Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MEET-Sepsis: Multi-Endogenous-View Enhanced Time-Series Representation Learning for Early Sepsis Prediction

Zexi Tan, Tao Xie, Binbin Sun, Xiang Zhang, Yiqun Zhang, Yiu-Ming Cheung

TL;DR

Sepsis prediction in ICU is hindered by subtle early signals and sparse time-series data. MEET-Sepsis introduces a dual-component framework: MERE, which generates multiple endogenous feature views, and CDTA, which performs multi-scale time-series representation learning with self-attention. The model is trained with a reconstruction plus prediction objective, enabling robust early predictions while maintaining interpretability, and demonstrates superior performance on PhysioNet/CinC 2019 qSOFA and SOFA datasets, particularly in early ESP windows. This approach promises faster, reliable early intervention opportunities and shows potential for extension to federated and multi-modal medical data settings.

Abstract

Sepsis is a life-threatening infectious syndrome associated with high mortality in intensive care units (ICUs). Early and accurate sepsis prediction (SP) is critical for timely intervention, yet remains challenging due to subtle early manifestations and rapidly escalating mortality. While AI has improved SP efficiency, existing methods struggle to capture weak early temporal signals. This paper introduces a Multi-Endogenous-view Representation Enhancement (MERE) mechanism to construct enriched feature views, coupled with a Cascaded Dual-convolution Time-series Attention (CDTA) module for multi-scale temporal representation learning. The proposed MEET-Sepsis framework achieves competitive prediction accuracy using only 20% of the ICU monitoring time required by SOTA methods, significantly advancing early SP. Extensive validation confirms its efficacy. Code is available at: https://github.com/yueliangy/MEET-Sepsis.

MEET-Sepsis: Multi-Endogenous-View Enhanced Time-Series Representation Learning for Early Sepsis Prediction

TL;DR

Sepsis prediction in ICU is hindered by subtle early signals and sparse time-series data. MEET-Sepsis introduces a dual-component framework: MERE, which generates multiple endogenous feature views, and CDTA, which performs multi-scale time-series representation learning with self-attention. The model is trained with a reconstruction plus prediction objective, enabling robust early predictions while maintaining interpretability, and demonstrates superior performance on PhysioNet/CinC 2019 qSOFA and SOFA datasets, particularly in early ESP windows. This approach promises faster, reliable early intervention opportunities and shows potential for extension to federated and multi-modal medical data settings.

Abstract

Sepsis is a life-threatening infectious syndrome associated with high mortality in intensive care units (ICUs). Early and accurate sepsis prediction (SP) is critical for timely intervention, yet remains challenging due to subtle early manifestations and rapidly escalating mortality. While AI has improved SP efficiency, existing methods struggle to capture weak early temporal signals. This paper introduces a Multi-Endogenous-view Representation Enhancement (MERE) mechanism to construct enriched feature views, coupled with a Cascaded Dual-convolution Time-series Attention (CDTA) module for multi-scale temporal representation learning. The proposed MEET-Sepsis framework achieves competitive prediction accuracy using only 20% of the ICU monitoring time required by SOTA methods, significantly advancing early SP. Extensive validation confirms its efficacy. Code is available at: https://github.com/yueliangy/MEET-Sepsis.
Paper Structure (12 sections, 5 equations, 3 figures, 1 table)

This paper contains 12 sections, 5 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Proposed Method
  3. MERE: Multi-Endogenous-View Representation Enhancement.
  4. CDTA: Cascaded Dual-Convolution Time-Series Attention.
  5. MEET-Sepsis: Model Training.
  6. Experiments
  7. Experimental Settings.
  8. Prediction Performance Evaluation.
  9. Ablation Study.
  10. Discussions.
  11. Concluding Remarks
  12. Acknowledgement

Figures (3)

  • Figure 1: Distribution of ICU patients in the class of "Sepsis" and "Non-Sepsis" over a 24-hour period. In the early stages (0 - 12h), the distribution difference of the two classes is subtle, while the death rate of the "Sepsis" class increases sharply.
  • Figure 2: Overview of the MEET-Sepsis framework.
  • Figure 3: MEET-Sepsis performance on SOFA (upper) and qSOFA (lower) datasets. The time slots when MEET-Sepsis ranked 1st are marked with red circles.