Determination of cardiopulmonary resuscitation quality based on machine learning algorithms using various biological...

Carotid blood flow (CBF) is a critical indicator during cardiopulmonary resuscitation (CPR), representing blood flow to the brain. In actual clinical settings, measuring it is almost impossible. In this study, we developed and evaluated machine learning models that estimate CBF using biological signals that can be measured during CPR.

To simulate various compression conditions in animal experiments, we employed robot manipulators and a commercial automated CPR machine. Five biological signals-arterial blood pressure, central venous pressure, photoplethysmography, intrathoracic pressure, end tidal carbon dioxide (ETCO2)-and CBF were measured. Various features were extracted from these biological signals to estimate CBF. Machine learning models were trained on various combinations of these features.

The models achieved a classification accuracy of up to 0.97, determining whether the CBF was above or below a predetermined threshold. Furthermore, the most influential biological signal highly correlated with CBF was identified as ETCO2.

This machine learning model holds great promise for improving treatment efficacy by assessing and providing feedback on CPR quality in clinical settings.

The online version contains supplementary material available at 10.1007/s13755-025-00419-6.