Deconvolving the laminar gradient echo activation profiles with the spatial PSF: an approach to revealing underlying activation patterns
Irati Markuerkiaga1 and David G. Norris1

1Donders Institute, Nijmegen, Netherlands

Synopsis

The specificity of GE-BOLD profiles is suspected to be degraded by intracortical veins. In this work experimentally obtained GE-BOLD profiles for different subjects are deconvolved with a laminar point spread functions obtained from a model of cortical vasculature. The obtained underlying activation profiles are closer to the activity profiles expected from electrophysiology for the type of stimulus used.

Purpose

Intracortical veins are suspected to degrade the laminar specificity of the Gradient Echo blood oxygenation level dependent (GE-BOLD) signal as they propagate the BOLD signal from lower layers to upper layers on their way to the pial veins. To overcome this lack of spatial specificity, the laminar point spread function (PSF) of the GE-BOLD signal could be used to deconvolve the measured profiles or in a forward model to obtain the underlying activation pattern across the cortex. In this work, experimental BOLD profiles at various echoes are deconvolved with the corresponding laminar PSFs calculated from a vascular model to obtain the underlying activation profiles.

Methods

Five subjects were scanned at a 7T Siemens scanner after informed consent was given. The functional scan used was a mutliecho 3D-FLASH sequence, with a 0.75 mm isotropic voxel size and 91s volume TR. The paradigm consisted of ten black screens interleaved with nine stimulus blocks of a flickering checkerboard at 7.5 Hz (see Koopmans et al 2011 for a thorough description). The average BOLD profiles across subjects are shown in Figure 1. The laminar PSFs were obtained in the following way: a vascular model of the primary visual cortex was developed following mainly histological observation in primates (Markuerkiaga et al 2014) and BOLD profiles could be predicted across the cortex making use of the BOLD signal model and relaxation parameters as proposed by Uludag et al 2009. This way, the laminar point spread function was calculated at 7 T for the echo times used in the experimental acquisition (Figure 2). The PSF is skewed towards the surface, it presents a peak in the activated layer and a rather flat plateau, which is helpful for the following deconvolution step. The acquired BOLD profiles, Y, can ideally be obtained as the multiplication of a matrix consisting of the laminar PSFs, X, and the laminar activation strength across the cortex, β: Y=X*β Hence, the activation patterns, β, is given by calculating X-1 *Y. In order to obtain more robust results against noise, a smoothness constraint was added to the transition of the activation strength between adjacent cortical depths. As the vascular model does not consider pial veins, which have an effect on the BOLD signal in the most superficial layer both by partial voluming and long ranging extravascular relaxation, BOLD signal in layer I (250 um below the pial surface), was excluded for the calculation of the activation through the cortex.

Results

The underlying activation profiles across echoes obtained are largely consistent for each of the subjects (Figure 3). In general, they show a stronger activation around the middle of the cortex compared to the lower and upper layers. The differences in the relative magnitude of the activation between subjects may be due to physiological differences between them.

Discussion

Intense and non-demanding visual stimuli are expected to generate a stronger activation in the granular layer, as it receives the input from the LGN(Bissig et al 2009). Due to the carry over effect through intracortical veins, the magnitude of the laminar GE-BOLD does not reflect this effect (Figure 1). The deconvolution of GE-BOLD profiles with the calculated laminar PSFs yields more realistic underlying activation patterns. For a given subject, the underlying activation is constant across echoes, and the activation profiles obtained are, as expected, consistent through the different echoes in all subjects. To conclude, the results shown here indicate that the use of prior knowledge about the BOLD PSF can improve the spatial resolution of GE-BOLD signal, while retaining its high sensitivity. Future work will be aimed at validating this approach in an independent set of BOLD profiles obtained under a different stimulus condition and/or field strengths.

Acknowledgements

This work was supported by the Initial Training Network, HiMR, funded by the FP7 Marie Curie Actions of the European Commission (FP7-PEOPLE-2012-ITN-316716)

References

1.Koopmans,et.al,Neuroimage(2011),56:1276-85;

2.Markuerkiaga,et.al,Proc.Intl.Soc.Mag.Reson.Med(2014);

3.Uludag,et.al, Neuroimage(2009),48:150-65;

4.Bissig,et.al, Neuroimage(2009),44:627-35

Figures

Figure 1: Experimentally obtained BOLD profiles for different echo times, averaged across five subjects.

Figure 2: Laminar BOLD point spread functions for layers VI, V, IV and II/III (B0 = 7T, TE=28ms).

Figure 3: Underlying activation profiles obtained for the different echoes for each of the subjects, (a) to (e), and the grand average profile across subjects and echo times (f). Each profile is normalized to the value obtained in the WM/GM boundary in order to compare across echoes and subjects.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
0637