Presentation

Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Poster Slams

Systems Supporting Clinically Salient Language Deficits

Poster C32 in Poster Session C, Friday, October 7, 10:15 am - 12:00 pm EDT, Millennium Hall

Sigfus Kristinsson¹, Chris Rorden¹, Roger Newman-Norlund¹, Dirk B. den Ouden¹, Argye Hillis², Gregory Hickok³, Leonardo Bonilha⁴, Julius Fridriksson¹; ¹University of South Carolina, ²Johns Hopkins University, ³University of California, Irvine, ⁴Medical University of South Carolina

Introduction Language is a complex higher-order cognitive function supported by two large systems: a ventral (temporal-parietal) network primarily responsible for semantic computations, and a dorsal (parietal-frontal) network involved with sequential processing and phonological mapping.1,2 Critically, everyday language functions like speech comprehension, naming, speech repetition, and spontaneous speech rely on shared components of the ventral and dorsal streams.2,3 Prior research in aphasia has historically examined neural correlates of each functional domain in isolation, ignoring their reliance on shared structural networks and the high correlation across functional abilities in these domains.4 In an effort to remediate this issue, we employed a singular value decomposition approach and partial least squares to test the simultaneous involvement of behavioral deficits onto damaged ventral and dorsal structures as a core mechanism to explain variance in impairment. Method A total of 93 participants with chronic (>12-months post-stroke) aphasia after one or more left hemisphere (LH) strokes were recruited for the current study. Participants underwent a detailed case history, language assessment (Western Aphasia Battery, WAB; Kertesz, 2007), and neuroimaging. Multiple facets of brain damage were defined by neuroimaging, namely: 1) lesion characteristics derived from T1-/T2-weighted imaging, 2) connectomics (functional and structural) derived from high-resolution diffusion tensor imaging tractography and resting-state fMRI, 3) gray and white matter tissue integrity, 4) task-based fMRI, 5) tissue microstructure derived from diffusion MRI, and 6) tissue perfusion measured with arterial spin labeling. A partial least squares approach was employed for model construction to evaluate the shared and unique aspects of both the dependent factors (WAB subscores: Speech Comprehension, Naming, Repetition, and Spontaneous Speech) and their predictors (neuroimaging data). A leave-one-participant-out machine learning algorithm was applied to assess model accuracy, while also determining the association between the latent decompositions of dependent factors and their predictors. Model accuracy was evaluated based on Pearson’s correlation between observed vs. predicted language scores. Results Spontaneous Speech performance emerged as the dominant component in all models irrespective of neuroimaging modality (loading range: -36 to 27). The most accurate prediction of Speech Comprehension was achieved by a model based on the structural connectome (r=.45, p<.01), followed by task-based fMRI (r=.35, p<.01). Performance on the Naming subtest was most accurately predicted based on white matter voxel-based morphometry (VBM; r=.39, p<.01), followed by the structural connectome (r=.37, p<.01). Manually demarcated lesion data composed the most accurate prediction of Speech Repetition scores (r=.38, p<.01), followed by a model based on tissue microstructure (fractional anisotropy, FA; r=.33, p<.01). Finally, the structural connectome model achieved the highest accuracy by far for the prediction of Spontaneous Speech scores (r=.42, p<.01). Models based on white matter VBM and FA values both achieved an accuracy of r=.01 (p<.01). Discussion Our findings indicate that speech production is a significant component in clinically salient speech impairments, and that variance in impairments is explained by lesion affecting both the ventral and dorsal language streams. These results highlight the importance of considering multiple neuroimaging modalities simultaneously to explore the intricate nature of language deficits for future research and clinical purposes.

Topic Areas: Disorders: Acquired, Computational Approaches