We are delighted to announce Tasos Papastylianou as one of our Data Impact Fellows for 2023. In this post Tasos shares a bit about his background, his current work and research and what he hopes to get out of the Fellows scheme.
Background
I am a Research Fellow at the Institute of Public Health and Wellbeing, at the University of Essex. I got my degree in Medicine at the University of Bristol in 2006, and initially worked as an NHS doctor for two years. At that point, having predicted that the future of Medicine would inevitably come to rely heavily on Machine Learning and Artificial Intelligence tools, I decided to take a year off from my NHS training, and undertake an MSc in Advanced Computer Science — Machine Learning and Data Mining, also at the University of Bristol. There I discovered my love for Computer Science and AI (not to mention Caramel Macchiatos!) and got to unleash my dormant inner computer scientist.
After some more work in the NHS as a physician, I decided to follow my inner computer-nerd calling and applied for a scholarship to undertake a DPhil in Biomedical Engineering, via the ‘CDT in Healthcare Innovation’ programme at the University of Oxford. My thesis work proved that current validation methods used in medical imaging yield unreliable results, and proposed more reliable alternatives. In addition, as part of that work I also proposed fuzzy methods that allow a clinician to incorporate a layer of clinical explainability on the machine-predicted results.
During my time in Oxford, I also co-founded Sentimoto Ltd, a (now dissolved) start-up company dealing with wearable tech for older adults. At the same time, I also took part in an international project which involved low-cost computerisation of water pumps in rural Kenya, enabling our team in Oxford, in collaboration with local Kenyan government, to monitor pump performance and predict imminent failures. This allowed local authorities to conduct prompt repairs to ensure continuous access of rural communities to clean water.
After completion of my PhD, I was hired as a postdoc at the University of Essex for two separate research projects. The first involved work on the NEVERMIND project, an EU project dealing with Biomedical Signal analysis from smartphones and wearables for the management of secondary depression in people with life-changing primary medical events. The second involved work with Brain-Computer Interfaces (BCI) in the context of the BARI project, a UK-US collaboration investigating optimal co-operation in hybrid teams of Human and AI agents.
My research, and role of the UKDS
In my current post as a Research Fellow in Health Informatics at the Institute of Public Health and Wellbeing, I am pursuing research which focuses on clinical and public health applications of Artificial Intelligence and Machine Learning. In particular, I’m interested in:
- Data analysis and information fusion from biological signals (e.g. Brain Computer Interfaces, wearables, hospital-derived, etc) and digitised health data;
- Medical image analysis, explainability, and appropriate validation;
- AI or tech-oriented public health interventions, and the use and benefits of free and open-source software in medicine.
As part of the role, I am also acting as a probationary Lecturer for the School of Computer Science and Electronic Engineering.
As my role at the Institute is still quite new, I cannot readily rely on the existence of data generated from previous projects; therefore, I have relied heavily on data made available via the UK Data Archive to kickstart my research. This has been particularly useful for conducting preliminary exploratory analyses on existing, good-quality, well-organised data, and judging feasibility for future research.
Areas where I’ve used this included:
- an investigation into childhood obesity issues, with a particular focus on coastline communities;
- an investigation into possible factors affecting body image in adolescents;
- child mental health clusters as seen through the Strengths and Difficulties Questionnaires collected through the Understanding Society dataset;
- and more recently a preliminary analysis of the role of unexpected clinical admissions on precipitating homelessness.
Other ongoing research involves more clinical or biosignal-oriented projects. For instance, I am currently involved in a Knowledge Transfer Partnership with Yulife Ltd, in a project that aims to maximize wellbeing and health outcomes for the company’s clients, using a data-driven, gamified approach to healthy lifestyle promotion. I am also a Co-Investigator in a recently submitted European Grant, which aims to provide healthcare organisations with digital tools that enable health workers to manage and maximize their resilience in the face of unexpected crises and adversity.
Future research plans
My research is very multidisciplinary by nature, and effectively split between the social and clinical domains. While the UKDS is less relevant to the clinical/biosignal side of things, it has proven invaluable in the social/public-health side of things. Therefore, the specific research I wish to focus on in the context of the Data Impact Fellows scheme, is the topic of how clinical admissions can on occasion inadvertently precipitate homelessness, and the subsequent impact this has on the individuals involved. The UKDA contains many studies and data that relate to this topic over the last few decades; I have already identified a few relevant datasets, which will initially form a nice student project for a Year 12 school/college student as part of the Nuffield Research Placements scheme that I am involved in as a supervisor. This will aim to get an idea of the extent to which clinical admissions precipitate homelessness, to gauge the extent of the problem, and how it compares to other routes to homelessness. Beyond this, I hope to collaborate with interested clinicians in the NHS, both in hospitals and on the GP side, to collect relevant data that could be used to train machine learning models, which could then be used at the hospital admission stage to stratify patient risk, and evaluate whether such a model could be used in the context of an alert system, flagging patients at risk for further action prior to discharge.