BOLD signals have been extensively studied to show their variability across subjects and even within the same subject across runs¹ and trials². While it has been reported that BOLD signals are reliable within individuals, spatial distributions of relative latency and temporal precision within small cortical regions have yet to be quantitatively described.

Here we used 3D inverse imaging (InI) method³ to characterize spatial distributions of relative latency and variability of the BOLD signal at human visual cortex with 0.1 s precision.

Subjects (n = 9) participated this study with written informed consents approved by the IRB of National Taiwan University Hospital. Events of transient wedge-shaped checkerboard flashing (full-contrast; 500 ms duration; 8 Hz checkerboard reversal) were presented randomly in the left (36 events), right (36 events), or both hemifields (108 events). InI data were collected on a 3T scanner (Skyra, Siemens, Erlangen, Germany) with a coronal projection acquisition (frequency encoding along superior-inferior direction) using a 32-channel head coil array. Imaging parameters were: TR =0.1 s, TE=30 ms, flip angle =30o. bandwidth = 2520 Hz/pixel. Four runs of functional data (304 s per run) and 3D T1-weighted structural images (MPRAGE sequence) were collected for each subject for visualization and localizing anatomical landmarks (FreeSurfer).

Functional data were first volumetrically reconstructed using the minimum-norm estimate³. Hemodynamic response functions (HRFs) for each hemifield stimulation were estimated by the General Linear Model (GLM) using finite impulse response basis functions. To estimate the variability of BOLD responses, we partitioned trials of the same stimulus randomly into two groups, each of which had one estimated HRF. This random partitioning was repeated 30 times, allowing for 60 HRF estimates from two partitioning groups and across 30 iterations.

The single-subject relative latency and temporal variability of the BOLD signal were calculated using a correlation analysis: first a template visual cortex HRF was created (averaging over V1 and V2 of the visual cortex ROI). Cross-correlation between the template visual cortex HRF and HRFs at each location in the visual cortex was calculated by temporally shifting the visual cortex template HRF (+4 and −4 seconds with 0.1 steps). Relative latency was defined as the shift corresponding to the highest correlation coefficient. BOLD variability was taken as the standard deviation of relative latency over the 60 HRF estimates from two partitioning groups and 30 iterations.

Significant BOLD signals (t-statistics) were found at the contralateral occipital lobe of the left and right hemispheres. Average and standard deviation of the relative latency in the left and right visual cortex were 0.09 (s) +/− 0.15 (s) and 0.14 (s) +/− 0.42 (s), respectively. The earliest BOLD response (with the most negative relative latency) was found along the fundus; longer latencies were observed further away from the fundus. The most delayed BOLD responses (about +1 s) were found at the crest of the calcarine sulcus and along the V1/V2 boundary. The variability of the BOLD response was mostly below 1 s in V1. The least variability (< 0.2 s) was found at the fundus of the calcarine sulcus. More variable BOLD responses were found at the crest of the calcarine sulcus. Average and standard deviation of the BOLD variability in the left and right visual cortex were 0.39 (s) +/− 0.17 (s) and 0.39 (s) +/− 0.30 (s). Regression analysis shows that the relative latency (x in seconds) is linearly related to BOLD variability (y in seconds): y = 0.29 + 0.71 x (r²=0.39; p< 0.001).We delineated spatial distributions of the relative latency and the variability of the BOLD responses at human visual cortex with high temporal precision. While being considered as the most reliable measure, single-subject BOLD signals can still vary a fraction of a second over repetitions. This variability characterization can be used as the prior information in Bayesian fMRI analysis⁴ when the HRF variability is considered. In this study, the relative latency and variability of the BOLD signal were evaluated based on individual’s visual cortex HRF, because previous studies have reported that the HRF from the same functional area differs significantly across subjects¹. The linear relationship between BOLD relative latency and variability and associated maps suggest that I) HRF latency and variability depend on the polar angle in the visual field, and II) the retinotopic location of the horizontal meridian has the least HRF variability and the fastest BOLD onset. HRF characteristics have been previously shown to vary systematically across V1⁵ and potentially reflect a relationship between retinotopic and vascular organization.