Premium
Robust spinal cord resting‐state fMRI using independent component analysis‐based nuisance regression noise reduction
Author(s) -
Hu Yong,
Jin Richu,
Li Guangsheng,
Luk Keith DK,
Wu Ed. X.
Publication year - 2018
Publication title -
journal of magnetic resonance imaging
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.563
H-Index - 160
eISSN - 1522-2586
pISSN - 1053-1807
DOI - 10.1002/jmri.26048
Subject(s) - reproducibility , intraclass correlation , noise (video) , analysis of variance , noise reduction , independent component analysis , region of interest , communication noise , regression analysis , medicine , linear regression , nuclear medicine , mathematics , statistics , artificial intelligence , radiology , computer science , linguistics , philosophy , image (mathematics)
Background Physiological noise reduction plays a critical role in spinal cord (SC) resting‐state fMRI (rsfMRI). Purpose To reduce physiological noise and increase the robustness of SC rsfMRI by using an independent component analysis (ICA)‐based nuisance regression (ICANR) method. Study Type Retrospective. Subjects Ten healthy subjects (female/male = 4/6, age = 27 ± 3 years, range 24–34 years). Field Strength/Sequence 3T/gradient‐echo echo planar imaging (EPI). Assessment We used three alternative methods (no regression [Nil], conventional region of interest [ROI]‐based noise reduction method without ICA [ROI‐based], and correction of structured noise using spatial independent component analysis [CORSICA]) to compare with the performance of ICANR. Reduction of the influence of physiological noise on the SC and the reproducibility of rsfMRI analysis after noise reduction were examined. The correlation coefficient (CC) was calculated to assess the influence of physiological noise. Reproducibility was calculated by intraclass correlation (ICC). Statistical Tests Results from different methods were compared by one‐way analysis of variance (ANOVA) with post‐hoc analysis. Results No significant difference in cerebrospinal fluid (CSF) pulsation influence or tissue motion influence were found ( P = 0.223 in CSF, P = 0.2461 in tissue motion) in the ROI‐based (CSF: 0.122 ± 0.020; tissue motion: 0.112 ± 0.015), and Nil (CSF: 0.134 ± 0.026; tissue motion: 0.124 ± 0.019). CORSICA showed a significantly stronger influence of CSF pulsation and tissue motion (CSF: 0.166 ± 0.045, P = 0.048; tissue motion: 0.160 ± 0.032, P = 0.048) than Nil. ICANR showed a significantly weaker influence of CSF pulsation and tissue motion (CSF: 0.076 ± 0.007, P = 0.0003; tissue motion: 0.081 ± 0.014, P = 0.0182) than Nil. The ICC values in the Nil, ROI‐based, CORSICA, and ICANR were 0.669, 0.645, 0.561, and 0.766, respectively. Data Conclusion ICANR more effectively reduced physiological noise from both tissue motion and CSF pulsation than three alternative methods. ICANR increases the robustness of SC rsfMRI in comparison with the other three methods. Level of Evidence: 2 Technical Efficacy: Stage 1 J. Magn. Reson. Imaging 2018;47:1421–1431.