An adaptive RV measure based fuzzy weighting subspace clustering (ARV-FWSC) for fMRI data analysis. (September 2015)

• Record Type: Journal Article
• Authors: Tang, Xiaoyan; Zeng, Weiming; Wang, Nizhuan; Yang, Jiajun
• Is Part Of: Biomedical signal processing and control. Volume 22(2015)
• Page Start: 146
• Page End: 154
• Publication Date: 2015-09
• Subjects: Fuzzy clustering analysis -- Fuzzy weighting subspace clustering -- RV measure -- Generated cube; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2015.07.006
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: We propose an adaptive RV measure based fuzzy weighting subspace clustering for fMRI data analysis. We apply a threshold to adaptively change the shape of the generated cube. We use a simple and effective denoising method. We adopt the RV coefficient to measure the distance between two multivariate sets. We modify the distance measure approach in objective function and updating formulas. Abstract: Conventional clustering methods for analyzing the fMRI data usually meet some difficulties, such as the huge samples, the slow processing speed, and serious noise effect. In this study, a novel adaptive RV measure based fuzzy weighting subspace clustering (ARV-FWSC) is proposed for fMRI data analysis. In this approach, the adaptive RV measure, different from the traditional distance measure like Euclidean distance or Pearson correlation coefficient, is applied to the clustering process, where the distance measure between two single voxels is converted into the adaptive RV measure between two sets of multi-voxels contained in the correspondingly generated cubes, whose shape is automatically updated by setting a threshold of the weighted template. Meanwhile, a simple denoising mechanism is also used to find noise points, whose datum generated cube only having one center voxel, and can directly exclude those noise voxels from the cluster. Furthermore, a modified fuzzy weighting subspace clustering is introduced to measure the importance of each dimension to a particular cluster, where the proposed algorithm could take the influence of different time points in each clustering process into account, besides having the advantage of ordinary fuzzy clustering like FCM (fuzzy c-means). Several evaluation metrics, e.g., coverage degree, ROC curve, and the number of clustering iteration, are adopted to assess the performance of the ARV-FWSC on real fMRI data compared with those of GLM (general liner model), ICA (independent component analysis), and FCM. Extensive experiment results show that the proposed ARV-FWSC for fMRI data analysis can effectively improve the clustering speed and raise the clustering accuracy.