Theme 1. Biomedical Signal Processing

Recent advances in entropy quantification algorithms for biomedical signals: Beyond univariate time series


Javier Escudero, University of Edinburgh, UK
Teresa S. Henriques, University of Porto, Portugal
Alberto Porta, University of Milan, Italy
Evangelos Kafantaris, University of Edinburgh, UK
Matthew W. Flood, Luxembourg Institute of Health, Luxembourg
Mirvana Hilal, University of Angers, France

Nonlinear analysis techniques are receiving an increasing degree of attention to analyse and understand the behavior of physiological systems. In particular, techniques based on entropy quantification are powerful to process real-world data due to their robustness to noise and suitability to short time series. These nonlinear techniques have extensively been used to analyse univariate time series related to the cardiovascular system and brain activity, among other time series, confirming that diverse pathological process disrupt the nonlinear characteristics of physiological recordings. Traditionally, entropy techniques have been applied to individual time
series. However, entropy-based techniques can reveal informative patterns in data other than univariate time series, such as multivariate time series and images. Thus, this session will provide researchers interested in nonlinear techniques or multivariate and/or multimodal physiological data with an overview of the state of the art in entropy quantification algorithms and with recent
methodological approaches that extend the concepts of entropy quantification beyond the study of simple univariate time series. The session starts with an introductory talk on entropy quantification algorithms by Dr Henriques, who has recently published reviews on this matter. This will be followed by talks by Prof. Porta, Mr Kafantaris, Dr Hilal, and Dr Escudero on recent technical extensions of
entropy quantification techniques for multivariate, multimodal, imaging, and graph data. The session will close with a talk by Dr Flood on the recent open-source toolbox entropyhub. This session will provide a venue to discuss cutting edge research in entropy quantification and to facilitate their application to diverse biomedical data.

Open research in Biomedical Signal Processing: Cuffless Blood Pressure Estimation Using the MIMIC-IV Database


Peter Charlton, University of Cambridge, UK
Tom Pollard, Massachusetts Institute of Technology, USA Benjamin Moody, Massachusetts Institute of Technology, USA Brian Gow, Massachusetts Institute of Technology, USA
Elisa Mejía-Mejía, City, University of London, UK
Panicos Kyriacou, City, University of London, UK

The field of Biomedical Signal Processing stands to benefit greatly from open research. Reproducible studies, accompanied by code and data, allow others to build on the state- of-the-art and to quickly translate between academia and industry. Openly available tools are widely used. Indeed, the MIMIC Waveform Database was referenced in 125 EMBC papers between 2016 and 2020.

The aim of this interactive workshop is to provide participants with the knowledge, skills and tools required to conduct open research in the field of biomedical signal processing, with a focus on the popular MIMIC Waveform Database. It will include a formal announcement of the release of the MIMIC-IV Waveform Database.

Firstly, the workshop will provide participants with an understanding of publicly available datasets containing physiological signals. We will summarise the datasets available online, focusing on those on PhysioNet. We will then introduce the MIMIC Waveform Database, including its clinical context, structure and formatting.

Secondly, the workshop will provide participants with essential skills for conducting high quality research with openly available data. Participants will work through interactive tutorials in the Python programming language using the WFDB Toolbox, a library of Biomedical Signal Processing tools. The tutorials will introduce key aspects of research study design and analysis, including: data exploration, selection and extraction; pre- processing; signal processing; analysis; and interpretation.

Finally, participants will work in groups on case studies replicating findings from the literature. This will provide experience with real-world analyses, and opportunity for networking. Researchers will be on hand to answer questions.

The workshop will be led by researchers who develop MIMIC and the WFDB toolboxes, and who have a track-record in reproducible research. It will be of great interest to students, researchers, and engineers from academia and industry.

Also of Interest…

The following sessions may be applicable to the theme Biomedical Signal Processing although they are primarily listed under a different theme.

Session title Primary theme
Signal Processing and Machine Learning Paradigms for Enabling the Digital Health Ecosystem 9