ASSESSMENT OF NEUROIMAGING DATA AND IDENTIFICATION OF ALZHEIMER'S DISEASE USING EXTREME LEARNING MACHINES

Abstract

Alzheimer's disease (AD), one of the most common forms of dementia, is a cognitive disorder that is progressive in nature and causes a dynamic deterioration of the mental state of an individual. It severely damages the brain cells, neurotransmitters, and nerves, leading to irreparable damage to the brain, which is one of the major causes of dementia. Early identification, assessment, and timely diagnosis are of paramount importance to slow down the progression of the disease, which calls for the design and development of algorithms and technology-aided tools for accurate detection, diagnosis, and prediction of the severity of Alzheimer's disease. To provide a solution to this, we propose an extreme learning machine (ELM) algorithm that is trained on neuroimaging data from longitudinal MRI scans obtained from the OASIS database. We adopt an extensive feature engineering pipeline to choose the most significant features for early identification of the onset of dementia. We obtain an overall accuracy of 98.3%, sensitivity of 0.956, specificity of 0.962, and F1 score of 0.972. We also show that our proposed ELM algorithm outperforms several other contemporary classifiers based on a range of evaluation metrics. The paper also provides a feasibility analysis of the proposed model for real-time clinical deployment.