This contribution addresses the extraction of atrial activity (AA) from real electrocardiogram (ECG) recordings of atrial fibrillation (AF). We show the appropriateness of independent component analysis (ICA) to tackle this biomedical challenge when regarded as a blind source separation (BSS) problem. ICA is a statistical tool able to reconstruct the unobservable independent sources of bioelectric activity which generate, through instantaneous linear mixing, a measurable set of signals. The three key hypothesis that make ICA applicable in the present scenario are discussed and validated: 1) AA and ventricular activity (VA) are generated by sources of independent bioelectric activity; 2) AA and VA present non-Gaussian distributions; and 3) the generation of the surface ECG potentials from the cardioelectric sources can be regarded as a narrow-band linear propagation process. To empirically endorse these claims, an ICA algorithm is applied to recordings from seven patients with persistent AF. We demonstrate that the AA source can be identified using a kurtosis-based reordering of the separated signals followed by spectral analysis of the sub-Gaussian sources. In contrast to traditional methods, the proposed BSS-based approach is able to obtain a unified AA signal by exploiting the atrial information present in every ECG lead, which results in an increased robustness with respect to electrode selection and placement.

Atrial fibrillation (AF) is the most common arrhythmia encountered in clinical practice. Neither the natural history of AF nor its response to therapy is sufficiently predictable by clinical and echocardiographic parameters. The purpose of this article is to describe technical aspects of novel electrocardiogram (ECG) analysis techniques and to present research and clinical applications of these methods for characterization of both the fibrillatory process and the ventricular response during AF. Atrial fibrillatory frequency (or rate) can reliably be assessed from the surface ECG using digital signal processing (extraction of atrial signals and spectral analysis). This measurement shows large inter-individual variability and correlates well with intra-atrial cycle length, a parameter which appears to have primary importance in AF maintenance and response to therapy. AF with a low fibrillatory rate is more likely to terminate spontaneously and responds better to antiarrhythmic drugs or cardioversion, whereas high-rate AF is more often persistent and refractory to therapy. Ventricular responses during AF can be characterized by a variety of methods, which include analysis of heart rate variability, RR-interval histograms, Lorenz plots, and non-linear dynamics. These methods have all shown a certain degree of usefulness, either in scientific explorations of atrioventricular (AV) nodal function or in selected clinical questions such as predicting response to drugs, cardioversion, or AV nodal modification. The role of the autonomic nervous system for AF sustenance and termination, as well as for ventricular rate responses, can be explored by different ECG analysis methods. In conclusion, non-invasive characterization of atrial fibrillatory activity and ventricular response can be performed from the surface ECG in AF patients. Different signal processing techniques have been suggested for identification of underlying AF pathomechanisms and prediction of therapy efficacy.

Heart sounds have been widely studied and have been demonstrated to have value for detecting pathologies in clinical applications. Over the last few decades, the use of heart sound signals has become increasingly uncommon and its practice in modern medicine somewhat diminished, although research into automated analysis has continued. Unfortunately, a comparative analyses of algorithms in the literature have been hindered by the lack of high-quality, rigorously validated, and standardized open databases of heart sound recordings. The 2016 PhysioNet/Computing in Cardiology (CinC) Challenge addressed this issue by assembling the largest public heart sound database, aggregated from eight sources obtained by seven independent research groups around the world. The database comprises a total of 4,430 recordings collected from 1,072 healthy subjects and patients with a variety of conditions, including heart valve disease and coronary artery disease. This editorial reviews the background issues for this Challenge, the design of the Challenge itself, the key achievements, and the follow-up research generated as a result of the Challenge, published in the concurrent special issue of Physiological Measurement. Additionally we make some recommendations for future changes in this the field of heart sound signal processing as a result of the Challenge. In the Challenge, participants were asked to classify recordings as normal, abnormal, or unsure. The overall score for an entry was based on a weighted sensitivity and specificity score with respect to manual expert annotations. To aid researchers, we provided a simple baseline classification method and a complex open source code base for segmenting the heart sounds, based on a hidden semi-Markov model. During the official phase of the Challenge, a total of 48 teams submitted 348 open source entries, with a highest score of 0.860 (Se=0.942, Sp=0.778). Subsequently, for this special issue, researchers reported the new highest score of 0.855 (Se=0.890, Sp=0.816) in the follow-up phase of the Challenge, indicating that the Challenge entrants achieved exceptional results which were extremely dicult to improve (even when there is a trade-off between Sp and Se) upon in the 4 months available post-Challenge. We expect that future researchers will be able to use the extensive database generated for the Challenge to significantly improve on the approaches detailed here.

Ablation is an effective therapy in patients with atrial fibrillation (AF) in which an electrical driver can be identified. The aim of this study was to present and discuss a novel and strictly noninvasive approach to map and identify atrial regions responsible for AF perpetuation. Surface potential recordings of 14 patients with AF were recorded using a 67-lead recording system. Singularity points (SPs) were identified in surface phase maps after band-pass filtering at the highest dominant frequency (HDF). Mathematical models of combined atria and torso were constructed and used to investigate the ability of surface phase maps to estimate rotor activity in the atrial wall. The simulations show that surface SPs originate at atrial SPs, but not all atrial SPs are reflected at the surface. Stable SPs were found in AF signals during 8.3% ± 5.7% vs. 73.1% ± 16.8% of the time in unfiltered vs. HDF-filtered patient data, respectively (P < .01). The average duration of each rotational pattern was also lower in unfiltered than in HDF-filtered AF signals (160 ± 43 ms vs. 342 ± 138 ms; P < .01), resulting in 2.8 ± 0.7 rotations per rotor. Band-pass filtering reduced the apparent meandering of surface HDF rotors by reducing the effect of the atrial electrical activity occurring at different frequencies. Torso surface SPs representing HDF rotors during AF were reflected at specific areas corresponding to the fastest atrial location. Phase analysis of surface potential signals after HDF filtering during AF shows reentrant drivers localized to either the left atrium or the right atrium, helping in localizing ablation targets.

Aims: Artificial intelligence (AI) has the potential to transform cardiac electrophysiology (EP), particularly in arrhythmia detection, procedural optimization, and patient outcome prediction. However, a standardized approach to reporting and understanding AI-related research in EP is lacking. This scientific statement aims to develop and apply a checklist for AI-related research reporting in EP to enhance transparency, reproducibility, and understandability in the field. Methods and results: An AI checklist specific to EP was developed with expert input from the writing group and voted on using a modified Delphi process, leading to the development of a 29-item checklist. The checklist was subsequently applied to assess reporting practices to identify areas where improvements could be made and provide an overview of the state of the art in AI-related EP research in three domains from May 2021 until May 2024: atrial fibrillation (AF) management, sudden cardiac death (SCD), and EP lab applications. The EHRA AI checklist was applied to 31 studies in AF management, 18 studies in SCD, and 6 studies in EP lab applications. Results differed between the different domains, but in no domain reporting of a specific item exceeded 55% of included papers. Key areas such as trial registration, participant details, data handling, and training performance were underreported (<20%). The checklist application highlighted areas where reporting practices could be improved to promote clearer, more comprehensive AI research in EP. Conclusion: The EHRA AI checklist provides a structured framework for reporting AI research in EP. Its use can improve understanding but also enhance the reproducibility and transparency of AI studies, fostering more robust and reliable integration of AI into clinical EP practice.

Background—: Ablation of high-frequency sources in patients with atrial fibrillation (AF) is an effective therapy to restore sinus rhythm. However, this strategy may be ineffective in patients without a significant dominant frequency (DF) gradient. The aim of this study was to investigate whether sites with high-frequency activity in human AF can be identified noninvasively, which should help intervention planning and therapy. Methods and Results—: In 14 patients with a history of AF, 67-lead body surface recordings were simultaneously registered with 15 endocardial electrograms from both atria including the highest DF site, which was predetermined by atrial-wide real-time frequency electroanatomical mapping. Power spectra of surface leads and the body surface location of the highest DF site were compared with intracardiac information. Highest DFs found on specific sites of the torso showed a significant correlation with DFs found in the nearest atrium (ρ=0.96 for right atrium and ρ=0.92 for left atrium) and the DF gradient between them (ρ=0.93). The spatial distribution of power on the surface showed an inverse relationship between the frequencies versus the power spread area, consistent with localized fast sources as the AF mechanism with fibrillatory conduction elsewhere. Conclusions—: Spectral analysis of body surface recordings during AF allows a noninvasive characterization of the global distribution of the atrial DFs and the identification of the atrium with the highest frequency, opening the possibility for improved noninvasive personalized diagnosis and treatment.