Classification of Mean Arterial Pressure Regimes in ICU Using a Model-Based Support Vector Machine: Acute Hypotensive, Critical and Survival Episodes

Abstract

In this study, a new pattern discrimination method for the classi cation of Mean Arterial Pressure (MAP) regimes in ICU via an appropriately regulated Radial Basis Function (RBF) Support Vector Machine (SVM) is described. The aim of this classi cation is to detect hazardous cardiogenic shock situations to prevent probable fatal failure of organs. To this end, rst, electrocardiogram (ECG) and Blood Pressure (BP) waveforms are processed via a Modi ed Hilbert Transform (MHT), and QRS complexes (equivalently obtaining heart rate-HR trend) and pressure pulses (equivalently obtaining trends of systolic, diastolic and mean arterial pressures) are detected, respectively. In the next step, a RBFSVM classi er is tuned using features obtained from the cardiogenic shock risk scoring model developed by Hasdai et al. (2000) to classify MAP regimes into three categories; survival (the status that will not fall into shock), critical (the transient status that may lead to shock or a return to the survival episode) and Acute Hypotensive Episode -AHE (meaning cardiogenic shock will certainly occur.) Then, the regulated RBF-SVM classi er is applied to 60 records of the Computers in Cardiology (CinC) Challenge 2009 and the values of Se = 92% and P+ = 93% are obtained for sensitivity and positive predictivity, respectively. As some results of this study, the proposed classi cation method recognized truly 15 subjects out of 15 normal (without shock episodes) subjects of the MIMICII database as belonging to the survival class", while the algorithm could classify 24 subjects as AHE", 3 subjects as of the critical class" and 3 subjects as in the survival" situation out of 30 shock containing records of the MIMICII database.