Early Detection of Sepsis Utilizing Self-guided Generative Adversarial Network

Sepsis is a severe illness that can potentially lead to death. According to statistics from the World Health Organization, approximately 6 million people die each year due to septic shock, with a high mortality rate of up to 50%. Early warning and timely intervention in the early stages of sepsis can prevent the occurrence of septic shock in the vast majority of cases. This paper presents a novel sepsis early warning system composed of a fully connected neural network, bidirectional long short-term memory, convolutional neural network, attention mechanism, cooperative network, and generative adversarial network. Among these components, the convolutional neural network can extract local features independent of position, bidirectional long short-term memory can capture time-dependent features of dynamic information using forward and backward operations, the fully connected neural network is suitable for analyzing key features within static information, and the attention mechanism can focus more on important features that enhance prediction performance. Additionally, the self-adjusting dual-channel module composed of collaborative networks and generative adversarial networks can generate random noise and enhance the robustness and generalization of features. In the experimental section, we validate our proposed method using the PhysioNet/Computing in Cardiology Challenge 2019 sepsis dataset, and the results confirm that our method outperforms other advanced sepsis prediction algorithms in terms of predictive performance.