Slide Slam

< Slide Slam Sessions

Slide Slam A11

Pre-treatment graph measures of resting-state functional connectivity predict language treatment outcome in aphasia

Slide Slam Session A, Tuesday, October 5, 2021, 12:30 - 3:00 pm PDT Log In to set Timezone

Isaac Falconer1, Maria Varkanitsa1, Swathi Kiran1; 1Boston University

Introduction: A great deal of variability exists in treatment-related behavioral outcomes in patients with post-stroke aphasia. In addition to known factors such as aphasia severity and lesion size and location, differences in the functional connectivity (FC) of intact brain regions are likely to play an important role. Graph measures of FC can be used to characterize properties of resting state networks. Integration and segregation measures have been found to be associated with behavioral deficits, but their relationship to treatment outcomes is not known. Here, we investigate this relationship and hypothesize that graph measures of FC in four functional networks, language (LN), default mode (DMN), dorsal attention (DAN), and salience (SN), predict treatment response in people with aphasia (PWA). Methods: Thirty chronic PWA (10F, time post stroke: mean=52 months, range=8–170 months) due to single left hemisphere stroke completed up to 12 weeks of semantic feature analysis treatment for word retrieval deficits. Mean baseline aphasia severity from the Western Aphasia Battery–Revised (WAB-AQ quotient) was 59.83 (range=11.7–95.2). T1-weighted anatomical scans and whole-brain resting-state fMRIs were collected at the baseline and preprocessed in FMRIPREP. Bivariate Pearson correlations between pairs of regions of interest (50 ROIs in total) were calculated in CONN. Patients’ FC matrices were entered in BRAPH and converted to weighted, undirected graphs (one graph per network). We then calculated the weighted variants of three global graph measures (i.e., degree, global efficiency, clustering). The relationship between global graph measures of FC and treatment outcome was examined using mixed effects logistic regression. The models included treatment session and the graph measure of interest as predictors of accuracy on naming probes. WAB-AQ was also included as a covariate. Results: PWA with lower average degree (p<0.001) but higher global efficiency (p<0.001) and clustering (p<0.001) in LN showed greater improvement over time compared to those with higher average degree and lower global efficiency and clustering. In DMN, PWA with lower average degree (p<0.001) and global efficiency (p<0.01) showed greater improvement over time. PWA with higher global efficiency (p<0.001) and clustering (p<0.001) in DAN and higher average degree (p<0.001), global efficiency (p<0.001), and clustering (p<0.001) in SN showed greater improvement over time. Conclusion: In LN, lower average degree is associated with greater treatment response whereas higher global efficiency and clustering are associated with greater treatment response. One interpretation of these findings is that moderately high FC throughout LN, which would result in a high average degree, is not beneficial, but rather a more organized network in which a few connections are very strong while others are weaker is needed to maximize benefit from treatment. As for the non-language networks, our results suggest that greater integration and segregation in DAN and SN are associated with greater treatment response, whereas lower integration in DMN is associated with greater response. Overall, our results demonstrate that connectivity in both language and non-language networks is a good predictor of treatment response and provide support for a framework of language recovery that accounts for differences in network topology.

< Slide Slam Sessions

SNL Account Login

Forgot Password?
Create an Account