Improved FastICA algorithm in fMRI data analysis using the sparsity property of the sources
Ruiyang Ge1, Yubao Wang2, Jipeng Zhang3
1State Key Laboratory of Cognitive Neuroscience and Learning & IDG/McGovern Institute for Brain Research, Beijing Normal University, Beijing 100875, China; College of Information Science and Technology, Beijing Normal University, Beijing 100875, China; Non-Invasive Neurostimulation Therapies (NINET) Laboratory, Department of Psychiatry, Faculty of Medicine, University of British Columbia, Vancouver, BC V6T 2A1, Canada.
SparseFastICA enhances independent component analysis (ICA) for fMRI data by incorporating source sparsity. This novel method improves robustness and spatial detection compared to standard FastICA, offering faster computation than Infomax for accurate brain network identification.
Area of Science:
- Neuroimaging
- Signal Processing
- Computational Neuroscience
Background:
- Independent Component Analysis (ICA) is crucial for blind source separation in functional magnetic resonance imaging (fMRI).
- Existing constrained and semi-blind ICA methods utilize temporal or spatial priors, but seldom incorporate source sparsity.
- Sparsity is an underutilized prior for improving ICA performance in fMRI data analysis.
Purpose of the Study:
- To introduce a novel SparseFastICA method for enhanced fMRI data analysis.
- To integrate source sparsity as a constraint within the FastICA algorithm.
- To evaluate the feasibility, robustness, and performance of SparseFastICA against existing methods.
Main Methods:
- Proposed SparseFastICA by adding source sparsity constraint to the FastICA algorithm.
- Estimated source sparsity using a smoothed ℓ0 norm method.
- Conducted experimental tests on simulated and real fMRI data, comparing SparseFastICA, FastICA, and Infomax ICA.
Main Results:
- SparseFastICA demonstrated feasibility and robustness for source separation in both simulated and real fMRI data.
- SparseFastICA exhibited superior robustness to noise and enhanced spatial detection power compared to FastICA.
- SparseFastICA achieved comparable spatial detection power to Infomax ICA but with a faster computation speed.
Conclusions:
- SparseFastICA offers a significant improvement over FastICA in robustness and spatial detection power for fMRI data.
- The method enables more accurate identification of brain networks compared to the standard FastICA algorithm.
- SparseFastICA presents a computationally efficient alternative comparable to Infomax for fMRI analysis.
