Abstracts

Rationale: Task-based fMRI is used for localizing language and motor functions prior to epilepsy surgery, yet not all patients are able to perform tasks in the scanner. This study used Independent Component Analysis (ICA) statistical methods on the resting state (rs) fMRI of healthy children and patients with epilepsy, to automatically delineate language and motor networks, and validate the results noninvasively with task fMRI activation.

Methods: Rs-fMRI and language Auditory Description Decision task (ADDT) fMRI data (5-6 minutes, TR=2s) from 60 subjects (49 patients age 7-21 and 11 healthy age 5-17), as well as rs-fMRI and motor task fMRI data from 38 subjects (27 patients age 10-21 and 11 healthy age 5-17), underwent fmriprep processing. ICA components were created using FSL melodic in native space at six different Total Number of Components (TNC, default and 20 to 60) and compared with language and motor target templates. The language template was generated by a Neurosynth.org meta-analysis of “language”, and was made symmetric across hemispheres. The motor template was a structural mask of the precentral and postcentral gyrus. The component of interest (COI) for each subject was determined by selecting the most stable component across thresholds within TNCs and with the highest Discriminability Index-based Component Identification (DICI, z(Hit Rate[% of overlap to target])-z(False Alarm[% of nonoverlap to non-target] Rate)) across TNCs. To validate the COI, first the hit rate for each subject was calculated between the COI and task activation map over a range of thresholds (task activation: top 10-20%, COI: z > 1 to 2) constrained to the language or motor template. Group mean hit rates were then calculated across subject groups and thresholds. Second, each subject’s success rate, i.e. whether the COI captured peak activation within canonical language (Broca’s:BA44,45, Wernicke’s: BA21,22,39) or motor (precentral, postcentral gyrus) regions was assessed at various z thresholds, and group percentages were then calculated.

Results: The mean hit rate between COI and ADDT activation across top % of activation and COI z thresholds for all subjects ranged from 52%-70%, while the mean hit rate between COI and motor task activation ranged from 85%-96%. The success rate of COI capturing the language activation peak is 69%-82% for Broca’s, and 85%-100% for Wernicke’s, and 78-100% for Motor activation peak across COI z thresholds and subject groups. The success rate of the ICA component in healthy subjects was 100% across Z thresholds in Wernicke’s and motor regions.

Conclusions: We demonstrate the sensitivity of an automated rs-fMRI ICA method in identifying eloquent cortical networks: motor > receptive > expressive language in healthy and pediatric epilepsy subjects. While sensitive, the COI extends beyond the margins of task activation and activation extends beyond the COI. Further investigation is necessary to determine the clinical value of the overlap as well as the unique areas mapped by each technique.

Funding: Please list any funding that was received in support of this abstract.: CTSI-CN Discovery Pilot Award UL1TR001876, Hess Foundation.