Enhancing the performance of multi-parameter patient monitors by homogeneous kernel maps

Abstract

A multi-parameter patient monitor (MPM) is a crucial appliance utilized in life-threatening care units of hospitals. It smells out a patient’s vital signs without the demand of continuous attendance by the nurses. The operation of the SVM algorithm closely depends on the kernel mapping and their corresponding parameters. MPM stays informed regarding the state of a patient utilizing the baseline vital parameters, heart rate (HR), blood pressure (NIBP), respiration rate (RR) and oxygen saturation (SPO2). A high exactness for sensitivity, specificity and overall classification is critical in giving quality social insurance to the patients. Support Vector Machine (SVM) is an influential classification technique that is successfully used in the improvement of MPMs. A cautious investigation of the baseline parameters uncovers that they are constantly positive, making them suitable to be used with a class of kernels, such as intersection, Chi-Squared, and Jenson-Shannon kernels that are previously used effectively with histograms of images for scene classification. In our baseline system, we have utilized 10 features (four vital parameters along with six correlated features) based on our earlier research using decision tree algorithms. Using the intersection kernel, we obtained an improvement of 2.14 % in sensitivity, 1.38 % in specificity, and 1.57 % in the overall classification accuracy, over the best baseline system using radial basis function (RBF) kernel. In this work, we experiment with the histogram kernels and showed that intersection kernel can effectively improve the sensitivity, specificity, and overall classification accuracy of MPMs.