Presentation
Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Lightning Talks
Improving presurgical language mapping by a method for optimally sorting independent components of resting-state fMRI
Poster B96 in Poster Session B, Tuesday, October 24, 3:30 - 5:15 pm CEST, Espace Vieux-Port
Beatriz Vale1, Alexandre Andrade1, Martin Lauterbach2, Diogo Duarte3, Ricardo Vigário4, Christopher Benjamin5, Pedro Vilela2; 1Instituto de Biofísica e Engenharia Biomédica (IBEB), Faculdade de Ciências da Universidade de Lisboa, 2Hospital da Luz Lisboa, 3Champalimaud Foundation, 4Universidade Nova de Lisboa, 5Yale School of Medicine
Introduction: Pre-operative language mapping is an important component of presurgical planning of resectable brain lesions in the vicinity of eloquent areas. Many advances have been conducted in the last few decades towards the development and implementation of resting-state independent component analysis (ICA) for presurgical mapping in clinical practice. Particularly, as an alternative to task-based functional magnetic resonance imaging (fMRI) and the gold standard electrical cortical stimulation. More recently, resting-state fMRI has been used since it is a simpler technique and does not require the patient to cooperate in complex cognitive tasks. However, the methods for resting-state fMRI analysis are not yet robust or of practical usage. This work proposes a method for optimally sorting independent components (ICs) resulting from ICA so that components representing language resting-state networks take the first places in the component order. Methods: We recruited 20 healthy, right-handed volunteers and acquired both resting-state fMRI and task-based fMRI using three linguistic paradigms: object naming, verbal responsive naming, and verb generation. Task data was processed using general linear model (GLM) analysis while resting-state networks were extracted using ICA. Furthermore, it was developed an automated sorting procedure for the resting-state extracted ICs based on three characteristics: spatial similarity with the Neurosynth “language” probability map, ratio of low/high frequency and IC reliability over several bootstrapping folds. Results: Task-related activation consistent with the language network was identified at the individual and group-level. Furthermore, the proposed algorithm is shown to sort ICs with a guarantee that the resting-state language maps appear among the first five with an accuracy of 75%. When considering a symmetric language probability map which allows to be taken into account, in the algorithm, not only individuals who have left hemispheric dominance but also right and atypical language hemispheric dominance, the resting-state language maps appear among the first five sorted with a decreased accuracy of 60%. Overall, there was a good overlay between the sorted ICs of relevance and the task subject-specific language maps. The Dice coefficient measured between task and rest maps was found to be significantly higher when determined within language regions of interest rather than whole-brain analysis. Comparison between task and resting-state language maps showed that resting-state networks were more specific, reporting more activation in language-specific critical areas with fewer extraneous non-language activations when compared to task-related activation maps. However, resting-state networks were less sensitive than task language maps. Conclusion: Our findings suggest that optimally sorting components can contribute to making ICA usage viable in the clinical context since language components are more likely to be presented first, efficiently reducing the time spent in clinical evaluation. We expect that sorting components can become an alternative reliable method for presurgical planning in patients who cannot follow a task-fMRI protocol. However, further research is required to validate the sorting method proposed in a cohort of patients with brain lesions.
Topic Areas: Computational Approaches,