School of Electronics, Electrical Engineering and Computer Science
& ECIT Global Research Institute
Proposed Project Title: Predicitive analytics in an ICU by processing streams of physiological data in real-time
Principal Supervisor: Dr Charles J Gillan
Second Supervisor: Dr Murali Shyamsundar (medicine)
Researchers at the ECIT Institute at QUB have worked closely with clinical colleagues in the QUB School of Medicine since 2013 to develop a prototype system for predictive analytics on streams of respiratory data from ventilators in an ICU. This PhD project will first of all educate the student in the wider field of high performance predictive analytics and then seek to apply it to the ICU setting. The aim is to generalise the work already done on respiration to other clinical parameters, and combinations of physiological parameters, thereby giving a more complete view of the condition of each monitored patient. This means that research will focus predictive analytics from values provided by several signals.
The monitoring and alert system has to run accurately, 24x7, for all ICU patients being monitored. Patients can arrive on the system and can leave from it at any point in time. Furthermore patients may be disconnected from the monitoring system for finite periods of time.
The project will be jointly supervised by Dr M. Shyamsundar, a QUB clinical academic and an NHS ICU Consultant.
The project will be based in the ECIT Institute and therefore is primarily a project developing research skills in: programming for a mission critical system, concurrent software programming and high performance computing. The student will develop extensive skills including in: C++, C, OpenCL, operating systems, parallel computing, distributed computing, code optimisation, numerical analysis, computational mathematics, databases and computational physiology. The general application is that of data analytics, a field that includes many mathematical methods from the field of statistics.
In order to minimise ethical issues of working with actual patient data, the project at least in the first instance seek to use databases of patient signals publically available. The following paragraphs are provided to give a prospective student a flavour of the kind of work that would be undertaken in the PhD project in so far as processing one physiological signal is concerned. To be clear, the ECG processing reported here is well known and would not in itself be part of the research. This is only an example of the kind of processing an algorithms that would be involved. Much research remains to be done on other physiological parameters.
The MIMIC (Multiparameter Intelligent Monitoring in Intensive Care) Databases contain physiologic signals and vital signs time series captured from patient monitors, and comprehensive clinical data obtained from hospital medical information systems. Within this set of data, the MIMIC III Waveform Database is freely distributed and can be used for research purposes. It contains several ElectroCardioGraph (ECG) traces. A typical cardiac waveform is shown below in which the PQRSTU parts of the waveform are identified.
Analysis of the continuous waveform, a process that can be carried out by computers, may reveal information on the health of the patient’s heart function to the clinician. For example, the P-wave represents the activation of the two atria, the upper chambers of the heart, while the QRS complex and T-wave represent the excitation of the lower chamber of the heart, the ventricles. The time interval between different features is a key piece of information for clinicians. For example:
- Bradycardia identifies as an R-R interval exceeding 1s while
- Tachycardia corresponds to an R-R interval less than 0.6s
ECG signal processing can be divided into two stages by functionality:
to removes or suppress noise from the raw ECG signal
- feature extraction
to obtain diagnostic information from the ECG signal 
Wavelet transforms  have proved in recent years to be particularly suited to analysis of the ECG signal. Similar analysis, performed with a variety of mathematical kernels, can be applied to other physiological signals.
Machine learning and decision support in critical care is receiving is a hot topic in medicine at this time. There are many issues of compartmentalization, corruption, and complexity involved in collection and pre-processing of critical care data . Understanding these will be a key part of the later stages of the project.
 M. Saeed, M. Villarroel, A.T. Reisner, G. Clifford, L. Lehman, G.B. Moody, T. Heldt, T.H. Kyaw, B.E. Moody, R.G. Mark. Multiparameter intelligent monitoring in intensive care II (MIMIC-II): A public-access ICU database. Critical Care Medicine 39(5):952-960 (2011 May); doi: 10.1097/CCM.0b013e31820a92c6.
 D. Balasubramaniam and D. Nedumaran, “Implementation” of ECG Signal Processing and Analysis Techniques in Digital Signal Processor based System,” MeMeA 2009 - International Workshop on Medical Measurements and Applications, Cetraro, Italy, May 29-30, 2009.
 P S Addison, Wavelet transforms and the ECG: a review, Physiological Measurement, 26 (2005) R155-99 doi:10.1088/0967-3334/26/5/R01
 A E W Johnson, M M Ghassemi, S Nemati, K E Niwhaus, D A Clifton and G D Clifford
Machine Learning and Decision Support in Critical Care, Proceedings of the IEEE, Volume 104, Issue 2, 2016, 444-66
Supervisor Name: Dr Charles J Gillan Tel: +44 (0)28 90971847
QUB Address: The ECIT Institute Email: email@example.com
Queen’s Road, Queen’s Island