Background: Functional magnetic resonance imaging (fMRI) is widely used to investigate dynamic brain functions in neurological and psychological issues; however, high noise level limits its applicability for intensive and sophisticated investigations in the field of neuroscience. New method: To deal with both issue (low sensitivity and dynamic signal), we used ensemble empirical mode decomposition (EEMD), an adaptive data-driven analysis method for nonstationary and nonlinear features, to filter task-irrelevant noise from raw fMRI signals. Using both simulations and representative fMRI data, we optimized the analytic parameters and identified non-meaningful intrinsic mode functions (IMFs) to remove noise. Results: We revealed the following advantages of EEMD in fMRI analysis: (1) EEMD achieved high detectability for task engagement; (2) the functional sensitivity was markedly enhanced by removing task-irrelevant artifacts based on EEMD. Comparison with existing method(s): Compared with other noise-removal methods (e.g., band-pass filtering and independent component analysis), the EEMD-based artifact-removal method exhibited better spatial specificity and superior Gaussianity of the resulting t-score distribution. Conclusions: We found that EEMD method was efficient to enhance the functional sensitivity of evoked fMRI. The same strategy would be applicable to resting-state fMRI signal in the general purpose.
- Ensemble empirical mode decomposition (EEMD)
- FMRI sensitivity
- Hilbert-Huang transform
- Resting-state fMRI
ASJC Scopus subject areas