Columbia University: Interdisciplinary research and innovation with a Greek “stamp”

Interdisciplinary research and innovation with a Greek “stamp”, in the field of its applications Artificial Intelligence in medicine and especially cardiology: this is the goal of the scientific movement “Heart-In-AI” (Heart Innovation and Artificial Intelligence), which was co-founded by Polydoros Kambaktsis, assistant professor of Cardiology at Columbia University, USA, in order for doctors and engineers of various specialties based either in Greece or abroad to work on joint projects, which put algorithms at the service of human health. After all, he himself, who has lived in New York for the last 11 years, has a dual scientific capacity: that of a cardiologist and that of an electrical engineer.

“When we talk about Artificial Intelligence (AI), a very critical element for the implementation of algorithms is to ask the right questions – goals. I consider this creative element to be compatible with the Greek mind and I believe that in Greece an important center of innovation can be created in this field. Our country has excellent engineers and IT scientists, who work both in Greece and abroad. I have formed the feeling – perhaps I am also biased – that creation suits us very well, as long as we are focused and guided by mentors, as it seems that disorganization and strife are related elements to the Greeks”, notes Mr. Kambaktsis, who co-founded the group with them Seraphim Moustakidis, PhD in electrical engineering from the company AiDEAS, Anastasios Drossou, PhD in electrical engineering and researcher at EKETA, Alexandros Briasoulis, assistant professor of Cardiology at EKPA and the doctors Ilias Doulamis and Aspasia Tzani.

Already, the scientific team – which remains open to new members, with the first point of contact being the social networks “Linkedin” and “Twitter” – has completed its maiden project, with the object of the usefulness of Machine Learning (MM) for the prediction of clinical outcome of patients undergoing heart transplantation. Now, he is entering “deeper waters”, with a second project, which aims to use MM to better or optimally match heart transplant patients to personalized immunosuppressive therapy – optimal immunosuppressive therapy ensures higher chances of survival and lower chances of miscarriage graft, as well as fewer side effects. Both projects draw on data from the US national database UNOS (United Network for Organ Sharing), drawing on over 18,000 transplant patients between 2010 and 2018.

Making minutes …seconds

But why is AI important in medicine? A simple example cited by Mr. Kambaktsis is that of heart ultrasounds: “For a cardiologist to read a patient’s heart ultrasound, he needs to see up to 100 “clips” in detail in order to make a complete diagnosis. This procedure can take from 10 minutes, as long as there is nothing pathological, up to 40 minutes. TN could handle extensive preprocessing in seconds to minutes before the clips reach the doctor’s eyes. It is obvious that this can help to improve and standardize the process, as well as to reduce time and costs,” says the professor, clarifying that he considers it impossible for AI to replace doctors now or in the future: “even when we go too far in this part (of AI), we can neither ethically nor scientifically entrust the treatment of a patient to a machine” appreciates.

Assisted diagnosis, as described above in the case of the heart ultrasound, is only one of the categories of TN applications in medicine and the one that – while still under development – has the greatest development and clinical applicability. For example, there is an intravascular TN imaging application for coronary arteries based on OCT (Optical Coherence Tomography) technology, which helps the interventional cardiologist to have an image of the vessels, the presence or absence of atherosclerotic plaque and its characteristics, so that correlation can be made and with a possible clinical event (heart attack). In fact, IT applications that help doctors in procedures related to diagnosis, radiation and pathological anatomy exist as equipment in public hospitals in Greece, perhaps not to the same extent as abroad, but certainly to a greater extent than in the past, he says.

…and entering treacherous “territory”

The other two groups of TN applications in clinical medicine are, according to Mr. Kabaktsis, the prediction of patient outcomes and the discovery of new knowledge, even knowledge in fields that are considered “uncharted territory” for researchers. “There are diseases that are quite complex, in which we are trying to understand where we should go. For example, heart failure with preserved ejection fraction and pulmonary arterial hypertension. They are two diseases, which are currently at the edge of research and we are trying to understand them, to see how we should approach them. MM helps us to collect and process a lot of information, which if combined with each other, can for example group categories of patients with specific characteristics, so that we can reach useful conclusions. In other words, MM is used as a starting point, giving further research directions – in general, this is what “Pfizer” also started in Thessaloniki, with the Digital Innovation Center it created in the city, where Big Data for medicines is used, among other things , with the goal of finding unknown correlations through AI,” he points out.

TN can rely on retrospectively collected data to determine exactly where clinical studies should be directed. This is very important because in many cases there is difficulty in designing clinical studies, either because they are very expensive or because the question of what to investigate has not been precisely defined. Thanks to TN, a clinical study can be designed more efficiently, so that phenomena such as what is sometimes recorded today are not observed, that is, a clinical study ends after two or three years of work and there is no usable result. Similar utility is offered by TN and MM in the production of hypotheses for rare diseases, the research on which often proves to be economically unprofitable, because they concern a limited number of patients.

The original idea of ​​the aforementioned first project of the group, for which Kambaktsis, Moustakidis, Briasoulis and Drossou worked, was to create – with data from the UNOS database – predictive models for the clinical outcome of patients undergoing heart transplantation. “The results were not dazzling, but the outcome prediction was improved compared to existing prediction models such as IMPACT. We are now trying, using data also from UNOS, to answer more complex questions, which have not yet been satisfactorily answered clinically, namely how we can achieve optimal matching of transplant recipients with their immunosuppressive therapy. We need a critical mass of data for safe conclusions, and we try to “clean” the data before feeding it to the algorithm to see what we get back. In the next quarter, I believe we will have the first results,” he notes.

The vulnerable Big Data of hospitals

Artificial Intelligence means Big Data processing, and Big Data in the health sector, especially as it relates to hospitals, is clearly vulnerable. “As a rule, big data cannot be produced by a single hospital, but by many different ones. So they need to be transferred to be combined, which makes them more vulnerable to cyber attacks. That is why it is important to create databases that gather them in a safe way, so that afterwards the MM models can process them without risk”, concludes Mr. Kambaktsis.

It is recalled that based on data presented last year by the professor of Piraeus University, Christos Xenakis, in 2021 the health sector had for the eleventh consecutive year the highest cost from cyber attacks, in terms of its economic impact. In the health sector, this cost actually amounts to $9.23 million/cyber incident, on average on the planet, far more than in the financial sector and banking._

RES-EMP

View the news feed and get the latest news.