MCAD-DCMGC: Multi-class Classification of Alzheimers Disease using DCGAN Combined with MobileNetV2 and Grad-CAM

Alzheimer is an advancing neurodegenerative disorder with different clinical stages, beginning with cognitively normal (CN) to mild cognitive impairment (MCI), further leading to Alzheimer’s dementia (AD).By using neuroimaging datasets, early and accurate classification of all these stages are crucial for timely intervention and medical treatment planning. In this paper, we pro- pose MCAD- DCMGC, a novel deep learning-based framework which uses MobileNetV2 for multiclass classification of AD stages from structural MRI slices. To overcome the problem of class imbalance and to enhance model generalizability, we augmented the dataset by using Deep Convolutional Generative Adversarial Networks (DCGANs), generating synthetic MRI samples for underrepresented classes. Furthermore, to increase the clinical interpretability, we are using Gradient-weighted Class Activation Mapping (Grad-CAM) for clear vision of different regions contributing to the multiclass classification decision. The experimental results with proposed light weight model achieved a multiclass classification accuracy of 92.62%, accompanied by insightful visual explanations. This pro- posed framework has potential to support clinical decision-making by improving both the accuracy and transparency of automated Alzheimer’s stage classification.