Modeling mechanical properties of the chest during the cardiopulmonary resuscitation procedure

The first step in improving the cardiopulmonary resuscitation (CPR) procedure is to understand how the force applied on the chest relates to the resulting blood pressure. In order to capture the mechanical properties of the chest and abdomen accurately, in this paper we proposed a nonlinear mass spring and damper model which consists of nonlinear spring up to 5th order, a combination of viscous and hysteresis damper. We then nondimensionalized the proposed model in order to enhance parametric study of the model for future use. In the next step we used gradient decent optimization method to identify the model parameters for force-compression data of 10273 CPR cycles collected from different pigs at the Children's Hospital of Philadelphia (CHOP). We used the mean square error (RMSE) between the estimated force and actual force for each cycle as an objective function to be minimized. Using the above method we were able to estimate the model parameters for each cycle seperately. In order to find a best set of estimated parameters we used K-nearest neighbor (K-NN) method. K-NN is an unsupervised learning technique which clusters the data into different groups based on the distance metric. The cluster center with lowest RMSE is selected as the estimated parameters for the entire cycles. The resulted MSE of testing entire cycles with the estimated parameters are 0.12 mean and 0.04 SD. Results show that the proposed model is the most accurate model of pig chest during the CPR.