Abstract
Little is known about how language functional MRI (fMRI) is executed in clinical practice in spite of its widespread use. Here we comprehensively documented its execution in surgical planning in epilepsy. A questionnaire focusing on cognitive design, imaging acquisition, analysis and interpretation and practical considerations was developed. Individuals responsible for collecting, analyzing, and interpreting clinical language fMRI data at 63 epilepsy surgical programs responded. The central finding was of marked heterogeneity in all aspects of fMRI. Most programs use multiple tasks, with a fifth routinely using 2, 3, 4 or 5 tasks with a modal run duration of five minutes. Variants of over fifteen protocols are in routine use with forms of noun-verb generation, verbal fluency, and semantic decision-making used most often. Nearly all aspects of data acquisition and analysis vary markedly. Neither of the two best-validated protocols were used by more than 10% of respondents. Preprocessing steps are broadly consistent across sites, language-related blood flow is most often identified using general linear modeling (76% of respondents), and statistical thresholding typically varies by patient (79%). The software SPM is most often used. fMRI programs inconsistently include input from experts with all required skills (imaging, cognitive assessment, MR physics, statistical analysis, brain-behavior relationships). These data highlight marked gaps between the evidence supporting fMRI and its clinical application. Teams performing language fMRI may benefit from evaluating practice with reference to the best-validated protocols to date and ensuring individuals trained in all aspects of fMRI are involved to optimize patient care.
