A decision support system for reducing false alarms in ICU
|Title||A decision support system for reducing false alarms in ICU|
|Summary||Developing a clinical decision support system using machine learning and biomedical signal analysis techniques for an ICU setting.|
|Keywords||Machine learning, Biomedical signal processing, Clinical decision support, Temporal data analysis|
|TimeFrame||Winter 2016 / Spring 2017|
|References|| 1. Liu C, Zhao L, Tang H, Li Q, Wei S, Li J. Life-threatening false alarm rejection in ICU: using the rule-based and multi-channel information fusion method. Physiological Measurement. 2016;37(8):1298.
2. Konkani A, Oakley B, Bauld TJ. Reducing hospital noise: a review of medical device alarm management. Biomedical Instrumentation & Technology. 2012;46(6):478-87.
3. Cvach M. Monitor alarm fatigue: an integrative review. Biomedical Instrumentation & Technology. 2012;46(4):268-77.
4. Goldberger AL, Amaral LAN, Glass L, Hausdorff JM, Ivanov PC, Mark RG, et al. PhysioBank, PhysioToolkit, and PhysioNet. Circulation. 2000;101(23):e215.
|Prerequisites||Good knowledge of applied mathematics and signal processing. An ability to implement state-of-the-art algorithms in a suitable programming environment. An interest in machine learning algorithms.|
|Supervisor||Sławomir Nowaczyk, Awais Ashfaq|
Background: With the advent of advanced patient monitoring systems in healthcare, alarms are ubiquitous and have been a subject of technical and psychological research for decades. False alarms in intensive care monitors have been reported to approach up to 86% (1). These ceaseless noises supplement the already stressful ICU environment causing sleep deprivation, sensory overload, desensitization to alarms, reduced quality of care and a depressed immune system. More about ICU alarm fatigue can be found in (1-3)
Objective: The desired objective of the project is to develop an intelligent decision support system that reduces the number of false alarms using ICU patient data. The datasets include the following:
- Waveforms: such as multi-channel ECG, multiple blood pressure recordings (like ICP, LAP, CVP, LAP etc.), Plethysmography and Respiratory waveforms.
- Numeric data: Body temperature, Cardiac Output, Heart Rate, Respiration Rate measured at equal or random intervals.
The student shall explore a wide range of exciting new ideas in the field of multi-channel data fusion and clinical decision support systems such as evidence-based learning, random forests, deep learning etc.
- How to combine different heterogeneous sources of temporal information?
- How to interpret such models to extract the underlying understanding of the classifier output?
Project Plan: The student will have a lot of freedom in choosing the focus of the project, however, the suggested plan is as follows:
- Investigate state-of-the-art Clinical Decision Support Systems (CDSS) and evaluate how they can be adapted to an ICU setting. These include knowledge based and non-knowledge based CDSS.
- Explore a wide range of machine learning algorithms that have been studied in the context of non-knowledge based CDSS and look for ways how they may be integrated with the domain knowledge.
- Develop a working prototype that can classify false alarms among all vital alarms.
- Test the developed prototype on pre-recorded ICU data, or if possible, in real ICU setting and report suggestions for improvements.
Data needed for the project:
- Labelled and unlabeled data for semi-supervised learning – available at (4)
- ICU patient data for testing – available at (4). The developed prototype might also be put to test in real ICU setting in Halland Hospital for quality analysis.
- Slawomir Nowaczyk (email@example.com)
- Awais Ashfaq (firstname.lastname@example.org)