Movie-watching fMRI for presurgical language mapping in patients with brain tumour

Introduction

The primary goal of presurgical language mapping is localising critical language areas with high sensitivity (i.e., capturing areas in which resection could lead to language deficits) and specificity (i.e., excluding non-language areas) and reliably determining language hemispheric dominance, on an individual basis. Language mapping is challenging due to the widely distributed functional organisation of language in the frontal, temporal and parietal lobes, and in neurosurgical patients the possibility of tumour-induced functional reorganisation.

A major drawback of conventional task-based functional magnetic resonance imaging (tb-fMRI) recommended for presurgical language mapping [1] is the contingency on patient performance of precisely timed tasks (e.g., antonym generation - AntGen). Drawbacks of task-free resting-state fMRI (rs-fMRI) include confounding effects of ‘mind wandering’ and sensitivity to motion artefacts. In contrast, movie watching is a rich, stimulating and naturalistic activity, predicted to constrain cognitive processes and engage the distributed, multimodal neural networks supporting language function in real life [2].

Our previous study demonstrated individual language mapping using movie-watching fMRI (mw-fMRI) in neurologically healthy subjects [3]. Here, we examine mw-fMRI language mapping in presurgical patients with a brain tumour encroaching on putative language cortex, and varying levels of language disruption. We hypothesise that mw-fMRI versus AntGen tb-fMRI, and rs-fMRI, will provide comprehensive language mapping at reduced burden, as determined by metrics of in-scanner head motion, and mapping specificity, sensitivity, and lateralisation.

Methods

Mw-fMRI was compared with clinically indicated AntGen tb-fMRI in 34 patients with brain tumour undergoing presurgical language mapping, and with rs-fMRI in 22 of these patients. See online supplemental methods for exclusion criteria, and online supplemental table S1 for demographic and clinical information. Language maps were generated from tb-fMRI using a general linear model, and from mw-fMRI and rs-fMRI using independent component analysis. For mw-fMRI, primary and secondary language components (LC1, LC2) with distinct time courses were identified by spatial correlation with a canonical language template comprising regions in frontal, temporal and parietal cortices, and temporal correlation with a model time course derived from an independent neurologically healthy group (online supplemental figure S1) [3]. For rs-fMRI, a language component was identified using spatial correlation with the language template. Language mapping metrics were calculated for each paradigm at 10% threshold (online supplemental figure S3 shows additional thresholds). All research procedures were approved by the Mass General Brigham Institutional Review Board.

Results

Head motion

Mw-fMRI data from one subject (2.9%), tb-fMRI data from four subjects (11.8%) and rs-fMRI data from two subjects (9.1%) were excluded from analysis due to excessive head motion. The extent of head motion (measured as framewise displacement) in the included patients was lower for mw-fMRI (0.13mm) than tb-fMRI (0.16mm, p=0.003), with no other differences found between paradigms.

Language maps

Data from one subject were excluded due to failure to identify a language component in both mw-fMRI and rs-fMRI. Maps representing the inter-subject activation overlap in tb-fMRI (n=30), mw-fMRI LC1 (n=32) and LC2 (n=30), and rs-fMRI (n=18), relative to a canonical language template, are shown in figure 1A. Examples of individual activation maps are shown in online supplemental figure S2.

Figure 1.

fMRI language mapping based on movie-watching (mw-fMRI), antonym generation (AntGen tb-fMRI), and resting-state (rs-fMRI) in patients with brain tumour. (A) Maps of intersubject overlap in activation for the mw-fMRI primary language component (movie-LC1, n=32), mw-fMRI secondary LC (movie-LC2, n=30), AntGen tb-fMRI general linear model (AntGen-GLM, n=30), and rs-fMRI LC (resting-LC, n=18). The overlap maps were thresholded at ≥5 subjects. The arrows on the colour bars indicate the greatest number of subjects with spatially overlapping activation in each paradigm. The functional overlap maps are shown overlaid on the canonical language mask (in white) and Montreal Neurological Institute−152 anatomical template. (B) Example of a subject-specific language mask, segregated into frontal (red), temporal (green) and inferior parietal (yellow) regions, overlaid on a gadolinium-enhanced T1-weighted MRI anatomical image, in a patient with a left temporal glioblastoma multiforme (WHO grade IV). (C) Comparison of sensitivity and specificity of movie-LC1 and movie-LC2 versus AntGen-GLM and resting-LC, in each language region. *P<0.05, **p<0.01, ***p<0.001 (corrected for multiple comparisons using the Bonferroni method). fMRI: functional magnetic resonance imaging; WHO: World Health Organization.

Sensitivity and specificity

Compared with tb-fMRI, the mw-fMRI LC1 showed higher sensitivity in bilateral temporal cortex and lower sensitivity in bilateral frontal cortex; LC2 showed higher sensitivity in parietal cortex. The mw-fMRI LC1 showed higher specificity than tb-fMRI. The mw-fMRI sensitivity was equivalent to that of rs-fMRI and the specificity was higher (figure 1C).

Laterality

There were significant and marginal differences in laterality between mw-fMRI and tb-fMRI, in parietal (p=0.039), temporal (p=0.063) and frontal (p=0.058) regions (online supplemental figure S4). Most disagreement cases showed left-hemispheric dominance in tb-fMRI and right-hemispheric dominance in mw-fMRI. There were no differences between mw-fMRI and rs-fMRI.

Discussion

Head motion

Head motion can confound functional connectivity estimates. Here, the finding of reduced head motion in mw-fMRI versus tb-fMRI (measured as fewer subjects excluded due to excessive motion, and smaller extent of motion in the included subjects), despite the mw-fMRI being administered last, is therefore important. Mw-fMRI may be associated with reduced motion because it helps reduce the mental discomfort of scanning.

Specificity and sensitivity

The higher language mapping specificity of mw-fMRI versus tb-fMRI and rs-fMRI indicates lower occurrence of false-positive activation outside of putative language areas. Practically, higher specificity may translate into increased confidence for neurosurgical tumour removal outside of the mapped language areas, which is important because extent of resection is a significant predictor of overall survival in patients with brain tumour. The lower specificity of the AntGen paradigm could be due to greater activation of visual areas in the task versus control condition, exemplifying that tb-fMRI depends on the choice of baseline.

Higher sensitivity of language mapping indicates lower occurrence of false-negative activation inside putative language areas, which is important because it could translate into reduced risk of post-surgical language deficits related to underestimation and accidental removal of functional tissue. The present findings suggest that mw-fMRI may provide superior mapping sensitivity in temporal and parietal cortices primarily associated with receptive language function, although lower mapping sensitivity in frontal areas primarily associated with expressive language function. This finding is consistent with prior work showing that phonologically demanding tasks specifically engage the dorsal auditory phonological stream and frontal cortex, whereas movie-watching engages a broad temporoparietal semantic network [4].

Lateralisation

Mw-fMRI showed good lateralisation agreement with rs-fMRI, but some degree of disagreement with tb-fMRI. The AntGen task may engage primarily the dorsal auditory phonological stream which is strongly left lateralised, whereas the naturalistic paradigms engage primarily semantic systems which are less lateralised [4,5].

Conclusion

We reported an mw-fMRI paradigm and language mapping pipeline that entail reduced burden on patients and staff. Mw-fMRI versus tb-fMRI and rs-fMRI demonstrated reduced or equivalent head motion, higher sensitivity in temporoparietal cortex and lower in frontal cortex, and higher specificity, supporting our hypothesis that receptive language can be mapped effectively using a naturalistic paradigm. Thus, mw-fMRI could complement or be an alternative to tb-fMRI in patients with a temporoparietal lesion, for whom high-precision presurgical mapping of this area is essential, and task performance may be challenging. In contrast, explicit phonological tasks may be better suited for mapping the frontal cortex with high precision, and for determining language hemispheric dominance.