Automated Cerebral Edema Detection using Electroencephalography in Post-cardiac Arrest Patients.

A research model found EEG patterns associated with diffuse brain swelling after cardiac arrest, including before radiographic recognition in a small matched analysis. It is an early-warning research signal, not a validated diagnostic or prognosis tool.

Stafford RA, Srivastava V, Altaf AZ, Tang L, Sachdeva AS, Elman D, Mallinger LA, Reinert AL, Madipally SSK, Salvati LR, Wahlster S, Chen D, Gilmore EJ, Greer DM, Cheng H, Ong CJ.
Neurocritical care · Publication date: 2026-09-28
Added to site: 2026-09-30
DOI: 10.1007/s12028-026-02668-z

Electronic publication date verified in Europe PMC. The record was indexed after the prior portal search, and this source date is separate from the site's addition date.

Modalities

EEG, AI

Clinical focus

Cardiac arrest, Cerebral edema, Clinical decision support

Population

Adults resuscitated from cardiac arrest who underwent routine neuroimaging and EEG monitoring from 2016 through 2024

Sample

124 patients in the detection cohort; 19 edema cases and 19 matched referents in the prediction analysis

Intervention / method

Transformer and LSTM models analyzed four- to eight-hour EEG segments obtained more than 24 hours after arrest to detect or predict diffuse cerebral edema

Comparator

Radiographic edema status; the prediction analysis matched referents by age, sex, witnessed arrest and EEG timing

Clinician summary

The best four-hour transformer detection model reported median AUC 83.5%, 75% accuracy, 80% sensitivity and 70% specificity. In 19 edema cases and 19 matched referents, the best eight-hour prediction model reported median AUC 80%, 70% accuracy, 80% sensitivity and 60% specificity.

How to interpret the findings

The retrospective single-center study used routine-care data and internal evaluation. The 38-person matched prediction analysis was small, specificity was 60%, and external validation, calibration, decision thresholds, prospective workflow effects and outcome benefit remain untested.

Equipment and disclosures

Routine clinical EEG and neuroimaging records were analyzed; acquisition hardware, channel configuration and transformer implementation details were not available in the checked abstract.

The indexed record reports American Heart Association and NIH/NINDS funding. Competing-interest statements were not available in the checked abstract.

Related evidence

Compare findings from the same clinical area, including mixed results.