Publication

Abstract : Predicting the physiological condition (normal/abnormal) of a patient is highly desirable to enhance the quality of health care. Multi-parameter patient monitors (MPMs) usingnbsp;heartrate, arterial blood pressure,nbsp;respirationnbsp;rate and oxygen saturation (S pO2) as input parameters were developed to monitor the condition of patients, with minimum human resource utilization. The Support vector machine (SVM), an advancednbsp;machine learningnbsp;approach popularly used for classification and regression is used for the realization of MPMs. For making MPMs cost effective, we experiment on the hardware implementation of the MPM using support vector machine classifier. The training of the system is done using the matlab environment and the detection of the alarm/noalarm condition is implemented in hardware. We used different kernels for SVM classification and note that the best performance was obtained using intersection kernel SVM (IKSVM). The intersection kernel support vector machine classifier MPM has outperformed the best known MPM using radial basis function kernel by an absoute improvement of 2.74% in accuracy, 1.86% in sensitivity and 3.01% in specificity. The hardware model was developed based on the improved performance system using Verilog Hardware Description Language and was implemented on Altera cyclone-II development board.