights for future research and clinical applications.