Ke Zhou1, Zhou Liu2, Wenguang He1, Jie Cai3, Lingjing Hu4. 1. School of Biomedical Engineering, Guangdong Medical University, Zhanjiang, 524023, China. 2. Department of Neurology, Affiliated Hospital of Guangdong Medical University, Zhanjiang, 524023, China. 3. School of Biomedical Engineering, Guangdong Medical University, Zhanjiang, 524023, China. caijie2013@gdmu.edu.cn. 4. Department of Medical Imaging Technology, Capital Medical University Yanjing Medical College, Beijing, 101300, China. hulj@ccmu.edu.cn.
Abstract
BACKGROUND: Patients with mild cognitive impairment (MCI) are a high-risk group for Alzheimer's disease (AD). Thus, a reliable prediction of the conversion from MCI to AD based on three-dimensional (3D) texture features of MRI images could help doctors in developing effective treatment protocols. METHODS: The 3D texture features of the whole-brain were deduced based on the gray-level co-occurrence matrix. Then, the embedded feature selection method based on least squares loss and within-class scatter (LSWCS) was employed to select the optimal subsets of features that were used for binary classification (AD, MCI_C, MCI_S, normal control in pairs) based on SVM. A tenfold cross validation was repeated ten times for each classification. LASSO, fused_LASSO, and group LASSO are used in feature selection step for comparison. RESULTS: The accuracy and the selected features are the focus of clinical diagnosis reports, indicating that the feature selection algorithm is effective.
BACKGROUND: Patients with mild cognitive impairment (MCI) are a high-risk group for Alzheimer's disease (AD). Thus, a reliable prediction of the conversion from MCI to AD based on three-dimensional (3D) texture features of MRI images could help doctors in developing effective treatment protocols. METHODS: The 3D texture features of the whole-brain were deduced based on the gray-level co-occurrence matrix. Then, the embedded feature selection method based on least squares loss and within-class scatter (LSWCS) was employed to select the optimal subsets of features that were used for binary classification (AD, MCI_C, MCI_S, normal control in pairs) based on SVM. A tenfold cross validation was repeated ten times for each classification. LASSO, fused_LASSO, and group LASSO are used in feature selection step for comparison. RESULTS: The accuracy and the selected features are the focus of clinical diagnosis reports, indicating that the feature selection algorithm is effective.
Authors: L Harrison; P Dastidar; H Eskola; R Järvenpää; H Pertovaara; T Luukkaala; P-L Kellokumpu-Lehtinen; S Soimakallio Journal: Comput Biol Med Date: 2008-03-14 Impact factor: 4.589
Authors: M Bobinski; J Wegiel; H M Wisniewski; M Tarnawski; M Bobinski; B Reisberg; M J De Leon; D C Miller Journal: Neurobiol Aging Date: 1996 Nov-Dec Impact factor: 4.673