Chelyabinsk State University (ChelSU) has developed and tested an intelligent system for predicting recurrent episodes of atrial fibrillation, the university's press service reported. The algorithm not only assesses the risk of recurrence but also shows the doctor which specific indicators influenced the decision.

The system was created by a researcher from the ChelSU Faculty of Mathematics under the guidance of university professors in cooperation with cardiologists from Regional Clinical Hospital No. 3. Data from 97 patients who received treatment from 2010 to 2013 were used for training. Initially, each case was described by 27 features; after processing, they were reduced to 18. A comparison of several machine learning algorithms showed that the support vector method provides the best efficiency. A patient is classified as high-risk only if the calculated probability of recurrence exceeds 91%.

Such a strict threshold reduces the frequency of false-positive conclusions and allows doctors to focus on the most severe cases. An important feature is the transparency of the algorithm's operation: it explains the contribution of each clinical sign to the final decision, which fundamentally distinguishes it from a “black box.” Currently, the development is a ready-made decision support tool for practical cardiology.