Presentation
Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Poster Slams
Longitudinal lesion-symptom mapping based on the linear-mixed model: a study of neural correlates of verbal fluency of stroke patients
Poster C70 in Poster Session C, Friday, October 7, 10:15 am - 12:00 pm EDT, Millennium Hall
Yanyu Xiong1, Mohamed Salah Khlif2, Natalia Egorova-Brumley2, Amy Brodtmann2, Brielle C. Stark1; 1Indiana University, 2University of Melbourne
Both voxel-based and multivariate lesion-symptom mapping approaches aim at establishing a statistical relationship between lesions of individuals and their impaired behavioral performances on a task. However, their applications in modeling longitudinal data are very limited due to methodological issues, such as incorporating repeated measures, missing data handling, and variance-covariance structure specification. Although recent research has used voxel-based repeated-measures analysis of variance to model the lesion-symptom relationship over time, its usefulness is still constrained by prerequisites like balanced research designs, no data clustering, and no missing data, etc. As many clinical longitudinal datasets call for more flexible models tailored to their specific data features, we proposed a novel voxel-based lesion-symptom mapping approach based on the linear-mixed model. In our study, we used this approach to investigate how white matter tract disconnections impact the semantic and phonemic fluency of stroke patients over time. The verbal fluency scores and anatomical T1 and T2 lesion data of 121 individuals at 3- (early chronic) and 12-months (chronic) post-stroke were obtained from the Cognition And Neocortical Volume After Stroke project with the lesion locus distributed on both hemispheres. The lesion prevalence across the subjects was on the right hemisphere. Each voxel-wise linear-mixed model included phonemic or semantic fluency scores as behavioral dependent variables. No significant differences were found between 3 and 12 months for each score (p >.05). The independent variables included the within-subject mean of lesions and the voxel lesion probability of each individual as predictors of interest. Only voxels damaged in at least 5% of the sample were included. The confounding effects of age and lesion volume were accounted for by running behavioral and lesion nuisance models before model fitting. As the simplest random intercept model excelled in model performance through hierarchical model selection, we obtained its corresponding volumetric statistical maps containing t and corrected p values of lesion predictors with the voxels significantly related to phonemic or semantic behavioral scores. The information on white matter disconnections was derived by overlapping the number of streamlines in each tract of the HCP842 tract atlas with the statistical maps at a threshold of 80%. Our results showed that white matter tract disconnections on the right hemisphere have different effects on phonemic and semantic fluency. The stroke patients with higher semantic fluency scores had lower risks of damage to the right arcuate fasciculus (AF) and right frontal aslant tract (FAT), providing the evidence that the right AF and FAT disconnections are related to the disrupted degree of semantic fluency. In contrast, stroke patients with enhanced phonemic fluency performance over time showed lower risks of damage to the right AF, suggesting its potential role in the neural recovery of phonemic fluency. To conclude, our study showed that lesion-symptom mapping based on the linear-mixed model could be employed as a more flexible approach to reveal both the cross-sectional and longitudinal effects of brain lesions on behavioral performance.
Topic Areas: Language Production, Disorders: Acquired