A Multimodal Approach for Alzheimer's Disease Detection and Classification Using Deep Learning

Abstract

Alzheimer's disease (AD) presents a substantial challenge to healthcare systems globally due to its progressive nature and the absence of effective treatments. The timely identification of AD is crucial for enabling interventions aimed at potentially slowing cognitive decline and enhancing patient outcomes. Recent advancements in medical imaging, notably PET and MRI, provide valuable insights into the subtle changes associated with AD pathology. This study investigates the utilization of the VGG16 deep learning model, recognized for its proficiency in image recognition tasks, to extract detailed features from MRI and PET scans. By leveraging the capabilities of deep learning, our objective is to reveal subtle patterns indicative of AD pathology. These extracted features are consolidated into a unified representation, which facilitates the training of machine learning classifiers. Employing various classifiers, such as Random Forest, Support Vector Machine, and K-Nearest Neighbors, we aim to exploit their strengths in managing complex data. The experimental outcomes demonstrate the effectiveness of this hybrid approach, with the Support Vector Machine emerging as the most successful classifier, achieving an accuracy of 84%. These findings underscore the potential of deep learning-assisted feature extraction and emphasize the significance of integrating advanced imaging techniques with sophisticated machine learning algorithms for improved AD detection and classification. Such initiatives hold promise for advancing our comprehension of AD pathology, enhancing diagnostic precision, and ultimately contributing to more effective management strategies for this debilitating neurological disorder.