Introduction: Alzheimer’s Disease (AD) is a progressive, incurable ailment of the Central Nervous System (CNS) that leads to the deterioration of cognitive functions, including memory, reasoning, and behavior. Methods: In this study, a Convolutional Neural Network (CNN) architecture, named ALZNetwork, is proposed for diagnosing Alzheimer’s Disease (AD). The model’s robustness lies in its training on a single dataset and testing on two different Magnetic Resonance Imaging (MRI) datasets, both consisting of 2D slices from Alzheimer’s Disease Neuroimaging Initiative (ADNI) sources, achieving impressive results. There were two main objectives of this research: (1) to create a novel 2D image dataset, and (2) to construct a robust convolutional framework on this dataset for AD diagnosis. Results: Comparisons were made between the ALZ-Network model and its pre-trained counterparts, i.e., Xception, InceptionV3, and EfficientNetB0. Model evaluation was done on a 3-class classification problem using standard metrics, i.e., accuracy, positive predictive value, sensitivity, and the harmonic mean of the latter two. InceptionV3 and Xception models achieved reduced training time with classification accuracies of 91.69% and 97.68%, respectively, whereas EfficientNetB0 and ALZ-Network achieved higher classification accuracies of 99.32% and 99.70%, respectively. Discussion: A web platform, i.e., Alzheimer’s Disease Detection System (ADDS), built on the proposed architecture and other models, was created (run on a local machine) to test unseen 2D images of AD. Conclusion: This platform can assist radiologists in screening for potential Alzheimer’s disease from MRI scans of subjects.
Sheikh et al. (Tue,) studied this question.