Health Data Scientist Yuri Rykov on Early Diabetes Diagnostics and AI in Medicine

You work both in industry and academia. Why did you decide to take this path?

For most of my life, I worked in academia: first, at HSE University and later, at Nanyang Technological University (Singapore). At one point, however, I decided to pivot towards the industry: in Singapore, I was invited to join a medtech startup, where I continued doing research; and in Russia, the tech field is strong and there are interesting teams and projects.

This year, I decided to join ITMO – despite having a nice job, I felt the lack of research activities and an inner drive that I wanted to devote to digital medicine.

Why did you choose this field?

When I did sociology research, I grew interested in online health-related communities. I wanted to understand how social media interactions affect people’s behavior in terms of health. It turned out that people with severe diagnoses are eager to interact with each other and share their experience; they can even change their opinions.

Later, this interest led me to the topic of digital biomarkers: the data on our bodies that is collected by wearables, such as step count, pulse, or skin temperature. This data can be used to monitor health and disease risks. I was intrigued by the idea that daily behavioral patterns can indicate early onset depression (1,2), cognitive impairments, metabolic syndrome, or flu. With highly detailed sensor data, we can use ML and AI to determine risk factors. In the future, we can create a wearable diagnostic center that will notify users about signs of disease.

What led you to focus on diabetes diagnostics?

I wanted to work with digital patients’ records – I was curious to see what information could be acquired during a visit to the clinic. Type 2 diabetes is a widely spread chronic disease, the causes of which are closely linked to lifestyle and behavioral risk factors.

Is it challenging to diagnose diabetes at an early stage?

The diagnostics procedure itself isn’t challenging – it’s just a blood test; what’s challenging is predicting the risk: currently, we don’t know the exact causes of type 2 diabetes or the full list of risk factors. Metabolism changes accumulate over time, and the disease can strike out of the blue. However, this can be prevented if patients turn for help during pre-diabetes – as this is a controllable stage.

We hope that over the course of our study, we will discover new pre-diabetes symptoms in patients’ records and thus will get better at predicting diabetes. Using these results, clinicians will be able to timely prescribe preventive treatments to patients.

Could you describe the potential result of your project in more detail?

The project’s main goal is to create a model that will accurately evaluate the risk of diabetes; so, if we are able to produce a high-accuracy solution, this would be a significant result. However, there is a deeper task: the existing diabetes risk evaluation algorithms use a limited number of predictors, like age, gender, weight, and waist circumference, and use linear models. We want to overcome these limitations by using more advanced machine learning methods and expanding the list of predictors. If we can prove that our model is more efficient than the available tools, it would be a solid outcome.

As for the long-term effect, it would be wonderful if our model transformed from a research case to a practical tool – to be used in hospitals. This sounds like a dream, but that’s what motivates me: to create something that will be useful outside academia.

Why did you choose ITMO as the place to implement your project?

I met my colleagues from ITMO while I was still working at HSE University – we collaborated on summer schools, at conferences, and at other events. ITMO has always had the positive image of a dynamic university that's open to collaboration. However, the decisive factor for me was that I personally knew the teams of the Center for Science Communication, the Laboratory for Digital Technologies in Public Health, and the Public Health Sciences Master’s program, particularly their respective heads Daria Denisova, Anton Barchuk, Anna Andreychenko, and Evgenia Sokolova.

Read also:

Changing the World for the Better: Thoughts From New Head of ITMO’s Public Health Sciences Master’s Program

To be honest, at first, I was surprised that ITMO has teams working in the field of public health – after all, the university is often perceived as solely IT-oriented. However, there are plenty of interdisciplinary projects here, so it’s no surprise that one of them is at the intersection of IT, medicine, and public health.

Moreover, our future project is related to ML methods, so this is definitely the right place for it. We hope that there will be students interested in the topic who will be willing to join us and maybe use the project as a testing ground for their ideas and hypotheses.

What kind of students are you looking for?

We would love to work with students interested in epidemiology and risk factors, as well as the development of predictive and survival models. I hope that students of the Public Health Sciences Master’s program will be interested, too, as the project falls into their main focus area.

We are also looking for Master’s and PhD students who are into natural language processing and large language models. We will be working with partially structured text databases, which we’ll need to turn into vectors to be used in machine learning. This isn’t just data labeling, but also information retrieval and data transformation.

Read also:

AI in Medicine: Obstacle or Boon?

Is it feasible to incorporate ML methods into regular clinical practice in the nearest future?

Digital medicine is a rapidly growing field, with thousands of medtech startups, especially AI-based. However, even the best ideas often don’t make it to implementation because clinicians don’t see it as beneficial or economically viable. 

Sometimes, they also don’t trust these technologies. So, the question of how digital technologies can be implemented into daily medical practice remains open. However, progress can’t be stopped – for instance, machine vision technologies are already helping us analyze X-ray images faster and more accurately. Thus, it's clear that AI isn’t just a way to cut down on the medical professional’s working hours, but also an opportunity to significantly improve diagnostics.

Read also:

Radiologist Olga Puchkova Talks About Medical Issues To Be Solved With AI 

How would you describe AI to a clinician to explain why they can trust new technologies?

If I were to talk to a medical professional who is skeptical about AI, I would appeal to evidence-based medicine and demonstrate the data of randomized clinical trials and meta-analyses that prove the benefits of AI (more accurate diagnoses and better outcomes for patients).

If the clinician fears that AI is there to take their job, I think this fear is speculative. With each new invention, something “old” was predicted to die: cinema would destroy theater, TV would destroy cinema. But in the end, these things coexist. In a capitalist society, to be honest, there will always be something to do – for all of us, including clinicians.

If you had unlimited resources, what would your dream research project be?

One of the hypotheses I am curious about is the connection between our behavior, environment, and health. I would love to conduct a large-scale longitudinal study, tracking people’s life-long social interactions and events, as well as monitoring their physiological and biological states, maybe even EEG. At the intersection of all of this data, we’d be able to see how our environment, status, and lifestyle affect our health, longevity, and predisposition to various diseases.