![](https://congressarchive.cimne.com/m2p_2023/assets/a4.png)
Identification of vital parameters during cardiopulmonary resuscitation
Please login to view abstract download link
Performing quick and reliable pulse checks during cardiopulmonary resuscitation (CPR) is a crucial, but difficult task. The current error-prone technique for the detection of a return of spontaneous circulation (ROSC) is the manual pulse palpation at some artery, augmented by end-tidal CO2-concentrations (etCO2). In this work, we try to determine the presence and the strength of cardiac output during cardiac arrest by a machine-learning based and a model-based approach. Accelerometers used as feedback devices during CPR are capable of detecting the apex beat of the heart against the chest wall. Employing this accelerometry signal and the ECG signal from 422 manually annotated defibrillator records from the German Resuscitation Registry, we trained a Support Vector Machine on a part of the data set and evaluated the predictions of the circulatory state on a test data set. To estimate the generalizability of the classifier we repeated this process on 50 different patient-wise split data sets. In mean (95% confidence interval), such an algorithm shows a balanced accuracy of 81.2, (74.9,87.5) %, a sensitivity of 80.6, (68.0, 93.0) %, a specificity of 81.8, (73.5, 90.1) % and a MCC of 0.614, (0.479,0.748) for classifying the correct circulatory state, keeping up with the best available pulse detection algorithms so far. This algorithm could be used as a valuable medical decision tool for cardiac arrest patients in the field. The usage of CO2 to determine cardiac output is the basis for the model-based approach. In presence of circulation the metabolic CO2 is transported from the tissue to the lungs and expired through the alveoli. Thus, increasing etCO2 levels indicate a strengthening circulation during cardiac arrest treatment, although other, practically uncontrollable variables like tidal volumes influence etCO2 -concentration too. Thus, we try to model these concepts in a simple ODE-model: With data from a porcine model, we try to estimate the parameters, including desired circulation strength q using Kalman filtering or a multishoot approach and compare the result to the measured cardiac activity. Due to the structure of the ODE model, we can split the model in the volume model, and the concentration model. So far, we only analyze the volume model, where we want to estimate Resistance,Compliance and some coefficients describing the intrathoracic pressure and use those to estimate the parameters in the concentration system afterwards.