Confronting the Challenges of Alzheimer’s Diagnosis: A Deep Dive into MRI-Based Early Detection Methods

Abstract

Alzheimer’s Disease (AD) is a brain disorder that causes progressive cognitive decline in older adults. It slowly destroys memory and thinking skills and eventually the ability to carry out the simplest task. Alzheimer’s patients experience changes in behavior and personality. There is currently no cure for AD. Although few medicines are approved, those work best for the early and middle stage patients of AD. As the aging population is growing, the occurrence of AD is steadily increasing. This is showing the urgent necessity for early diagnosis and intervention. Additionally, identifying symptoms early enables timely therapeutic intervention, aiding in symptom management and enhancing their quality of life. AD is notorious for gradually eroding memory and cognitive abilities, underscoring the pivotal yet challenging nature of early detection and management. In this paper, our aim is to detect the early signs of AD by harnessing the power of machine learning and deep learning to filter through Magnetic Resonance Imaging (MRI) images for detecting the early signs of AD. We utilized the Open Access Series of Imaging Studies (OASIS) dataset, sourced from Kaggle, comprising MRI brain images, to train and validate our developed model. We employed various models, including Random Forest and Logistic Regression from shallow learning techniques, as well as Extra Trees from ensemble learning. Additionally, we leveraged Convolutional Neural Networks (CNN), a deep learning model, in our analysis. While it’s apparent that Deep Learning, particularly CNNs, may yield superior performance, we also explore whether comparable results can be achieved through shallow or ensemble learning approaches. We conducted evaluations using different metrics, including common ones such as accuracy, precision, recall, and AUC for all models. Additionally, we specifically calculated training and validation accuracy and loss for the CNN model. Extra Trees and CNN demonstrated nearly comparable performance, with CNN achieving the highest Accuracy and AUC scores, while Extra Trees excelled in precision and recall metrics.