Deep Learning in Early Prediction of Sepsis and Diagnosis

Abstract

Identifying sepsis early can help prevent further complications by identifying possible risks. The detection of sepsis early on, With support vector machines (SVM) and Long Short-Term Memory (LSTM), We integrated a recurrent neural network to derive the results. An infection can cause sepsis, a life-threatening disorder that results in organ failure, tissue destruction, and death. Approximately 30 million people will get sepsis every year, and one-fifth of them will die from it. Sepsis can often be prevented by early detection and prompt treatment. Analyze patient data from the intensive care unit to predict whether or not they have Sepsis Disease using a deep neural network. This study aims to detect sepsis early by using physiological data. As inputs, vital signs, laboratory results, and demographic information are collected from patients. Choosing the most appropriate hyperparameters for the training phase and the probability threshold for inference, we used an LSTM and an SVM. Our study proposes a model based on LSTM and SVM to predict sepsis among ICU patients. By utilizing RF and LR approaches as well as LSTM networks, we created a data pipeline for cleaning and processing data. AUC-ROC score of 0.696 was achieved.

Ähnliche Arbeiten