Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Inverse symptom–lesion mapping in human aphasia: Predicting left-hemisphere damage location from behavioral phenotype
Poster Session D, Thursday, October 1, 4:30 - 6:30 pm, Wangari Maathai
Roger Newman-Norlund1, Kalil Warren2, Saeed Ahmadi1, Nadra Salman2, Yong Yang4, Regan Willis4, Xiang Guan4, Leonardo Bonilha3, Julius Fridriksson1; 1University of South Carolina, Department of Communication Sciences and Disorders, 2University of South Carolina, Linguistics Program, 3University of South Carolina, School of Medicine, Department of Neurology, 4University of South Carolina, Department of Computer Science and Engineering
Introduction. For more than 160 years, since Broca and Wernicke, lesion–symptom mapping (LSM) has inferred brain function by running from damage to deficit. The inverse direction, symptom–lesion mapping (SLM), in which lesion location is predicted from behavior alone, has remained a theoretical aspiration rather than a working method. Two obstacles have stood in the way: naturally occurring stroke follows vascular territories rather than cytoarchitectonic boundaries, and typical cohorts contain too few participants with damage in any given atlas region to support stable inverse estimation. Here we present, to our knowledge, the first operational human SLM, built on the Center for the Study of Aphasia Recovery (CSTAR) cohort, one of the largest and most deeply phenotyped chronic aphasia databases assembled to date. Participants and measures. We analyzed 296 chronic post-stroke participants (≥6 months post-onset; unilateral left-hemisphere or subcortical stroke). Lesions were MRI-confirmed and segmented into the 189-region JHU atlas. The behavioral phenotype comprised roughly 200 features: Western Aphasia Battery subscores and Aphasia Quotient; Philadelphia Naming Test and WAB sentence-repetition error proportions across six categories; NIHSS items; WAIS; Boston Naming Test; Apraxia Battery; Pyramids and Palm Trees; discourse measures; demographics; and five aphasia-syndrome composites. Modeling. A multi-task Elastic Net jointly predicted per-ROI lesion load across the 64 left-hemisphere ROIs passing a 10% damage-prevalence cutoff. Evaluation used leave-one-out cross-validation (296 fits, each holding out a single patient). Results. Of 64 LH ROIs, 40 achieved leave-one-out R² > 0.1, 27 exceeded 0.2, and 16 exceeded 0.3. The dorsal language stream and its supporting white matter led: superior longitudinal fasciculus (R² = 0.47; AUC = 0.89), precentral gyrus (R² = 0.40; AUC = 0.83), posterior insula (R² = 0.40; AUC = 0.85), retrolenticular internal capsule (R² = 0.39; AUC = 0.85), posterior superior temporal gyrus (R² = 0.37; AUC = 0.86), superior temporal gyrus (R² = 0.36; AUC = 0.88), posterior middle temporal gyrus (R² = 0.34), and IFG pars opercularis (R² = 0.34; AUC = 0.85). Predictable ROIs clustered in classical perisylvian language cortex and underlying white matter, while visual, limbic, and memory regions remained at chance, an implicit negative control indicating that the model tracks language-specific signal rather than global lesion burden. Elastic Net coefficients showed a clean anterior/posterior dissociation: anterior-cortex ROIs loaded on WAB fluency and a Broca composite; posterior-cortex ROIs loaded on comprehension, repetition, and Wernicke/Anomic composites; motor ROIs additionally drew on NIHSS right-arm motor scores. Total LH lesion volume was predicted at r = 0.72. Conclusions and clinical implications. Human SLM is tractable for the classical left perisylvian language network. For the patient sitting in a rural clinic, in an acute stroke bay, or anywhere neuroimaging is intractable, this means that a trained clinician with a behavioral battery can now reason quantitatively about where the damage sits, and where to aim stimulation. Aphasiology has had a forward map for 160 years. It finally has an inverse.
Topic Areas: Disorders: Acquired, Computational Approaches