Using new artificial intelligence and machine learning techniques, it will be possible to accurately determine where certain cardiac arrhythmias are triggered. Establishing the place of origin of these arrhythmias will increase the effectiveness of one of the most common procedures for their treatment: radiofrequency ablation, which essentially consists of inserting a catheter into one of the heart ventricles, with which the necessary radiofrequency is emitted to eliminate the alteration of the heart rhythm. So states a scientific article recently published in the journal Frontiers in Cardiovascular Medicine, thanks to research led by the Physense research group of the UPF BCN Medtech Research Unit.

More specifically, the research proposes applying AI and machine learning techniques to pinpoint the origin of a specific type of ventricular arrhythmia. All ventricular arrhythmias are caused by alterations in the normal rhythm of the heart that begin in its internal chambers (the ventricles), but different types exist. This research concerns ventricular arrhythmia of the outflow tract, the region that connects the ventricles with the main arteries of the heart. Outflow tract ventricular arrhythmia (OTVA) is the most common manifestation of so-called idiopathic ventricular arrhythmias, whose causes cannot be identified by means of conventional methods or in patients without structural heart disease, hence it is more difficult to specify the reasons.

Currently, apart from medication to correct heart rhythm alterations, the most common treatment for OTVA is radiofrequency ablation. To apply this technique, it is first necessary to map the electrical circuit that causes cardiac arrhythmia to then position the radiofrequency emitting catheter in the area of the disorder. This generates the temperature increase required to remove the specific part of heart tissue where the arrhythmia is triggered.

Higher probabilities of success and shorter treatment duration of some ventricular arrhythmias, among the objectives of the new AI-based model

To date, radiofrequency ablation treatment has proved less effective than would be desirable. To improve it, the place of origin of the arrhythmia needs to be better established, so that the radiofrequency-emitting catheter can act on the exact area where it originates. This would increase the chances of successful treatment and reduce intervention times and relapse rates.

This is the challenge addressed in the scientific article recently published in the journal Frontiers in Cardiovascular Medicine. On behalf of the Physense research group of the UPF BCN Medtech Research Unit, Álvaro J.  Bocanegra-Pérez (lead author), Gemma Piella, Guillermo Jimenez-Perez and Oscar Cámara (principal investigator) have participated in the research. On behalf of Centre Mèdic Teknon de Barcelona, also co-authoring the paper are Giulio Falasconi, David Soto-Iglesias and Antonio Berruezo. Other co-authors are Rafael Sebastian, from the CoMMLab  at the University of Valencia; Andrea Saglietto (Department of Medical Sciences at the University of Turin), and Diego Penela (Humanitas Research Hospital in Milan).

In the article, they propose a model based on AI and machine learning, which can contribute to substantially improving the accuracy of current diagnoses and treatments of OVTA. Current diagnostic methods are fundamentally based on the analysis of electrocardiograms (ECGs) performed prior to the operation, through visual inspection by medical professionals. Despite their experience, visual inspection is subject to human error and can lead to incorrect or inaccurate diagnoses, which may in turn reduce treatment effectiveness to below optimal levels.

In recent years, more advanced methods have been developed to try to overcome the limitations of the visual inspection of electrocardiograms using computational models and machine learning (ML) approaches. However, the methods devised so far still have limitations when it comes to determining precisely where OTVAs originate or with regard to obtaining specific, personalized information from each patient or the interpretation and application of results in real clinical scenarios.

An AI model to analyse clinical data and electrocardiograms together, thus enhancing the accuracy of diagnoses

To overcome these limitations, the study led by the UPF Physense research group proposes a holistic approach that will allow more effective diagnoses and treatments of OTVA. Based on AI and machine learning techniques, it will be possible to analyse real clinical data regarding the patient’s age, sex and medical history in an integrated and automatic manner, -especially concerning whether or not they have previously suffered from high blood pressure- and electrocardiograms (both real and simulated by computational methods). This will make it possible to establish the place of origin of the arrhythmias in each particular case, reduce the margin of error with respect to visual inspections, and facilitate the interpretation of the results.

Regarding the contributions of the research, Álvaro J. Bocanegra-Pérez, a researcher with the UPF Physense research group, says: “The proposed method has proved effective in differentiating OTVAs of left and right origin. The methodology used makes the system robust and ensures interpretability for any subsequent analysis, for example, the identification of the specific place of origin of the arrhythmia. This multimodal and interpretable approach is key to multidisciplinary work between physicians and engineers, as it allows making contributions to the methodology by both parties”.

The research conducted has demonstrated the effectiveness of this method in ventricular arrhythmias originating in both right and left ventricles, based on a study that combined the analysis of 2,496 simulated cases with that of real patients. From this latter group, the cases of 114 patients from Teknon Hospital in Barcelona, 31 from Hospital Clínic de Barcelona and a further 334 corresponding to a study carried out in China (Zheng et al.) have been examined. Beyond this research, it will be necessary to continue working in this line of investigation, analysing data from a greater number of patients to develop a more robust system that might be generalized to clinical practice.

Reference article:

Álvaro J. Bocanegra-Pérez, Gemma Piella, Rafael Sebastian, Guillermo Jimenez-Perez, Giulio Falasconi, Andrea Saglietto, David Soto-Iglesias, Antonio Berruezo, Diego Penela and Oscar Camara. Automatic and interpretable prediction of the site of origin in outflow tract ventricular arrhythmias: machine learning integrating electrocardiograms and clinical data. Front. Cardiovasc. Med., 20 March 2024, Sec. Cardiac Rhythmology, Volume 11–2024. https://doi.org/10.3389/fcvm.2024.1353096