A Lightweight and Accurate Deep Convolutional Neural Network Method based on MRI for the Diagnosis and Classification of Alzheimer's Disease

The Alzheimer's disease, a progressive brain disorder, necessitates timely detection for effective management due to the current diagnostic methods' limitations. The study presents an efficient convolutional neural network (CNN) designed for classifying brain magnetic resonance imaging (MRI) images into four categories related to Alzheimer's disease. To enhance diagnosis, this study proposes two distinct approaches at different stages: (1) using optimized data pre-processing; and (2) designing a lightweight CNN architecture with low complexity and fewer parameters that simultaneously possesses good accuracy, computational efficiency, and excellent performance. The proposed method achieved outstanding results, with a final accuracy of 99.22%, a macro average F1 score of 0.99, an MCC of 0.9870, and a Cohen kappa score (CKS) of 0.9870. In addition to accuracy, the complexity of the proposed model, including the comparison of model size, time elapsed, the number of FLOPs, and the trainable and non-trainable parameters of the proposed method were also thoroughly investigated. This model, with the advantages of high accuracy, reduced FLOPs, faster execution time, and lower memory requirements outperforms other deep learning methods used in recent studies.