Search Abstracts | Symposia | Slide Sessions | Poster Sessions
A five-factor decomposition of narrative speech in chronic aphasia and its neural correlates using lesion symptom mapping
Poster Session E, Sunday, September 14, 11:00 am - 12:30 pm, Field House
Sophie Steinberg1,3, Catherine R. Kelly2,3, Alycia B. Laks2,3, Peter E. Turkeltaub1,2,3,4, Andrew T. DeMarco1,2,3; 1Georgetown University Medical Center, 2Department of Neurology, Georgetown University Medical Center, 3Center for Brain Plasticity and Recovery, Georgetown University Medical Center, 4MedStar National Rehabilitation Hospital
Introduction Narrative speech is representative of daily language use and therefore may provide insights into brain-behavioral relationships of realistic speech and language use in aphasia. Prior studies have applied dimensionality reduction to discourse samples to identify underlying deficits in narrative speech. For example, Alyahya et al. (2020) found three components (verbal quantity, verbal quality, motor speech), while Casilio et al. (2024) found four (paraphasia, logopenia, agrammatism, and motor speech). Both prior studies implicated the precentral gyrus, temporal, and frontal regions as loci associated with their respective components, with some variability in exact localization. Here we attempt to reconcile some of these differences in prior work by applying principal components analysis (PCA) on our own large dataset, followed by voxel-based lesion-symptom mapping of the resulting components. Methods We analyzed 59 controls (31M/28F; mean age 59.9 years) and 95 left-hemisphere stroke survivors (55M/40F; mean age 60.1 years; median lesion volume 74.95 cm3, mean time-since-stroke 43.3 months, mean Western Aphasia Battery-Aphasia Quotient 76.76). All participants described the WAB-R Picnic Scene and the Cookie Theft scene. For each participant, we tabulated core lemmas (based on controls), closed class words, pseudowords, unique nouns, unique verbs, other open class words (open class words minus nouns and verbs), and grammatical morphemes– all per utterance, per minute, and total. These 21 variables were reduced using PCA, with varimax rotation (patients and controls together). Anatomical lesions were traced from high-resolution MPRAGE/FLAIR scans. Multivariate lesion-symptom mapping was used to examine lesion correlates of each principal component (corrected voxelwise p<.005 and clusterwise p<.05, based on 5000 permutations). Results Dimensionality reduction produced five components, which we interpret as 1) speech quantity (totals of measures), 2) speech rate/fluency (primarily measures calculated per minute), 3) grammar (primarily measures calculated per utterance), 4) phonological processing (pseudowords per minute, per utterance, and total), and 5) intricacy of speech (other open class words per minute and per utterance). When components were submitted to LSM, two were significant. “Speech quantity” yielded a significant cluster (p=.04, vol=6.15cm3, MNI x=-39.9,y=-13.5,z=42.3) spanning the precentral gyrus, from the hand area through the frontal operculum. “Speech rate/fluency” yielded a significant cluster (p=.03, vol=7.94cm3, MNI x=-29,y=-7.9,z=5.4) overlapping the basal ganglia and posterior insula, plus the white matter between them. Summary These findings capture additional important dimensions in discourse compared to prior studies of aphasia (for instance intricacy, representing adverb and adjective usage). Our lesion findings are partly consistent with prior studies associating the precentral gyrus with speech quantity and the basal ganglia and surrounding white matter with motor speech deficits in narrative speech. The subcortical lesion results suggest involvement of white matter pathways in a discourse network. Future steps will be to explore structural connectivity related to our PCA findings.
Topic Areas: Disorders: Acquired, Language Production