Epoch versus impulse models in the analysis of parametric fMRI studies

Objective In parametric fMRI studies the relationship between the amplitude of the hemodynamic response and electrophysiological or behavioral parameters is commonly analyzed using the general linear model (GLM). We examined ways of using single-trial response time (RT) in the analysis of a decision-making task to better isolate task-specific activation. Methods fMRI and RT data were recorded in twenty-one subjects performing a visual-oddball-task. Four explanatory variables (EVs) were generated for the GLM-analysis: A conventional (constant impulse) EV, a constant epoch EV informed using subjects’ average RT, a variable impulse EV and a variable epoch EV both informed using single-trial RT. EVs were tested individually and as orthogonalized pairs. Results The individual EVs all detected similar extensive patterns of activation, while orthogonalized EVs were mainly correlated with BOLD signal variance in sensorimotor and parietal areas. Orthogonalizing the variable epoch EV to the constant epoch EV isolated cortical regions resembling the “dorsal frontoparietal attention network” from activation detected by the conventional (i.e., constant impulse) analysis. Conclusion For short event durations, the activation detected by individual EVs is very similar, but orthogonalized, parametrically informed EVs can improve isolation of task-specific BOLD signal change. Significance Different approaches for integrating parametric timing measures in fMRI analyses can significantly influence outcomes, refining or confounding findings.