Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Intrinsic Connectivity within a Functionally Defined Language Network Predicts Phonological Competence in Chronic Post-Stroke Aphasia
Poster Session A, Wednesday, September 30, 11:00 am - 1:00 pm, Wangari Maathai
Olga Boukrina1,2, Chirag Motwani1, Tamar Truzman3, Elizabeth B. Madden4, Brian M. Sandroff1,2, Melissa Rosahl1, Glenn Wylie1,2, William W. Graves5; 1Kessler Foundation, West Orange, NJ, USA, 2Rutgers New Jersey Medical School, Newark, NJ, USA, 3University of Cambridge, Cambridge, UK, 4Florida State University, Tallahassee, FL, USA, 5Rutgers University, Newark, NJ, USA
Reading deficits are prevalent in chronic post-stroke aphasia (Brookshire et al., 2014) and limit functional independence. Although lesion location contributes to impairment, it does not fully account for variability in aphasia outcomes, suggesting a role for functional interactions among surviving cortical regions (Gleichgerrcht et al., 2017). Resting-state functional connectivity (rsFC) studies in chronic aphasia have demonstrated disruptions across language and domain-general systems (Falconer et al., 2024; Klingbeil et al., 2019); however, prior work has focused on whole-brain or large-scale network alterations, obscuring dysfunction within core language-selective regions (Johnson et al., 2025). We examined whether intrinsic connectivity within a functionally defined language network predicts reading-related performance after stroke. Sixteen individuals with chronic left-hemisphere stroke, aphasia, and reading deficits (3 females; mean age=59.31±12.49 years) and sixteen age-matched healthy adults (6 females; mean age=57.31±12.55 years) underwent MRI scanning on a 3T Siemens scanner. Resting-state fMRI data were preprocessed using fMRIPrep. Analyses were restricted to cortical regions defined by the overlap of two independently validated language localizers (Fedorenko et al., 2010; Graves et al., 2024). The resulting regions were mapped onto Gordon cortical atlas parcels for parcel-wise analyses. Lesions were segmented using nnU-Net–based models trained on multi-site stroke MRI datasets (Truzman et al., n.d.) to exclude lesioned voxels. Mean ROI time series were extracted following nuisance regression, detrending, motion scrubbing, and bandpass filtering (0.01–0.08 Hz), and Fisher z-transformed connectivity matrices were computed for each participant. Behavioral measures included word and nonword reading aloud, as well as phonological, orthographic, and semantic processing. Global mean connectivity analyses across network edges showed weak associations with behavior. Therefore, multivariate predictive modeling was performed using elastic-net regression with leave-one-subject-out cross-validation (LOOCV) to determine whether distributed rsFC patterns predicted behavioral performance while accounting for lesion volume. Phonological task accuracy (pseudoword rhyme judgement) was most strongly related to rsFC patterns. An FC-only multivariate model significantly predicted phonological performance (LOOCV Pearson r=0.62, p=0.01; Spearman rho=0.56, p=0.025), indicating that intrinsic connectivity patterns were behaviorally informative independent of lesion burden. Prediction further improved in a lesion-adjusted model (Pearson r=0.79; Spearman rho=0.76, both p’s<0.001), suggesting complementary contributions of lesion burden and distributed rsFC organization. Stable predictive edges involved bilateral inferior frontal and temporoparietal regions, including left angular gyrus connectivity with left and right inferior frontal regions. Positive predictive edges were observed primarily among dorsal language-related connections, whereas increased right frontal and bilateral frontal coupling showed negative associations with phonological performance, potentially reflecting maladaptive compensatory recruitment. A direction-weighted composite FC score derived from stable predictive edges strongly correlated with phonological performance (Spearman rho=0.84, p<0.001) and remained significant after controlling for lesion volume. Compared to healthy controls, stroke participants showed increased right and bilateral frontal coupling, increased left temporoparietal–frontal connectivity, and reduced interhemispheric frontoparietal and left frontal connectivity, although group differences did not reach statistical significance in the current sample. Thus, chronic reading impairment may reflect specific patterns of adaptive and maladaptive intrinsic language-network reorganization rather than globally reduced connectivity. Functionally constrained rsFC measures may provide sensitive biomarkers of phonological network integrity after stroke.
Topic Areas: Disorders: Acquired, Reading