Marginal Spectrum Modulated Hilbert-Huang Transform: Application to Time Courses Extracted by Independent Vector Analysis of Resting-State fMRI Data

Hilbert-Huang transform (HHT) can reveal abnormal activations impacted by mental disorders from regions of interest (ROIs) based functional magnetic resonance imaging (fMRI) data with high temporal and frequency resolutions. However, this advantage has not been extended to the time courses extracted by data-driven methods such as independent vector analysis (IVA) from fMRI data. This study explores HHT to analyze IVA separated time courses and improves HHT via multiplying the HHT spectrum with the marginal HHT spectrum (named as marginal spectrum modulated HHT) to enhance the difference between patients and controls. We evaluated the proposed HHT using resting-state fMRI data collected from patients with schizophrenia and healthy controls, and compared with time series of ROIs defined according to Brodmann areas. Experimental results showed that the proposed HHT improved inter-group differences compared to HHT, in terms of the clustering precision, e.g., 7.7% higher (81.2% vs. 73.5%) for the temporal lobe, providing new evidence for potential imaging biomarkers for schizophrenia in the time-frequency domain.