Empirical Evaluation of Deep Learning Architectures in the Early Detection of Alzheimer's Disease through MRI Data Analysis

Abstract

Alzheimer's disease (AD) poses a significant challenge to healthcare systems worldwide, necessitating early detection and intervention for effective management. In this paper, various Deep Learning (DL) architectures such as CNN, Residual Neural Networks (ResNet), and U-Net with Gating (UGNet), and VGG16 for detecting AD at an early stage by applying MRI data which is acquired from the Open Access Series of Imaging Studies (OASIS) dataset.  In this method, MRI data is firstly preprocessed for extracting the relevant features which are used for training these models. Paramters of the evaluation metrics such as Classification accuracy, precision, specificity, and area under the receiver operating characteristic curve (AUC-ROC) are applied for assessing the performance of four DL models while AD patients. From the simulation results, it has been observed that high classification accuracy and AUC-ROC scores has been achieved in VGG16 as compared with other models. However, ResNet performs better when it comes on classifying complex image tasks in detecting AD. When convolutional and deconvolutional pathways with gating mechanisms are considered, UG-Net performs a notable performance in comparing to CNN and ResNet. These findings have supported the potentiality of DL techniques in detecting early diagnosis and intervention strategies while contributing to much-improved patient care.