l1 Regularization Based Random Vector Functional Link Network for Alzheimer's Disease Diagnosis

Abstract

Alzheimer's disease (AD) is a neurological condition primarily impacting the elderly and is known for its progressive decline in normal brain functioning. The magnetic resonance imaging (MRI) modality enables disease diagnoses by identifying atrophy patterns and structural changes. Imaging captures anatomical details effectively using axial, sagittal, and coronal planes. The axial plane of MRI provides a cross-sectional view, whereas the coronal plane allows visualization in the anterior-posterior direction. The sagittal plane offers a lateral view and aids in examining asymmetry and bilateral structures of the brain's anatomy. In this paper, the features of the sagittal plane are extracted using Resnet-50, a deep-learning network. These extracted features are fed to the classifier for AD diagnosis. This paper presents a l1 regularization-based random vector functional link network (RVFL) classifier for AD diagnosis. The optimization problem of the proposed l1 regularization-based RVFL is solved using the Split Bregman iterative method. l1 regularization-based RVFL classifier is more generalizable and produces sparse output. The sparse output indicates a lower number of non-zero elements in the output compared to the standard RVFL network which uses l2 regularization. The experiments are performed on the publicly accessible Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, compared with other feed-forward networks. Results highlight the better performance of the proposed model for AD diagnosis than state-of-the-art approaches. © 2025 IEEE.