Population receptive field and connective field models were generated using VISTASOFT (GitHub,
https://github.com/vistalab/vistasoft), according to Haak et al.
9 and Gravel et al.
10 After the population receptive fields were modeled, eccentricity and polar angle maps were derived from them and used to delineate the primary visual cortex (V1), V2d, V2v, V3d, V3v, human V4 (hV4), and object-associated (lateral–occipital, LO) and motion-associated (temporal–occipital, TO) areas, as described by de Best et al.
11 The number of voxels representing the cortical surface area was compared between groups, with no significant differences being found between cohorts. For connective field modeling, the activity of a voxel in the target area (e.g., V2) was predicted by blood oxygen-level dependent (BOLD) activity in the cortical surface of the source area (e.g., V1), termed V1 to V2 connective fields, as described in detail in Haak et al.
9
Briefly, the preprocessed BOLD responses were first high-pass filtered by removing the linear trend from the data. The reason for choosing only high-pass filtering for this dataset was to avoid filtering out the stimulus-associated signal along with the noise. Specifically, filtering from 0.01 to 0.1 Hz would mean that the parts of the signal that happen outside this time range would be filtered out. The target area BOLD response was then predicted by folding a circular Gaussian connective field model, defined by position (
v0) and size (σ), over the cortical surface of the source region of interest and calculating the weighted sum of voxel responses. For each voxel, these parameters were adjusted until the error between the model-predicted and the observed fMRI time-series was minimized. The connective field size was corrected for ipsilateral visual representations and was termed sampling extent, as in Haak et al.
9
Similarly, these connective field models were also applied to resting-state fMRI scans; however, these scans were bandpass filtered to include frequencies between 0.01 and 0.1 Hz, rather than just high-pass filtered, in order to clean out both baseline drift and high-frequency physiological noise, as described by Gravel et al.
10 Classical resting-state data cleaning procedures usually involve more steps, such as independent component analysis-based denoising of the white matter and cerebrospinal fluid in order to reveal small functional differences among areas that could be otherwise hidden. In connective field modeling, however, frequencies that are shared among white matter, cerebrospinal fluid, and cortex might contribute to the connective field measurements; therefore, we did not discard them.
For both stimulation and resting state-based connective fields, we assessed V1 to V2/V2v/V2d/V3/V3v/V3d/hV4/LO/TO (i.e., V1 to region
x) and V2/V2v/V2d/V3/V3v/V3d/hV4/LO/TO to V1 connective fields (i.e., region
x to V1). In these models, voxels with ≥20% variance explained were further analyzed. Connective field sizes were searched by the model in the following descriptive steps: 0.0001 mm, 0.2 mm, 0.4 mm, 0.6 mm, and so on until 10 mm. When 0.0001 mm is chosen (fulfills the role of 0), that means that all other higher values explained less variance and neighboring voxels did not contribute to the model; that is, neighboring voxels did not have any shared response that could be detected (combining them into a surface). This indicates that there was no “connective field area,” only a one-on-one connection; thus, connective fields with sizes smaller than or equal to 0.0001 mm were excluded. Median sampling extents, weighted by variance explained, were calculated over eccentricities 0.5° to 7.5°. The median connective field size was calculated as such because connective field size was reported to be constant across eccentricities within an area.
9
The differences between the sampling extents of V1 to region x and region x to V1 were calculated to indicate the convergence magnitude. Here, higher values indicate a larger spatial extent, whereas positive and negative values, respectively, indicate convergence and divergence from V1 to region x. The two-sided Wilcoxon rank-sum test was applied to convergence magnitudes for comparisons between PCA patients and controls regarding the connections between V1 and the nine extrastriate areas. Bonferroni multiple-comparisons correction was applied for these nine comparisons.
We also assessed convergence magnitude along the visual hierarchy. In order to do so, we derived hierarchical levels from Haak et al.
17,18 in order to give some ordinal value to each visual area. In these studies, V1, V2, V3, hV4, LO1, LO2, TO1, and TO2 were, respectively, at hierarchical levels 0, 1, 2, 3, 3, 4, 5 and 6. Because we assessed LO1/LO2 and TO1/TO2 together (referred to as LO and TO in the rest of the text), we assigned them the average values of, respectively, 3.5 and 5.5. We assessed these levels in both streams together (i.e., V1 ↔ V2, V3, hV4, LO, and TO), in the dorsal stream (i.e., V1 ↔ V2d, V3d, and TO), and in the ventral stream (i.e., V1 ↔ V2v, V3v, hV4, and LO). For each participant, convergence magnitude was linearly regressed using these levels to assess the slope of convergence magnitude along the visual hierarchy. We assessed whether there was a slope in each group (using a single-sample signed-rank test to compare it to 0) and whether it differed between patient and control groups (using the Wilcoxon rank-sum test).
P values were Bonferroni corrected for the three comparisons (both streams, dorsal stream, and ventral stream).
We compared differences between V1 to region x and region x to V1 in goodness of fit, number of voxels that were included in the assessment (after excluding connective field sizes of 0.0001 or smaller; referred to as number of voxels in the results), and target area eccentricity between PCA patients and controls, as changes in these variables could potentially bias the results. If a difference was found, we assessed the Spearman correlation coefficient between the biasing variable and the convergence magnitude. Here, too, Bonferroni correction was used.