Poster Presentation

Search Abstracts | Symposia | Slide Sessions | Poster Sessions

Neural–behavioral dissociations in cross-task lesion–symptom mapping of naming errors in chronic aphasia

Poster Session A, Wednesday, September 30, 11:00 am - 1:00 pm, Wangari Maathai

Kalil Warren¹, Roger D. Newman-Norlund², Seyed Saeed Ahmadi², Nadra Salman³, Yong Yang⁴, Regan Willis⁴, Xiang Guan⁴, Leonardo Bonilha⁵, Julius Fridriksson²; ¹Psychology Department, University of South Carolina, ²Communication Sciences and Disorders Department, University of South Carolina, ³Linguistics Program, University of South Carolina, ⁴Computer Science and Engineering Department, University of South Carolina, ⁵Neurology Department, USC School of Medicine

Introduction. Spoken naming errors in aphasia are widely used as windows into the language production system, and lesion–symptom mapping (LSM) has linked specific error types to damage in distinct brain regions. However, it remains unclear whether the lesion correlates of a given error type generalize across tasks, or whether neural and behavioral indices of cross-task stability necessarily align. We addressed both questions by comparing lesion–error associations across two tasks with shared retrieval demands but differing contextual constraints — confrontation naming (Philadelphia Naming Test; PNT) and sentence completion (Western Aphasia Battery–Revised; WAB-R) — in one of the largest cohorts assembled for error-type LSM in chronic aphasia. Methods. Participants were individuals with chronic post-stroke aphasia from a single left-hemisphere ischemic stroke (N = 214 PNT; N = 212 WAB-R; N = 183 with both). All spoken responses were transcribed at the item level and classified by expert consensus coding into seven mutually exclusive categories (correct, semantic, phonemic, mixed, neologism, unrelated, no-response) using a unified, theoretically grounded taxonomy. Lesion masks were manually delineated on T2-weighted MRI, normalized to MNI space via enantiomorphic normalization, and intersected with the JHU atlas to yield lesion load values for 24 left-hemisphere ROIs spanning dorsal-stream, ventral-stream, insular, and major white-matter structures. For each error type and task, mass-univariate OLS regressions with permutation testing (4,000 permutations) and FDR correction identified ROI-level lesion–error associations. Cross-task generalizability was then assessed at three complementary levels: (1) ROI-level neural correspondence (Spearman ρ of ROI rankings by absolute t-statistic across tasks); (2) within-subject behavioral consistency (Spearman ρ of error proportions across tasks); and (3) cross-task prediction via ridge regression trained on one task and evaluated on the other. Results. Cross-task ROI correspondence revealed a clear gradient: correct, phonemic, and semantic responses showed strongly conserved neural correlates across tasks (ρ = 0.57–0.70, all p ≤ .003), whereas mixed, neologism, and unrelated errors showed weak, non-significant correspondence (ρ = 0.11–0.30). A group-level permutation test confirmed the distinction (Δ = 0.48, p = .0496), and pairwise Steiger tests yielded 9 of 12 significant comparisons across groups. Two informative neural–behavioral dissociations emerged. Semantic errors showed strong cross-task neural correspondence (ρ = 0.57) but the weakest significant within-subject behavioral consistency (ρ = 0.17, p = .022), indicating that ventral-stream regions are recruited in a task-general fashion while individual error rates vary substantially between tasks. Unrelated errors showed the complementary pattern: significant behavioral consistency (ρ = 0.37, p < .001) alongside non-significant neural correspondence (ρ = 0.30, p = .15), suggesting a patient-level trait whose neural expression is partially modulated by task demands. Conclusion. Cross-task stability of naming-error correlates differs systematically across error types and across levels of analysis. Neural and behavioral indices of generalizability can diverge in opposite directions for different error categories — a finding that constrains both clinical interpretation of error profiles and neurobiological models of lexical access.

Topic Areas: Disorders: Acquired, Speech-Language Treatment