Axial deoxy-hemoglobin variation as a source of BOLD lag structure: evaluation on HCP-lifespan data

2620 

Abstract Submission 

Toshihiko Aso¹, Takuya Hayashi²

¹RIKEN Center for Biosystems Dynamics Research, Kobe, Hyogo, ²RIKEN Center for Biocystems Dynamics Research, Kobe, Hyogo

Toshihiko Aso  
RIKEN Center for Biosystems Dynamics Research
Kobe, Hyogo

Takuya Hayashi  
RIKEN Center for Biocystems Dynamics Research
Kobe, Hyogo

The longstanding controversy regarding the global signal regression (GSR) in resting-state fMRI stems from the lack of an integrated model. Despite the literature revealing cardiorespiratory components, they are not separable without special settings (Birn 2006; Chang 2009). In addition, the global component is not spatially uniform; GS can be decomposed into a set of spontaneous low-frequency oscillation (sLFO) with varying lag across voxels (Fig. 1A-B). sLFO was originally found as a systemic fluctuation common to NIRS and fMRI signals (Tong 2011), but the lag tracking allows estimation of sLFO from the fMRI data (Fig. 1C) with the "lag maps" reflecting vascular anatomy. Removal of this whole structure is a "tailored" GSR reducing more variance (-40%) than GS (Erdoğan 2016; Aso 2019). Since the mechanism of BOLD sLFO remains unsettled (Tong 2019), especially in the non-brain tissues, we proposed an axial deoxy-hemoglobin variation model (Aso 2019) which explains the absence of lag structure in CBV-based fMRI (Aso & Hayashi 2020). Indeed, physiological SaO2 or recently found hematocrit fluctuation can directly affect the tissue susceptibility:

M = TE ･ B₀ ･ CBV₀ ･ [deoxy-Hb]^β (Hoge 1999)

This model states equivalence of the sLFO phase with moving blood, which is yet to be proven. Here we take advantage of the quality and sample size of HCP-lifespan data to elucidate several issues: 1) Are respiration volume per unit time (RVT) and heart rate variation (HR) the sources of correlation between GS amplitude and lag map robustness? (Aso & Hayashi 2023), 2) Is sLFO the source of correlation between GS and RVT/HR?, and 3) Is the age-related change of lag map (Aso 2020) reproducible in the high quality data?

The 22.5 minutes of rsfMRI data from 863 subjects (HCP-A:714; D:149, >12 years) were processed. Only 582 had good physiological traces and were used for the main analysis, but the aging analysis involved all subjects. The perfusion lag mapping was done after minimally brain-masked preprocessing and smoothing at 4 mm FWHM, with 7.2-s tracking range both up- and downstream from the GS phase (https://github.com/RIKEN-BCIL/BOLDLagMapping). RVT was calculated using Power's toolbox (https://www.jonathanpower.net) after de-spiking the original sweeps. HR was obtained by finding the pulse rate for every peak and linearly interpolating the value (Fig. 1D).

Lag mapping was successful (>70% of voxels) in most subjects, especially those with high GS amplitude (Fig. 1E). GS amplitude strongly followed RVT amplitude compared to HR, while both were significant (Fig. 1F). The known temporal relationships between RVT/HR and fMRI signals were replicated (Fig. 2A-B) but, interestingly, sLFO was more correlated than GS indicating the direct relationship. Furthermore, the best fitted sLFO waveform with HR/RVT was in the arterial side of the vasculature, in agreement with the model that sLFO exists in the inflow (Fig. 2C). Finally, the voxel-wise regression revealed both up- and downward shift of timing in the deep and superficial venous territory, roughly replicating previous works (Fig. 2D-E). This change was not accounted for by atrophy (Fig. 2F)

The known relationship of GS with RVT/HR were found to be due to sLFO; it runs through the cerebral vasculature to create a major part of the GS. The result also suggests the sLFO estimated from the fMRI data, without peripheral measurement, is valid. The higher correlation of RVT/HR with sLFO from the arterial side of the vasculature is also a new finding supporting an intrinsic signal source of fluctuating deoxy-Hb that is correlated with, but cannot be fully accounted for by RVT or HR. This model allows a simple interpretation of the age-related change confirmed here, a lifelong alteration of the venous drainage pattern starting in adolescence. Such a perfusion component would systematically affect the functional connectivity analysis and therefore needs to be treated.

Exploratory Modeling and Artifact Removal ²

Neurophysiology of Imaging Signals ¹

Aging

Blood

Cerebral Blood Flow

Data analysis

fMRI CONTRAST MECHANISMS

MRI