Objective metrics for language lateralization of fMRI examinations: a new model for the classification of hemispheric dominance in healthy subjects and epileptic patients

TL;DR

This study compares methods for calculating hemispheric language dominance in fMRI, introducing a sigmoidal model that improves classification accuracy and consistency with clinical reports across healthy and epileptic subjects.

Contribution

A new sigmoidal model for classifying language lateralization in fMRI, demonstrating high agreement with clinical assessments and applicability to both healthy and epileptic populations.

Findings

LI AVE and LI VOL yield similar classification results.

The sigmoidal model achieves up to 100% agreement with clinical reports.

The model provides thresholds for dominant and co-dominant classification.

Abstract

Purpose: to compare different methods to calculate Laterality Index (LI), a metric which allows to evaluate hemispheric brain language dominance in functional MRI examinations (fMRI). Methods: Two methods were considered for calculating LI: LI AVE and LI VOL , respectively based on the differences between measurements of average and volume of fMRI signal in brain hemispheres. Laterality curves were obtained calculating values of LI VOL with increasing thresholds of fMRI signal and fitted with sigmoidal functions. A model for dominant and co-dominant classification based on fit parameters has been developed. The two methods and the sigmoidal model were applied to two cohorts of 93 epileptic patients and 27 healthy subjects undergoing language fMRI examinations with association, understanding and fluency tasks. Results: Despite the different definitions, LI AVE and LI VOL resulted in equivalent classification of language lateralization. The agreement of neuroradiological clinical reports with classification of language lateralization resulting from the proposed methods ranged from 94.6% to 89.2% for LI metrics and up to 100% for the sigmoidal model. The fit parameters of the sigmoidal function defined empirical thresholds useful for classification between dominant and co-dominant, providing similar values for subjects and epileptic patients for fluency and association tasks. This result supports the idea of a unique model for language lateralization classification in epileptic patients and healthy subjects. Conclusions: Language lateralization in fMRI can be effectively assessed by objective metrics. A novel approach based on sigmoidal fit of laterality curves resulted in higher agreement with clinical reports providing further information about the strength of language lateralization.