Abstract:Using the unique organization of an achiasmic individual's visual cortex, we showed that the amplitude of an evoked BOLD response is proportional to the amplitude of the evoked neural response raised to a power of approximately 0.5 (Bao, Purington & Tjan, 2015, eLife). Specifically, we found that in V1-V3 of an achiasmic individual, there are two nearly identical but non-interacting neural populations that are finely intermingled in the same cortical location. Their population receptive fields are spatially disjointed, allowing independent control of each population with visual stimuli. Since the neural populations do not interact, presenting two identical stimuli, each to one of the receptive fields, doubles the local neural activity relative to presenting just one. This in-vivo system lets us identify the nonlinearity between neural and BOLD responses independently from any nonlinearity between stimuli and neural responses. In the current study, we analyzed data from several experiments with the achiasmic participant and found that a simple linear-nonlinear model can approximate the neural-BOLD relationship in time. This model postulates a hemodynamic "control signal" that is linearly related to the evoked neural response during a stimulus event. The resulting BOLD time course is the sum of such control signals, raised to a power of 0.5 in a sign-preserving manner. This model suggests that nonlinearities associated with spatial summation, adaptation, and surround suppression observed with fMRI can be considerably contaminated by hemodynamic nonlinearity, an observation supported by recent findings (Bao et al., 2014VSS). The model also predicts that functional connectivity inferred from task states by linearly regressing out task-related activities can be misleading, since task harmonics produced by the nonlinear hemodynamics will be present in the residual. Squaring a baseline-subtracted BOLD response before applying any general linear model and using a modified hemodynamic response function can reduce the effects of hemodynamic nonlinearity on results. Meeting abstract presented at VSS 2016

Analysis of fMRI Time-Series Revisited—Again

Analysis of fMRI Time-Series Revisited

Fast Orthogonal Search Method for Modeling Nonlinear Hemodynamic Response in Fmri

Fmri Time Series Analysis Based on Stationary Wavelet and Spectrum Analysis

Time Series Analysis of fMRI Data: Spatial Modelling and Bayesian Computation

Temporal Autocorrelation in Univariate Linear Modeling of FMRI Data

Validity and power in hemodynamic response modeling: A comparison study and a new approach

Treatment of Baseline Drifts in fMRI Time Series Analysis

Unbiased estimation of the coefficient of determination in linear models: an application to fMRI encoding model comparison

Problems Associated with a Nonlinear Relationship Between Neural and Fmri BOLD Responses and a Solution

Assessment and Optimization of Functional MRI Analyses

HRF estimation improves sensitivity of fMRI encoding and decoding models

fMRI Signal Restoration Using a Spatio-Temporal Markov Random Field Preserving Transitions

A new fMRI data analysis method using cross validation: Negative BOLD responses may be the deactivations of interneurons

Modelling the data and not the images in FMRI

Prototype-Based Spatio-Temporal Probabilistic Modelling of fMRI Data.

A Bayesian General Linear Modeling Approach to Cortical Surface fMRI Data Analysis

The Statistical Analysis of fMRI Data

FMRI Clustering in AFNI: False-Positive Rates Redux

Revisiting the importance of model fitting for model-based fMRI: It does matter in computational psychiatry

Cluster failure: Why fMRI inferences for spatial extent have inflated false-positive rates