Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Early Prediction of Sepsis using Heart Rate Signals and Genetic Optimized LSTM Algorithm

Alireza Rafiei, Farshid Hajati, Alireza Rezaee, Amirhossien Panahi, Shahadat Uddin

TL;DR

This work tackles the challenge of predicting sepsis onset outside hospital wards by using only heart-rate signals from wearable-capable devices. It evaluates four machine learning architectures (LGB, MLP, LSTM, LSTM-FCN) whose structures are optimized with a genetic algorithm, then extends the one-hour predictions to four hours via transfer learning. Using the PhysioNet 2019 dataset, the authors balance a heart-rate–only feature set through forward filling and noise augmentation to train the models on 12-hour windows around sepsis onset. Results show the LSTM achieves the highest accuracy (AUROCs around $0.96$ for 1 hour and $0.92$ for 4 hours) but with substantial model size, while LGB offers a compact alternative with competitive AUROCs; overall, the study demonstrates feasibility for real-time wearable sepsis warning systems and highlights trade-offs between accuracy, latency, and memory. The findings point to potential for real-world deployment, with future work needed to incorporate non-ICU data and additional features for robust out-of-hospital sepsis detection.

Abstract

Sepsis, characterized by a dysregulated immune response to infection, results in significant mortality, morbidity, and healthcare costs. The timely prediction of sepsis progression is crucial for reducing adverse outcomes through early intervention. Despite the development of numerous models for Intensive Care Unit (ICU) patients, there remains a notable gap in approaches for the early detection of sepsis in non-ward settings. This research introduces and evaluates four novel machine learning algorithms designed for predicting the onset of sepsis on wearable devices by analyzing heart rate data. The architecture of these models was refined through a genetic algorithm, optimizing for performance, computational complexity, and memory requirements. Performance metrics were subsequently extracted for each model to evaluate their feasibility for implementation on wearable devices capable of accurate heart rate monitoring. The models were initially tailored for a prediction window of one hour, later extended to four hours through transfer learning. The encouraging outcomes of this study suggest the potential for wearable technology to facilitate early sepsis detection outside ICU and ward environments.

Early Prediction of Sepsis using Heart Rate Signals and Genetic Optimized LSTM Algorithm

TL;DR

This work tackles the challenge of predicting sepsis onset outside hospital wards by using only heart-rate signals from wearable-capable devices. It evaluates four machine learning architectures (LGB, MLP, LSTM, LSTM-FCN) whose structures are optimized with a genetic algorithm, then extends the one-hour predictions to four hours via transfer learning. Using the PhysioNet 2019 dataset, the authors balance a heart-rate–only feature set through forward filling and noise augmentation to train the models on 12-hour windows around sepsis onset. Results show the LSTM achieves the highest accuracy (AUROCs around for 1 hour and for 4 hours) but with substantial model size, while LGB offers a compact alternative with competitive AUROCs; overall, the study demonstrates feasibility for real-time wearable sepsis warning systems and highlights trade-offs between accuracy, latency, and memory. The findings point to potential for real-world deployment, with future work needed to incorporate non-ICU data and additional features for robust out-of-hospital sepsis detection.

Abstract

Sepsis, characterized by a dysregulated immune response to infection, results in significant mortality, morbidity, and healthcare costs. The timely prediction of sepsis progression is crucial for reducing adverse outcomes through early intervention. Despite the development of numerous models for Intensive Care Unit (ICU) patients, there remains a notable gap in approaches for the early detection of sepsis in non-ward settings. This research introduces and evaluates four novel machine learning algorithms designed for predicting the onset of sepsis on wearable devices by analyzing heart rate data. The architecture of these models was refined through a genetic algorithm, optimizing for performance, computational complexity, and memory requirements. Performance metrics were subsequently extracted for each model to evaluate their feasibility for implementation on wearable devices capable of accurate heart rate monitoring. The models were initially tailored for a prediction window of one hour, later extended to four hours through transfer learning. The encouraging outcomes of this study suggest the potential for wearable technology to facilitate early sepsis detection outside ICU and ward environments.
Paper Structure (20 sections, 5 figures, 1 table)

This paper contains 20 sections, 5 figures, 1 table.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORKS
  3. MATERIAL AND METHODS
  4. DATABASE
  5. PRE-PROCESSING
  6. PROPOSED MODELS
  7. Light gradient boosting (LGB)
  8. Multilayer perceptron (MLP)
  9. Long short-term memory (LSTM)
  10. LSTM-FCN
  11. RESULTS
  12. LGB
  13. MLP
  14. LSTM
  15. LSTM-FCN
  16. ...and 5 more sections

Figures (5)

  • Figure 1: The pre-processing workflow.
  • Figure 2: The workflow of finding the optimal number of neurons for the LSTM model.
  • Figure 3: The architecture of LSTM-FCN model.
  • Figure 4: Results from the test dataset were evaluated one hour before sepsis onset: (a) ROC curves; (b) PR curves; (c) Calibration curve.
  • Figure 5: Evaluation results of all models on the test dataset: (a) AUROC at different prediction windows; (b) AP at different prediction windows.