Parametrically regularised population receptive field (pRF) mapping
Poster No:
2556
Submission Type:
Abstract Submission
Authors:
Michael Woletz1, David Linhardt2, Siddharth Mittal1, Christian Windischberger3
Institutions:
1Medical University of Vienna, Vienna, Vienna, 2Medical University of Vienna, Wien, Vienna, 3Medical University of Vienna, Vienna, Austria
First Author:
Co-Author(s):
Introduction:
Population receptive field (pRF) mapping (Dumoulin, 2008), is a method for mapping the visual cortex in terms of input from the visual field. By sampling the visual field using a stimulus, the BOLD response in the visual cortex can be used to assign each voxel a receptive field, typically modelled as a Gaussian with three parameters (µxx, µyand σ). In a standard analysis, this will be done for each voxel individually. The retinotopic organization of the visual cortex will in this case lead to neighbouring voxels having similar parameters, with gradual changes throughout the visual cortex. This topological organisation is not typically enforced, so large variations between neighbouring voxels may exist due the presence of measurement noise. While the topological organisation might be enforced during the fitting process (Tu, 2021), this enforcement is only possible on surface data and might strongly influence the results. In this work we propose a simpler solution, by adding a regularisation factor to the parameter estimation, penalising differences between the parameters between neighbouring voxels/vertices.
Methods:
The proposed model optimises the sum of the explained variances of all voxels/vertices. We added a regularisation component that penalises the squared difference of the pRF parameters of neighbouring voxels/vertices. This regularisation is additionally weighted by the explained variance of each neighbour. In this way, the parameters of a voxel/vertex might be adjusted with respect to its neighbours, especially if the neighbours' pRF models fit the data well.
The proposed model was implemented in the Python programming language and applied to the measured data of three subjects (3T, Siemens Prisma Fit) using a moving bar stimulus while acquiring EPI images (1s/38ms TR/TE, 1.5 mm isotropic, multiband 3). Algebraic properties of the model fitting process were used to enable the computation of the error function's gradient and the L-BFGS-B optimiser as implemented in SciPy was used to fit the model with and without regularisation to the data.
The proposed model was implemented in the Python programming language and applied to the measured data of three subjects (3T, Siemens Prisma Fit) using a moving bar stimulus while acquiring EPI images (1s/38ms TR/TE, 1.5 mm isotropic, multiband 3). Algebraic properties of the model fitting process were used to enable the computation of the error function's gradient and the L-BFGS-B optimiser as implemented in SciPy was used to fit the model with and without regularisation to the data.
Results:
Eccentricity and polar angle maps of the three subjects in V1 of the left hemisphere are shown in Figure 1. The regularisation led to smoother maps and fewer outliers compared to the results without regularisation.
Conclusions:
In this work, we presented a new method of adding a regularisation to pRF mapping. This method allows to better preserve the topological structure of the visual cortex and is able to reduce otherwise spurious outliers in the map. Further research is necessary to ensure that the regularised maps are consistent across runs and how the regularisation parameters should be best adjusted.
Modeling and Analysis Methods:
Activation (eg. BOLD task-fMRI)
Methods Development 2
Novel Imaging Acquisition Methods:
BOLD fMRI
Perception, Attention and Motor Behavior:
Perception: Visual 1
Keywords:
FUNCTIONAL MRI
Modeling
Vision
1|2Indicates the priority used for review
Provide references using author date format
Tu, Y. (2021), 'Topological Receptive Field Model for Human Retinotopic Mapping'. Med Image Comput Comput Assist Interv. pp. 639-649