Coil Sensitivity-Guided Denoising of Non-Brain-Origin Fluctuations in fMRI
Yul-Wan Sung1, Masaki Fukunaga2, Uk-Su Choi2
1From the Kansei Fukushi Research Institute (Y-W.S., S.O.), Tohoku Fukushi University, Kunimigaoka 6-149-1, Aoba, Sendai, Miyagi 9893201, Japan; Section of Brain Function Information (M.F., Y.K.), National Institute of Physiological Sciences, Myodaiji,Okazaki 444-8585,Japan; Medical Device Development Center (U-S.C.), Daegu-Gyeongbuk Medical Innovation Foundation, Cheombok-ro 80, Dong-gu, Daegu 41061, S. Korea; Department of Radiology (D.K.), Mayo Clinic, 200 First St. SW, Rochester, MN, 55905 and Neuroscience Research Institute (K.-N.K.), Gachon University, Namdong-daero, Namdong-gu, Incheon 24 774, S.Korea. sung@tfu-mail.tfu.ac.jp sungstone@gmail.com.
None:
Noise suppression is essential for improving the reliability of functional MRI (fMRI), particularly at ultra-high field strengths where system- and receiver-related fluctuations become prominent. We introduce a hardware-informed denoising method, Coil Sensitivity Filter (CsFilter), which exploits spatial sensitivity relationships among receiver coil elements to identify and suppress non-brain-origin signal components. CsFilter operates in the frequency domain by retaining temporal fluctuations whose coil-amplitude ordering is consistent with receiver sensitivity profiles. Using 7T task-based fMRI data acquired with a 32-channel head coil, CsFilter produced robust increases in voxel-wise temporal signal-to-noise ratio relative to conventional high-pass filtering, with more than 62% of voxels showing greater than twofold improvement. Task-based analyses showed increased statistical strength, with mean t-values rising from 4.68 to 5.74 and median t-values from 4.39 to 5.28. These results indicate that coil-sensitivity-guided filtering provides an effective and complementary strategy for suppressing additive non-brain-origin fluctuations in multi-channel fMRI data.
