Large Margin and Local Structure Preservation Sparse Representation Classifier for Alzheimer's Magnetic Resonance Imaging Classification
Author(s): Liu, RM (Liu, Runmin); Li, GJ (Li, Guangjun); Gao, M (Gao, Ming); Cai, WW (Cai, Weiwei); Ning, X (Ning, Xin)
Source: FRONTIERS IN AGING NEUROSCIENCE Volume: 14 Article Number: 916020 DOI: 10.3389/fnagi.2022.916020 Published: MAY 25 2022
Abstract: Alzheimer's disease (AD) is a progressive dementia in which the brain shrinks as the disease progresses. The use of machine learning and brain magnetic resonance imaging (MRI) for the early diagnosis of AD has a high probability of clinical value and social significance. Sparse representation classifier (SRC) is widely used in MRI image classification. However, the traditional SRC only considers the reconstruction error and classification error of the dictionary, and does not consider the global and local structural information between images, which results in unsatisfactory classification performance. Therefore, a large margin and local structure preservation sparse representation classifier (LMLS-SRC) is developed in this manuscript. The LMLS-SRC algorithm uses the classification large margin term based on the representation coefficient, which results in compactness between representation coefficients of the same class and a large margin between representation coefficients of different classes. The LMLS-SRC algorithm uses local structure preservation term to inherit the manifold structure of the original data. In addition, the LMLS-SRC algorithm imposes the l(2,1)-norm on the representation coefficients to enhance the sparsity and robustness of the model. Experiments on the KAGGLE Alzheimer's dataset show that the LMLS-SRC algorithm can effectively diagnose non AD, moderate AD, mild AD, and very mild AD.
Accession Number: WOS:000807986900001
PubMed ID: 35693338
Author Identifiers:
Author Web of Science ResearcherID ORCID Number
Cai, Weiwei AAH-5456-2020 0000-0001-6795-6152
ISSN: 1663-4365
Full Text: https://www.frontiersin.org/articles/10.3389/fnagi.2022.916020/full