Related Experiment Video
Updated: Sep 2, 2026
Point-of-Care Ultrasound for Peripheral Veno-Arterial Extracorporeal Membrane Oxygenation Without Left Ventricular Venting
Published on: January 17, 2025
A Review of the Role of Artificial Intelligence in Patients on Extracorporeal Membrane Oxygenation: A Scoping Review
Zahra Asadi1, Mahmoud Shiri Kahnouei2,3, Fathiyeh Bahramnejad3
1Department of ICU Nursing, School of Nursing & Midwifery Tehran University of Medical Sciences Tehran Iran.
Background And Aims:
ECMO is used to deliver cardiopulmonary support in extreme failure where standard therapies are unfruitful. Although ECMO is associated with high costs and significant risks, advances in technology and clinical management have improved its safety and patient outcomes. AI/ML may provide decision support through data analysis to alert clinicians to potential safety issues, detect complications, and help inform pump/ventilator settings. The primary objective of this scoping review is to examine the applications of AI and machine learning in patients receiving various ECMO modalities, specifically veno-arterial (VA), veno-venous (VV), and extracorporeal cardiopulmonary resuscitation (ECPR).
Methods:
We carried out a scoping review based on Arksey and O'Malley, encompassing 20 inaugural AI/ML ECMO studies (no review/theoretical/non-English). The databases were searched till December 2025. Expressed data covered study characteristics, AI methods, inputs, objectives, performance, and limitations. Algorithm applications were categorized based on patient characteristics and clinical use (selection, timing, monitoring, weaning, and readmission). The findings were expert-validated to identify trends and gaps.
Results:
Across studies (n ≈85,509 ECMO patients; mean per study ≈4275), AI applications were identified across selection, timing, monitoring, weaning, and readmission. Models (Random Forest, XG Boost, Light GBM, deep nets) Reported AUROCs ranged from approximately 0.70 to 1.00 for ECMO need prediction, neurological outcome, mortality, and bleeding. Imaging data were underutilized in most studies; most studies were retrospective in design. While predictive performance has been reported, direct hard outcome improvement wasn't consistently shown; AI primarily functioned as decision support.
Conclusion:
AI/ML are promising for predicting ECMO needs and outcomes and may support clinical decision-making, but retrospective data and limited prospective validation temper conclusions. Prospective and real‑world validation studies are required to demonstrate clinical outcome benefits and support the integration of AI into ECMO practice and guidelines.
