Thanasis Tsanas

Image

What is your research focus?  

My work is broadly focused on data science, and specifically on time-series analysis, signal processing, and statistical machine learning. I develop and apply signal processing and statistical machine learning algorithms to explore data and decipher complicated concealed statistical relationships, primarily in healthcare applications. The algorithms I develop are directly driven by and validated on complicated real-world problems, aiming to facilitate interpretation of the underlying key mechanisms of the modelled system. My work is inherently multi-disciplinary tackling challenging problems primarily in healthcare domains from neurodegenerative disorders and mental health, to asthma, cardiovascular disease, and neonatal monitoring. At the same time I develop novel generic signal processing and statistical machine learning algorithms with wider impact and applicability, for example in areas such as information fusion, feature selection, multimodal signal processing, and statistical learning.

It is difficult to pinpoint one thing where it all started in the directions I have taken for my work, so I will have to provide you with a long story for how different pieces of the puzzle came together for me.

To a large extent it all started as an undergraduate student in Biomedical Engineering in Athens in 2005, when I had a great supervisor (Fanis Maniatis) with whom I developed my first algorithmic models to explore the cardiovascular system. At the same time, I was working as a biomedical engineering trainee on a 6-month internship at one of the top hospitals of Greece, the Onassis Cardiac Surgery Centre. That gave me an opportunity to work alongside some of the top clinicians in the country, including attending often at the operating theatre, and made me realize the potential of these interactions and people working with different types of expertise. Coming back to Fanis, he would leave his work and drive to where I was doing my internship so that we would have our supervisory meetings typically after 18:30 – a testament of how supportive a supervisor he was. He had spent an enormous amount of time to train me and I will always be grateful to him – unfortunately he decided to leave academia for a prestigious industrial post and we have not had the opportunity to work on publishing something together. I was the only student he had supervised.

In my second undergraduate degree, in Electrical Engineering, I had the opportunity to work with Yannis (John) Goulermas at the University of Liverpool (2005-2007), who has been one of the most influential people in my work (and life). He pushed me really hard on developing solid mathematical foundations combining algorithms and biomedical engineering, and this led to my first journal publication – which is quite unusual for an undergraduate thesis. Parenthetically, at the time I was an undergraduate student working with him, I felt he was an extremely demanding supervisor, however I always deeply appreciated he was very direct, meticulous and supportive. We have been in regular contact following the completion of my undergraduate degree, and in time he became my most trusted life-long mentor and friend. I was devastated when he passed away in May 2022. I have so many things I carry on from my discussions with Yannis, that are still shaping my day-to-day research decisions.

In my MSc (2007-2008) I studied in more depth signal processing and learned about how to process signals and transmission protocols etc. At the time it seemed to be taking me away from the (bio)medical applications, however it all came to fit nicely later on, as I picked up very useful skills that I would use for processing and telemonitoring.

When I joined the Oxford Centre for Industrial and Applied Mathematics at the University of Oxford for my PhD in 2008, I was suddenly surrounded with some brilliant mathematicians and I felt intimidated at first. I had the opportunity to work with a great supervisor, Max Little, who was very meticulous and supportive particularly at the beginning of my PhD (in my second year he left Oxford to join MIT). We decided early on that it would be better to be moving from mechanistic models (that I had been working on previously) towards data science, signal processing and statistical machine learning. The work I did in my PhD focused on developing signal processing and statistical machine learning algorithms to provide insights into Parkinson’s disease using primarily speech signals. 10+ years on, I still publish papers that I had started thinking/working on as a PhD student.

My work since then has diversified in terms of working with different physiological signals (e.g. electrocardiograms, electroencephalograms) and in particular data collected from wearable sensors. In all cases, the aim is to develop tailored algorithms to extract information from potentially massive datasets.

Similarly, I have worked in very different applications, from mental health, neurodevelopmental and neurodegenerative disorders, asthma, cardiology, and endometriosis. Moreover, I have worked on problems (e.g. as a consultant) outside the healthcare remit, for example in forensic phonetics, deciding on the sports odds numbers for betting companies (i.e. betting strategy optimization to “beat the market” and maximize profit from the estimated games’ results and inciting participation from people who are betting), risk disclosure etc. Many of these projects are currently ongoing, and I have different members of my team and collaborators who are the main people doing the day-to-day work. Recently, I worked on a project to understand the implementation of different hospital systems for successful management of medications, ensuring they have good mitigation strategies taking into account patient-specific characteristics. It is often challenging when working on a new problem, but it is definitely never boring!

These seemingly very different areas are, in fact, very similar under the prism of what I see as a mathematician/engineer/data scientist: the unifying theme is the development of an algorithmic framework to provide insights the problem on the basis of the collected data.

What is your innovation idea? 

I believe there is enormous untapped potential in the use of wearable technology and smartphones for healthcare applications.

Developing a framework to analyze the diverse types of data we could collect and provide new insights into aspects of daily living, including about physical activity, sleep, mental health aspects, in a way that can be actioned by domain experts (e.g. clinicians) or carers can truly revolutionize contemporary healthcare delivery.

Why does this idea matter, what impact will it have on the world and what problem will it solve? 

With the population growing older overall, there are increasing healthcare demands arising from chronic diseases and.  Given that various devices (from computers to smartphones and wearables) are becoming increasingly more widespread and overall more affordable, there is an opportunity to affect millions of lives. What used to be considered e.g. a playing machine (a computer) or something that merely tells you the time (watch), or a bulky device to call people (mobile phone) we now see these devices can record and provide (potentially in real-time) diverse pieces of information. With appropriate data processing we can provide unique insights.

The beauty of all these is that it does not solve a single problem in the traditional sense, e.g. towards addressing people with some chronic condition e.g. Parkinson’s. The potential of what I envision is much broader and affects virtually any aspects of healthcare, from facilitating the monitoring of pregnant women and the normal development of the fetus, to monitoring neurodevelopment in young children, developing decision support for different ranges of diseases across different ages (e.g. childhood asthma to dementia), up to palliative care – truly the full spectrum of the life cycle.

To give three indicative specific examples with different devices which we have already demonstrated: (1) using speech (in fact sustained vowels, which are generalizable across any language) collected over the standard phone network to demonstrate how we can assess neurodegenerative disorders at scale across 7 countries (more than 20,000 samples collected within 6 months); (2) using passively collected information from smartphones such as geolocation to assess depression; (3) using information collected from smartwatches to objectively characterize interventions to improve mental health.

I am acutely aware of the danger that these developments and plans will disproportionately affect people from different economic backgrounds who might not be able to access expensive equipment, and I am actively trying to ensure we develop solutions which are affordable using minimal resources where possible.

I believe democratizing access to healthcare and improving general well-being is a noble goal, which will ultimately help people live longer, healthier lives.

What is the future of your research?

A large part of my work relies on colleagues who collect large scale data in collaboration in different domains, so to some extent I am relying on the high quality data that can be provided and this can take me in different directions.

Most of my work is about developing new ways to extract information from the collected data. There is an enormous amount of work in terms of pursuing new ways to visualize data, developing approaches to mine the different types of data (often considering these data modalities jointly), developing robust generalizable models, and ultimately packaging any developed algorithsm we have in a from that can be externally deployed in practice.

What motivated you to apply to the Bayes Innovation Fellows programme and what do you hope to gain from it?

I have followed the traditional academic path so far working primarily on solving problems and publishing work. I am excited at the prospect of dedicating time and effort to potentially being able to translate many of these ideas in practice.

I hope I will get useful advice on how to explore commercialization opportunities, learn from pitfalls and successes of collegues and enterpreneurs who have walked similar paths in the past, and the support of the cohort and the University (e.g. exploring partnerships) towards developing something that I think could be game-changing. 

Links

 The Usher Institute

Edinburgh Medical School