Deep learning-based ROSC prediction and ECG phenotyping in out-of-hospital cardiac arrest.
Resuscitation · 2026-03-06 · Method & validation
Abstract
INTRODUCTION: Electrocardiogram (ECG) signals during cardiac arrest contain detailed information on cardiac rhythm characteristics and have been associated with resuscitation outcomes. We aimed to develop a convolutional neural network (CNN) model to predict return of spontaneous circulation (ROSC) and identify novel ECG phenotypes in patients with out-of-hospital cardiac arrest (OHCA). METHODS: This retrospective cohort study used Korean OHCA Registry and ECG data from Seoul emergency medical services between July 2021 and December 2023. Adult patients with nontraumatic OHCA who had prehospital ECG signals were included. Five-second ECG segments obtained during resuscitation were analyzed. A one-dimensional CNN was developed to simultaneously predict the probability of ROSC within 2 min and to classify the concurrent ECG rhythm as shockable or non-shockable. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC). ECG phenotypes were identified by applying K-means clustering to feature representations extracted from the trained CNN. RESULTS: 3452 patients (median age, 73 years [interquartile range, 61-81]; 34.4% female) were included. The AUCs for predicting ROSC within 2 min and shockable rhythm in the test set were 0.921 (95% confidence interval [CI], 0.897-0.945) and 0.983 (95% CI, 0.979-0.988), respectively. Five ECG phenotype categories with varying morphologies and ROSC probabilities were identified. Phenotype 1 consisted of shockable rhythms with high ROSC probability (30.4%), whereas Phenotype 2 showed shockable rhythms with low ROSC probability (4.8%). Phenotypes 3 and 4 included pulseless electrical activities with relatively high (5.2%) and lower (0.5%) ROSC probabilities, respectively. Phenotype 5 primarily consisted of asystole or near-asystole rhythms. Transitions between ECG phenotypes were associated with CPR quality. CONCLUSIONS: The CNN-based model accurately predicted ROSC and shockable rhythm from ECG signals and identified five novel ECG phenotypes in OHCA. These findings can enable accurate ROSC prediction and individualized prehospital resuscitation for patients with OHCA.
Abstract © the authors or publisher, via PubMed. Reuse and removal
Tags
Decision support tools · Quality of life & function
Related articles
- Comparative performance of machine learning and conventional scoring systems for neuroprognostication in out-of-hospital cardiac arrest survivors. (J Formos Med Assoc, 2025)
- Prediction of neurologic outcome after out-of-hospital cardiac arrest: An interpretable approach with machine learning. (Resuscitation, 2024)
- Overcoming data scarcity in life-threatening arrhythmia detection through transfer learning. (Commun Med (Lond), 2025)
- A Machine Learning-Based Decision Support System for the Prognostication of Neurological Outcomes in Successfully Resuscitated Out-of-Hospital Cardiac Arrest Patients. (J Clin Med, 2024)
- Severity of metabolic derangement predicts survival after out-of-hospital cardiac arrest and the likelihood of benefiting from extracorporeal life support. (Emergencias, 2024)
- Derivation and validation of a blood biomarker score for 2-day mortality prediction from prehospital care: a multicenter, cohort, EMS-based study. (Intern Emerg Med, 2023)