A new density-ratio based approach for patient-specific biomedical monitoring

In order to denote the abnormalities of the patients, we propose a novel approach to detect anomaly in biomedical monitoring using density ratio values as the Patient Status Index (PSI). The key idea of the proposed method is to define the ratio of training and testing data densities, where training dataset only consist of normal data and testing dataset consist of both normal and abnormal data, and identify irregular samples for testing patients' dataset. Furthermore, we define four inequalities to denote the interval values of density ratio and give the corresponding status for patients. In addition, the applied Kullback-Leibler based algorithm for calculating density ratio values without involving density estimation is equipped with a cross validation (CV) model selection procedure, allowing us to objectively optimize values of tuning parameters. We select training and testing data from Physionet database to do our pilot experiment. The experimental results for 11901 beats show that the density-ratio based approach work very well in terms of specificity and sensitivity.