Clinical fMRI is becoming standard for mapping language function prior to brain surgery. However, variability in paradigms and analyses across institutions has made it difficult to assess the relative validity and reliability of different approaches or to use the data collected to increase our understanding of the brain’s underlying language mechanisms. This is further complicated by increasing recognition that language processing is more complex than suggested by the classical model of anterior expressive (Broca’s) and posterior receptive (Wernicke’s) areas, instead involving a network of regions within both the dominant and non-dominant hemispheres.
To address this lack of quantitative data we performed a retrospective analysis of all pre-surgical fMRI language mapping cases at UCSF over the past 7 years. Our goals are threefold. First, to compare clinical assessments of language dominance made at the time of testing with the aid of a scanner-based FDA-approved software package (BrainWave, GE Healthcare) to results derived from post-hoc processing of the raw BOLD data using a standard ‘research pipeline’. Second, to explore how variability in statistical thresholds and regions of interest influence the laterality-index (LI), a metric widely used to quantify language dominance. Third, to apply an optimized-LI approach to compare the relative efficacy of different masks and paradigms in determining language dominance, and validate these LI-based results using a group-level analysis of population data of the type standard in the research literature.
By computing laterality indices using a functionally derived language mask and thresholding based on percentiles of individual subject’s maximum activation we achieved good correlation between laterality indices and clinical assessments of language dominance. These correlations peaked at thresholds of approximately 90% of maximum subject activation and then declined, suggesting that fixed thresholds may be poorly suited to data collected in the context of pre-surgical fMRI. Correlations were also much higher when based on a functional language mask derived from the Neurosynth meta-analytic platform than when restricted to the anatomically defined regions that correspond to the classical model of ‘Broca’s’ and ‘Wernike’s’ areas.
Comparing laterality indices across tasks as a function of threshold we found clear differences in performance among different language tasks in determining hemispheric dominance for language, with verb generation tasks outperforming all others. We confirmed this relative strength of language network activation across different tasks through between-task contrasts derived from an entirely separate group-level analysis of pair-wise population data. Overall our results suggest that laterality indices computed using a functionally derived language mask and thresholds based on individual subject’s maximum activation provide a robust quantitative approach for assessing language dominance and for comparing results across subjects and paradigms.