Journal of Information Communication Technologies and RoboticApplications:A Novel Architecture Optimization Using DeepLearning for Alzheimer’s Multiclass Disease Detection

Abstract

Neurodegenerative Disorders (NDD) directly affect the nervous system, causing the gradual loss of neurons and cognitive decline. Alzheimer’s disease (AD) and its precursor, Mild Cognitive Impairment (MCI), affect a large proportion of the world’s population. Deep learning models, such as Convolutional Neural Networks (CNNs), are primarily employed for AD detection using MRI-based biomarkers. However, their performance is constrained by challenges such as overfitting, class similarity, and limited labeled data. The number of convolutional layers, the number of filters per layer, and combinations of hyperparameters also affect the model’s performance. Trying every possible combination, including critical hyperparameters, to obtain the best model requires substantial training time and effort. This study explores the performance of different CNN architectures on the Alzheimer's Kaggle dataset. It is a multiclass dataset with different stages of Alzheimer’s disease, including Mild-Demented, Moderate-Demented, Non-Demented, and Very-Mild Demented. A thorough performance-based analysis was carried out to determine the optimal CNN architecture. The findings offer important insights into refining and optimizing deep learning algorithms for accurate Alzheimer’s disease detection. Various combinations of parameters, including learning rate, batch size, dropout rate, and optimizer, have been tuned to mitigate overfitting and improve generalization. The impact of all optimization techniques is analysed to select the best model among eleven CNN versions. The proposed approach achieves 90.6% accuracy, 96% specificity, and an F1-score of 90%, demonstrating a balanced and therapeutically appropriate performance quality.