According to estimates, up to half a million people in the Czech Republic suffer from atrial fibrillation. The diagnosis is associated with an increased risk of serious complications and therefore requires not only regular patient monitoring but also long-term treatment and, in some cases, hospitalization, which generates significant costs. Early access to appropriate therapy is crucial for prognosis. A research team led by Ing. Jakub Hejč, Ph.D., from the International Clinical Research Center (ICRC) has decided to contribute to this effort, setting a goal to reduce the number of repeat interventional procedures by one third.
The primary aim of the research is to identify new biomarkers and hidden nonlinear relationships associated with the development of fibrosis and the subsequent burden of atrial fibrillation. “These relationships would remain undiscovered using conventional statistical methods. Our model therefore does not only answer the question ‘what will happen’, but also seeks to explain ‘why it happens’. This approach is analogous, for example, to modern applications in oncology, where it helps uncover disease mechanisms or responsible gene mutations,” explains Ing. Hejč.
The specific morphological clusters of electrophysiological signals identified so far correlate with treatment failure, even in regions where no abnormalities have been demonstrated using conventional biomarkers.
The model’s contribution to personalized treatment lies not so much in defining which patients should undergo ablation, but rather in deciding how to perform the procedure as effectively and as gently as possible for each individual patient. By creating an individualized electrophysiological profile and identifying specific foci that lead to recurrence, the intervention can be more precisely targeted.
The research does not rely on standard clinical parameters, as these represent only an indirect reflection of pathophysiology. Instead, it focuses on the analysis of high-dimensional data from 3D electroanatomical mapping. Ing. Hejč explains the principle: “These are essentially point clouds in space recorded directly inside the heart, carrying detailed information about chamber anatomy and the propagation of electrical signals through tissue. It is within these complex maps that we search for hidden patterns capable of predicting arrhythmia recurrence.”
Ing. Hejč points out that the AI model differs from those designed for early diagnostic support not only in its goal, but also in how it is integrated into the clinical workflow. “If we manage to fully interpret the hidden patterns identified by the model and understand their electrophysiological basis, we will not need to deploy a computationally demanding neural network. The aim is to extract new, clearly definable indices from these patterns. In clinical practice, this would mean integrating only a simple and understandable algorithm for calculating these new biomarkers, which increases the likelihood of rapid adoption by physicians and potentially by commercial companies as well,” he says.
What makes the Czech project unique compared to others? “Most similar AI projects focus on binary prediction of arrhythmia recurrence, typically using ambulatory ECG recordings combined with clinical risk scores. Our approach is specific in that it incorporates survival analysis directly into the model optimization process. This allows us to derive a unique risk profile for each individual relative to other patients, thereby refining the learning process itself,” explains Ing. Hejč.
He also acknowledges the data-centric orientation of the research: “Personally, I see the greatest potential in developing methods that can generalize from a small amount of high-quality data and connect knowledge across disciplines. I consider this direction key to partially opening the proverbial black box, as it naturally reduces model uncertainty. Essentially, we force the model to learn causal patterns rather than statistical correlations typical of models trained on massive, noisy datasets.”
The model outputs undergo expert validation, and the research team is also considering measures that would allow the tool to be applied to data other than those on which it was trained at St. Anne’s University Hospital, with which the team has a long-standing collaboration.
Editorial Team, Medscope.pro
Did you like this article? Would you like to comment on it? Write to us. We are interested in your opinion. We will not publish it, but we will gladly answer you.
Don‘t have an account? Create new account
Enter the email address that you registered with. We will send you instructions on how to set a new password.
