An End-To-End 1D-ResCNN Model For Improving The Performance Of Multi-parameter Patient Monitors

Abstract

Multi-parameter patient monitors (MPMs) are widely used medical devices for continuous observation of a patient’s physiological conditions in a hospital. Early warning score (EWS) is an existing system used in monitors that have low accuracy. Hence, the monitors’ performance must be improved to generate meaningful alarms. In this work, we have used a Residual neural network (ResNet) along with bottleneck features extracted from convolutional neural networks (CNNs) to improve the alarm accuracy. The accuracy, sensitivity, and specificity of MPMs can be improved by capturing the intrinsic relationship between the vital parameters which is achieved by using different kernels. Thus, the overall performance of the ResNet model is noted to be 98.43% of sensitivity, 99.96% of specificity, and 99.60% of overall performance accuracy. Compared to the baseline system, the proposed system has a performance improvement of 0.16% (sensitivity) alarm accuracy, 0.18% (specificity)no-alarm accuracy, and 0.17% overall accuracy