INVESTIGATING THE EFFICIENCY OF ALZHEIMERS DISEASE CLASSIFICATION BY RADIOMIC FEATURES & DEEP LEARNING TECHNIQUES

Abstract

Alzheimer’s Disease (AD) poses a significant global health challenge, underscoring the need for advanced tools facilitating early and precise diagnosis. Previous studies have em- ployed various traditional machine learning techniques, transitioning from image decom- position methods such as principal component analysis to more sophisticated non-linear decomposition algorithms. The advent of computer vision and neural networks has further propelled advancements in the biomedical field. Given the absence of a definitive cure for Alzheimer’s in the medical industry and the significance of formulating effective treatment strategies through early diagnosis, the relevance of early detection is paramount. This study centers on Alzheimer’s disease diagnosis through MRI data analysis, leveraging radiomic features. It introduces a novel 3D deep learning model integrated with attention modules to tackle the vanishing gradient problem. Additionally, it outlines the development of a machine learning (ML) model for classification using radiomic features extracted from NIFTI files. The project introduces Alz3Dnet, a network that integrates attention modules and addresses vanishing gradient challenges, achieving a validation accuracy of 92.22%. Furthermore, an ML ensemble model utilizing radiomic features attains an accuracy of 96%