Abstract Objective Neurologic complications are a leading cause of morbidity and mortality in children supported with extracorporeal membrane oxygenation (ECMO). Antemortem neuroimaging is limited, while postmortem neuropathology provides definitive but rarely available assessment. Correlating imaging with pathology may improve diagnostic accuracy and monitoring strategies. Design We performed a retrospective, single-center cohort study of neonates and children (0–18 years) supported with ECMO (2009–2024) who underwent cerebral autopsy. Standardized neuropathologic sampling across six brain regions classified injuries as mild, moderate, or severe. Antemortem imaging (computed tomography CT or head ultrasound HUS) within 7 days of death was reviewed by a blinded neuroradiologist and scored using a Neurologic Injury Severity (NIS) framework. Imaging–pathology correlation was assessed in patients with both datasets. Measurements and Main Results A total of 54 patients underwent neuropathologic evaluation; 26 had qualifying neuroimaging. Neuropathology revealed frequent ischemic and hemorrhagic injuries, most commonly in the frontal lobes, deep grey matter, and temporal lobes. Severe injury was more frequent in venoarterial ECMO and in infants/toddlers. NIS scores strongly correlated with global and regional histopathologic severity, particularly in frontal, temporal, and deep grey regions. Discordance was observed in parieto-occipital, pontine, and cerebellar regions, where immunohistochemistry (GFAP, CD68) detected subtle injuries not visible with hematoxylin and eosin (H&E). Severe pathology consistently corresponded to NIS ≥ 10, while mild pathology aligned with low scores and limited involvement. Conclusions Structured imaging severity scoring, especially with CT, correlates strongly with neuropathology in pediatric decedents of ECMO. Findings support NIS scoring as a surrogate for underlying pathology, while underscoring the need for refined histologic methods and adjunctive neuromonitoring to optimize neurologic surveillance.
Tabet et al. (Tue,) studied this question.