Severe Patients Shock Detection Model Based on Convolutional Neural Networks

Medical research has shown that the prolongation of Capillary Refilling Time (CRT) is significantly associated with the shock caused by various pathogenesis, and the shock detected in critically ill patients as earlier as possible has important clinical significance. Utilizing capillary refill time measuring instruments, we can implement standardized and mechanized measurement of CRT, and get stable and reliable photoplethysmogram (PPG) raw signal data of CRT. However, there are many challenges in extracting clinical parameters from PPG raw signal data and analyzing the shock occurs or not depending on these parameters, such as great affection by noise, poor ability to cope with variations, and underutilization of raw data. To cope with these challenges, we propose an end-to-end shock detection model based on Convolutional Neural Networks (CNNs). In this paper, the convolutional kernel of dense neural network and mathematical network are proposed innovatively. Our shock detection model achieves 0.89 F1-Score using clinical CRT data, which can provide direct and exact medical verdict indicating the patient be in shock or not, which has clinical value.