Advancing Population Receptive Field Mapping to Ultra-High Spatial Resolution in fMRI

Abstract

Functional magnetic resonance imaging (fMRI) has become the prime method for mapping function in the human brain. It also enables the investigation of retinotopic organization in the human visual cortex using population receptive field (pRF) mapping.pRF mapping is an approach for estimating the aggregate receptive field properties of neuronal populations within individual voxels by modeling their responses to systematically varying visual stimuli. Advances in MR hardware and sequence design as well as in reconstruction and denoising methods have enabled the progression of fMRI into the sub-mm regime. Although protocols in this range are typically limited by longer scanning times, poor coverage and lower signal-to-noise ratio, the improved spatial resolution has enabled layer-specific investigations of the cortex. This laminar fMRI is a rapidly growing field, but so far only a handful of papers have been performing pRF mapping in the submillimeter range. It has previously been shown that three-dimensional echo planar imaging (3D-EPI) is in principle capable of delivering higher signal-to-noise ratio (SNR) than standard 2D-EPI at high-resolutions. There is, however, no dedicated study that explores whether this SNR advantage also translates into superior variance explained values in pRF mapping. The thermal denoising algorithm NORDIC improves SNR by singular value thresholding, was, however, found to affect pRF parameters.In this thesis, extensive testing and parameter optimisation of a novel 3D-EPI sequence has been performed at 7 Tesla in order to obtain a robust 3D-EPI protocol with 1.2 mm isotropic resolution. This protocol was compared in a group of healthy participants to a standard 2D-EPI protocol with similar temporal and spatial resolution to examine potenial benefits of the 3D acquisition approach. The resulting retinotopic maps and the obtained pRF parameters were analyzed both with and without application of NORDIC denoising in the primary (V1), secondary (V2) and tertiary (V3) visual cortex. Furthermore, an ultra-high-resolution 3D-EPI protocol at 0.7 mm isotropic resolution was implemented and compared to the 1.2 mm acquisition to assess the stability and quality of the resulting pRF estimates. Finally, the data from the ultra-high-resolution runs were upsampled and subsequently used to divide the voxels into cortical layers, allowing the investigation of depth-dependent characteristics of the pRF estimates.Here it is shown that the optimised 3D-EPI protocol delivers superior performance compared to 2D-EPI. Wilcoxon signed rank testing revealed significantly higher explainedvariance in the 3D-EPI data (V1: d=0.56, V2: d=0.44, V3: d=0.47). Furthermore, its pRF parameters were being less affected by the application of NORDIC. Artificial increase in pRF size was stronger for the 2D-EPI (V1: d=0.66, V2: d=0.45, V3: d=0.22) than for the 3D-EPI (V1: d=0.51, V2: d=0.31, V3: d=0.19) data. It can therefore be recommended to choose 3D-EPI over 2D-EPI for high-resolution pRF mapping. Utilizing this sequence, together with NORDIC denoising, allowed for the optimization of a ultrahigh resolution protocol at 0.7 mm spatial and 2 s temporal resolution with full coverage of the early visual cortex, delivering high-quality retinotopic maps and pRF estimates comparable with those at lower resolutions. The acquired data was suitable for layerspecific pRF analysis which revealed depth-dependent changes of pRF size throughout the early visual cortex. Those results help facilitate the advance of pRF mapping to the sub-mm regime, as comparable spatiotemporal resolutions with full coverage of the early visual cortex have not previously been reached.