A Novel Method for Multi-subject fMRI Data Analysis: Independent Component Analysis with Clustering Embedded (ICA-CE)

• Published in: 2023 45th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC) Vol. 2023; pp. 1 - 5
• Main Authors: Du, Yuhui, Zhu, Wenchao, Zhang, Yuduo
• Format: Conference Proceeding Journal Article
• Language: English
• Published: United States IEEE 01.01.2023
• Subjects: Biology; Brain - diagnostic imaging; Cluster Analysis; Connectome; Data analysis; Data mining; Estimation; Functional magnetic resonance imaging; Humans; Independent component analysis; Magnetic Resonance Imaging - methods; Task analysis
• ISSN: 2694-0604
• DOI: 10.1109/EMBC40787.2023.10339989

The analysis of multi-subject functional magnetic resonance imaging (fMRI) data and the extraction of accurate brain functional networks (FNs) are of great importance. However, traditional independent component analysis (ICA) methods perform analysis on multi-subject fMRI data under the condition of known or assumed classes of subjects, which may decrease its ability to extract accurate individual brain FNs. Although a previous method named clusterwise ICA (C-ICA) clusters subjects and obtains shared FNs in group-level for each class, its clustering performance on complex data is not ideal. To address the issues, we propose a novel method called independent component analysis with clustering embedded (ICA-CE) that can achieve both the estimation of individual FNs and the clustering of subjects in an unsupervised or semi-supervised manner. Using the simulated data with different properties, ICA-CE achieved better clustering performance than group ICA followed by K-means and C-ICA, and the mean accuracy of extracted individual FNs obtained by ICA-CE was greater than 90%. Using the task-related fMRI data from Human Connectome Project (HCP), our method also achieved higher clustering accuracy, while extracting task-related class-specific FNs. In summary, ICA-CE is effective in estimating accurate brain FNs while achieving the clustering of multiple subjects.Clinical Relevance- Our method is promising in estimating accurate brain functional networks for patients with brain disorders and outputting related class label for each subject.