Related Experiment Video
Updated: Jul 16, 2025
07:42
A Data-Driven Approach to Quantifying Immune States in Sepsis
Published on: February 7, 2025
246
Feature augmentation and semi-supervised conditional transfer learning for early detection of sepsis
Yutao Dou1, Wei Li2, Yucen Nan2
1College of Computer Science and Electronic Engineering, Hunan University, Changsha, 410082, China; Centre for Distributed and High Performance Computing, School of Computer Science, The University of Sydney, Darlington, NSW, 2008, Australia.
Computers in Biology and Medicine
|September 16, 2023
Summary
Early sepsis detection is vital. New machine learning models, ITFG and SAC-TL, improve early sepsis identification using physiological data, significantly enhancing patient outcomes and reducing mortality.
Area of Science:
- Medical Informatics
- Machine Learning in Healthcare
- Critical Care Medicine
Background:
- Sepsis poses a significant public health risk, causing substantial morbidity and mortality.
- Early detection is critical for improving patient outcomes.
- Current methods like Sequential Organ Failure Assessment (SOFA) face challenges in early sepsis detection due to data acquisition limitations.
Purpose of the Study:
- To develop an interpretable machine learning model for early sepsis detection within six hours of onset.
- To introduce a Semi-supervised Attention-based Conditional Transfer Learning (SAC-TL) framework to improve model generalizability.
- To address challenges of feature sparsity and missing data in sepsis prediction.
Main Methods:
- Proposed an interpretable machine learning model, Interpretable Tree-based Feature Generation (ITFG), utilizing feature correlations for sepsis identification.
- Introduced a Semi-supervised Attention-based Conditional Transfer Learning (SAC-TL) framework for enhanced generality and early warning capabilities.
- Leveraged continuous physiological measures and addressed data sparsity and missing data issues.
Main Results:
- Achieved an Area Under the Curve (AUC) of 97.98% on the MIMIC dataset.
- Obtained an AUC of 86.21% on the PhysioNet dataset.
- Demonstrated effective sepsis detection across different data environments and achieved state-of-the-art early detection results.
Conclusions:
- The proposed ITFG and SAC-TL approaches effectively enable early sepsis detection using physiological data.
- These methods address data challenges and offer practical solutions for various generalizability needs.
- The study highlights the potential of advanced machine learning for improving sepsis management and patient outcomes.
