Hyperparameter Tuning of the Shunt-murmur Discrimination Algorithm Using Bayesian optimization

Patients undergoing hemodialysis generally have shunts implanted in their bodies; a number of other problems, such as vascular stenosis, can be encountered. Patients undergoing hemodialysis can inspect the effective functioning of their shunts by listening to the shunt murmur. However, this manual inspection is difficult and requires experience. In this paper, we propose a method of exploring the hyperparameters of the shunt-murmur discrimination algorithm using Bayesian optimization. The resistance index(RI) obtained from the ultrasound system is used as a class label. The normalized crosscorrelation coefficients, Mel frequency cepstrum coefficients, and frequency power percentage were the features to be trained by a random forest (RF). Bayesian optimization was used to explore the hyperparameters of the RF, achieving a significant accuracy improvement.