Figure 1
(a) pSFT fMRI scan design. The default scan consists of 6 40-s stimulus blocks surrounded by 10-s blank periods. In each stimulus block, 40 SF bandpass-filtered stimuli are presented in a random order while participants perform a luminance change detection task at fixation. (b) Stimulus generation. Every stimulus presented begins as a random sample of uniform white noise. A 2D fast Fourier transform (FFT) is performed on the noise sample, which is then masked by a circularly symmetric narrow-band filter (filter width is exaggerated here for visualization purposes). The resulting image is returned to the spatial domain and converted to 8-bit grayscale. (c) Examples of bandpass-filtered stimuli. (d) Examples of stimulus SF energy as a function of SF. The target SF (cpd) is provided above the peak of each energy profile.
Figure 2
(a) Voxel-wise estimation of population spatial frequency tuning. A sequence of SFs is fed into a log-Gaussian function to produce a neural response. The neural response is convolved with a HRF to synthesize a BOLD signal that can be compared to the measured BOLD signal. This process is repeated and the pSFT parameters optimized until the sum of squares error (SSE) between the predicted and measured BOLD is minimized. (b) pSFT as a function of pRF eccentricity. pSFT and pRF parameter estimates are from a sample dataset of visual areas V1–V3 [11]. n = 200 in every subplot.
Figure 3
Computational Validity. (a) Simulated BOLD. Each subplot depicts the simulated BOLD signal of a voxel with one of three BOLD noise levels. Black curves are the clean BOLD signal, while the green, yellow, and red curves are the BOLD signal with noise. (b) Parameter recovery. Each subplot depicts simulated vs. estimated pSFT parameters at each BOLD noise level from 1000 simulated voxels (top row, peak; bottom row, bandwidth). The correlation coefficients are displayed on the bottom right of each subplot.