Naturalistic fMRI for Presurgical Language Mapping

2031 

Abstract Submission 

Yanmei Tie¹, Laura Rigolo², Colin Galvin², Alexandra Golby³, Einat Liebenthal⁴

¹Brigham and Women's Hospital, Harvard Medical School, Boston, MA, ²Brigham and Women's Hospital, Boston, MA, ³Harvard Medical School, Boston, MA, ⁴McLean Hospital, Harvard Medical School, Belmont, MA

Yanmei Tie  
Brigham and Women's Hospital, Harvard Medical School
Boston, MA

Laura Rigolo  
Brigham and Women's Hospital
Boston, MA

Colin Galvin  
Brigham and Women's Hospital
Boston, MA

Alexandra Golby  
Harvard Medical School
Boston, MA

Einat Liebenthal  
McLean Hospital, Harvard Medical School
Belmont, MA

Task-based functional MRI (task fMRI) is conventionally used for presurgical language mapping[1]. However, neurosurgical patients may have difficulty performing the language tasks satisfactorily, particularly if they have pre-existing language or other cognitive deficits[2], which could in turn affect the quality of task fMRI language mapping. To tackle this problem, we adopted naturalistic fMRI using movie clips as stimuli (movie fMRI), based on the hypothesis that movies depicting real-life conversation scenes can more completely engage the language networks supporting every-day language communication. We previously demonstrated the feasibility and effectiveness of movie fMRI for mapping language areas in individual healthy subjects[3], and for presurgical language mapping in individual brain tumor patients[4]. The present work compares different analytic approaches to movie fMRI for language mapping, in relation to task fMRI.

We reanalyzed data from our initial study[3] in 22 right-handed healthy native English speakers (11 females, mean age=26.3 yrs, range: 19-39 yrs). All subjects underwent movie fMRI with a 7-min excerpt from the film "The Parent Trap", which included 7 conversation segments interleaved with 6 non-speech scenes. Subjects also underwent language task fMRI consisting of 7 blocks of antonym generation and 7 blocks of letter case categorization, interleaved with fixation intervals.

Two main analytical approaches were investigated for movie fMRI: (1) a hypothesis-driven general linear model (GLM) with (a) a regressor of weighted word count derived from the movie subtitles; (b) a regressor of conversation segments derived based on the weighted word count convolved with a hemodynamic response function. (2) a Data-driven group independent component analysis (ICA), with 34 components extracted using the Infomax algorithm, of which one language component was identified based on temporal correlation with the conversation regressor. In addition, an inter-subject correlation (ISC) analysis was performed[5].

Figure 1a-d shows the movie group results. For GLM with the weighted word count as a linear regressor, one activation cluster was revealed in the left superior temporal sulcus (Fig. 1a), consistent with an area specialized for sub-lexical phonemic processing[6]. For GLM with the conversation regressor, positive activation was seen in multiple putative language areas in the left inferior frontal and bilateral temporoparietal cortex, and there was also negative activation (i.e., greater in non-conversation segments) in visual areas (Fig. 1b). The ICA derived language map (Fig. 1c) was similar to the conversation GLM map (Fig. 1b) but showed no activation in visual areas (Fig. 2b). The ISC map revealed extensive areas of synchronized activation across subjects (Fig. 1d). Within the language network, the strongest cross-subject synchronization was seen in the left posterior middle temporal gyrus (Fig. 2a). The task fMRI map showed extensive activation in the frontal cortex and basal ganglia and limited activation in left posterior middle and inferior temporal cortex (Fig. 1e). Compared to task fMRI, movie fMRI revealed more superior and middle temporal activation and less insulae activation (Fig. 2c-d).

The results demonstrate the efficacy of movie fMRI language mapping. First, different regressors can be used in GLM analyses of movie fMRI to map different aspects of language processing, as exemplified here for phonemic perception (Fig. 1a). Second, data-driven analysis of the movie fMRI data provided more comprehensive mapping of temporal language regions than task fMRI. Third, the data-driven analysis of movie fMRI returned a specific language component, whereas the GLM-derived language map included residual visual activation. Finally, ISC analysis revealed varying levels of cross-subject synchrony during movie watching within the language network, which we will further investigate in future work.

Language Comprehension and Semantics ²

Speech Perception

Activation (eg. BOLD task-fMRI)

fMRI Connectivity and Network Modeling

Task-Independent and Resting-State Analysis ¹

Aphasia

FUNCTIONAL MRI

Language

Multivariate

Neurological

Other - Naturalistic fMRI