Capturing Individual Variability in Spatial and Temporal Properties of Resting-State fMRI: A Comparison of Data-Driven Methods
- Junlin Jing,
- Pan Wang,
- Benjamin Klugah-Brown
, - Andrew M Michael,
- Bharat B Biswal
Junlin Jing
Clinical Hospital of Chengdu Brain Science Institute, School of Life Science and Technology, MOE Key Laboratory for Neuroinformation, University of Electronic Science and Technology of China
Pan Wang
Clinical Hospital of Chengdu Brain Science Institute, School of Life Science and Technology, Science and Technology of China, MOE Key Laboratory for Neuroinformation, University of Electronic, Clinical Hospital of Chengdu Brain Science Institute, School of Life Science and Technology, MOE Key Laboratory for Neuroinformation, University of Electronic Science and Technology of China
Benjamin Klugah-Brown
Clinical Hospital of Chengdu Brain Science Institute, School of Life Science and Technology, Science and Technology of China, MOE Key Laboratory for Neuroinformation, University of Electronic, Clinical Hospital of Chengdu Brain Science Institute, School of Life Science and Technology, MOE Key Laboratory for Neuroinformation, University of Electronic Science and Technology of China
Corresponding Author:[email protected]
Author ProfileAndrew M Michael
Duke Institute for Brain Sciences, Duke University, Clinical Hospital of Chengdu Brain Science Institute, School of Life Science and Technology, MOE Key Laboratory for Neuroinformation, University of Electronic Science and Technology of China
Bharat B Biswal
Department of Biomedical Engineering, New Jersey Institute of Technology, Clinical Hospital of Chengdu Brain Science Institute, School of Life Science and Technology, MOE Key Laboratory for Neuroinformation, University of Electronic Science and Technology of China
Abstract
Data-driven methods such as group independent component analysis (GICA), group information guided ICA (GIG-ICA), and independent vector analysis (IVA-GL) have been instrumental in the exploration of spatial and temporal brain functional networks. The previous comparative studies have mainly focused on the linear and nonlinear mixed variability, making it difficult to identify consistent patterns within the population. This study aimed to comprehensively investigate the extent to which linear inter-subject variability (L-ISV) is present in the spatial and temporal domains of brain functional networks and specifically examine the performance of GICA, GIG-ICA, and IVA-GL in characterizing this variability. Resting-state fMRI data was obtained from the Open Access Series of Imaging Studies (OASIS-3), which included 100 healthy controls (HC) and 53 participants with mild cognitive impairment (MCI) Differences in L-ISV between GIG-ICA and IVA-GL were found for spatial properties only in the MCI group, but not for temporal properties. A similar pattern was observed between GICA and IVA-GL in the spatial properties. The MCI group showed increased variability as evidenced by greater amplitude and widespread clusters in the spatial properties compared to HC. For the combined spatial and temporal features, the default mode network (DMN) and sensorimotor network (SMN) showed consistent performance in classifying MCI and HC. Specifically, IVA-GL achieved a classification accuracy of 75.80% for the DMN and 77.80% for the SMN, compared to 71.90% and 70.59% for GICA, respectively. Also, for DMN and SMN, GIG-ICA achieved 75.16% and 71.90%, respectively, compared to IVA-GL with 75.82% and 77.78%, respectively. IVA-GL outperformed GICA and GIG-ICA in classification, indicating its efficiency in discriminating networks with high L-ISV. In conclusion, the study revealed substantial L-ISV in the spatial properties of the MCI group. IVA-GL was more effective in determining
