Intelligent Sepsis Detector Using Vital Signs Through Long Short-Term Memory Network

Abstract

Sepsis has become a primary source of mortality of patients treated in intensive care units. The timely detection of sepsis assists in decreasing the mortality rate, as it becomes difficult to treat the patient if the symptoms get worsen. The primary objective of this work is to early detect sepsis patients by utilizing a deep learning model, and then perform a comparative analysis of the proposed system with other modern techniques to analyze the performance of the proposed model. In this work, we employed the long short-time memory model on the sepsis patient dataset. The three different performance metrics are used to evaluate the performance of the proposed system, i.e., accuracy, specificity, and AUROC. The results were obtained in three different windows after the patient was admitted to the intensive care unit, such as 4, 8, and 12 h window sizes. This proposed system achieved accuracy, specificity, and AUROC of 77, 75, and 91%, respectively. The comparison of the proposed system with other state-of-the-art techniques is performed on the basis of the above-mentioned performance metrics, which demonstrated the significance of the proposed system and proved that this system is reliable to implement in real-time environments.