Publication

Abstract : pMulti-parameter patient monitors (MPMs) are extensively used in health care to monitor the condition of patients. It is very important to use MPMs with good classification accuracy, sensitivity and specificity to identify the patients health deterioration at the right time. The four vital parameters that MPM uses are heart rate, respiration rate, oxygen saturation and blood pressure. It is well known that, there exists an intrinsic relationship between vital signs of a healthy person. We captured this intrinsic relationship using correlation features (CFs). We then developed a baseline system with four vital signs as input to SVM back-end classifier and then checked the performance of MPM with CFs as input. It was seen that the performance of the MPM with CFs as it's input outperforms the baseline system. However combining the vital signs with the CFs to make the number of input features of MPM to ten were shown to deteriorate the MPM performance. We then fused the output scores of baseline system and CF system. The fused system obtained a performance of 0.58% absolute in classification accuracy, 0.68% absolute in sensitivity and 0.56% absolute in specificity with respect to the baseline system. MIMIC II database was used for all experiments discussed in this work and SVM back-end uses RBF Kernel. © 2017 IEEE./p