A Resilient, Non-neuronal Source of the Spatiotemporal Lag Structure Detected by BOLD Signal-Based Blood Flow Tracking

Abstract

Recent evidence has suggested that blood oxygenation level-dependent (BOLD) signals convey information about brain circulation via low frequency oscillation of systemic origin (sLFO) that travels through the vascular structure (“lag mapping”). Prompted by its promising application in both physiology and pathology, we examined this signal component using multiple approaches. A total of 30 healthy volunteers were recruited to perform two reproducibility experiments at 3 Tesla using multiband echo planar imaging. The first experiment investigated the effect of denoising and the second was designed to study the effect of subject behavior on lag mapping. The lag map's intersession test-retest reproducibility and image contrast were both diminished by removal of either the neuronal or the non-neuronal (e.g., cardiac, respiratory) components by independent component analysis-based denoising, suggesting that the neurovascular coupling also comprises a part of the BOLD lag structure. The lag maps were, at the same time, robust against local perfusion increases due to visuomotor task and global changes in perfusion induced by breath-holding at the same level as the intrasession reliability. The lag structure was preserved after time-locked averaging to the visuomotor task and breath-holding events, while any preceding signal changes were canceled out for the visuomotor task, consistent with the passive effect of neurovascular coupling in the venous side of the vasculature. These findings support the current assumption that lag mapping primarily reflects vascular structure despite the presence of sLFO perturbation of neuronal or non-neuronal origin and, thus, emphasize the vascular origin of the lag map, encouraging application of BOLD-based blood flow tracking.

Axial variation of deoxyhemoglobin density as a source of the low-frequency time lag structure in blood oxygenation level-dependent signals

“…Finally, the lag structures in the S 0 (or non-599 BOLD) component did not correlate with that from T2*, either spatially or temporally. We 600 also found a vascular region-dependent change in the T2* sLFO, with a decreased 601 amplitude in the outlet part close to major veins, in contrast to the S 0 response that 602 remained constant; this finding replicates a previous observation in the raw BOLD signal 603 (Aso et al, 2017a). The S 0 component exhibited a unique brain region-dependent response 604 to the respiratory phase, suggesting that certain perfusion parameters specifically contribute 605 to this component, but not the perfusion lag.…”

“…The similarity between the low-90 frequency phase map and perfusion MRI in healthy participants was a milestone in this 91 direction, as it presented the perfusion time lag embedded in the BOLD signal (Tong et al, 92 2017). The resilient nature of the lag map against the fMRI task condition was shown, 93 further supporting its non-neuronal origin (Aso et al, 2017a). In parallel, a number of 94 clinical studies have established the phase delays as a marker of cerebrovascular disorders 95 (Amemiya et al, 2013;Lv et al, 2013;Christen et al, 2015;Ni et al, 2017;Nishida et al, 96 2018; Khalil et al, 2018).…”

“…In the present study, the lag structure was treated as a broadly distributed, structured 705 noise for fMRI. Indeed, it can be partly eliminated by sophisticated denoising techniques 706 (Aso et al, 2017a). However, the specificity of lag mapping in isolating information on a 707 purely vascular origin remains unclear.…”

“…For example, measurements of velocity changes 708 critically depend on a recursive lag-tracking method that incorporates the gradual change in 709 LFO over regions (Tong & Frederick, 2014). Adaptation for changes in the waveform that 710 may arise from different paths of the blood was demonstrated to increase the 711 reproducibility of the lag map (Aso et al, 2017a). However, as the changes in waveform 712 can also reflect NVC, removing the whole lag structure may lead to type II errors in the 713 fMRI results.…”

“…The same pulse sequence 222 program was used in Experiment 2 but with multi-echo settings: TE1 = 7.76 ms and TE2 = 223 25.82 ms for the first 6 participants and TE1 = 11.2 ms and TE2 = 32.78 ms for the 224 remaining 15 participants. A smaller multiband factor of 2 was selected to allow for the 225 short TE in combination with parallel imaging using GeneRalized Autocalibrating Partial 226 information (Aso et al, 2017a). A dual-echo gradient-echo dataset for B0 field mapping 234 was also acquired after the BOLD scan in the same orientation.…”