Search

Funding Acknowledgements Type of funding sources: Public grant(s) – National budget only. Main funding source(s): NWO Rubicon (452019308) Amsterdam Cardiovascular Sciences Background Left Ventricular Ejection Fraction (LVEF) is suboptimal as a sole marker for predicting sudden cardiac death (SCD) or benefit from an ICD. Machine (ML) and deep (DL) learning provide new opportunities for personalised predictions using complex, multi-modal physiological data. Objective We hypothesise that risk stratification for ICD implantation can be improved by ML and DL models that combine clinical variables with time series features from 12-lead electrocardiograms (ECG). Methods We present a multicentre study of 1010 patients with an ischaemic, dilated or non-ischaemic cardiomyopathy and LVEF≤35% implanted with an ICD between 2007 and 2021 for primary prevention of SCD (64.9 ±10.8 years, 73.2% male) in two academic hospitals. For each patient, raw 12-lead, 10-second ECG-recordings obtained Results At 3-year follow-up, 16.0% of patients had died of whom 72.8% met the criteria for ICD non-benefit. Extreme gradient boosting models identified subjects with ICD non-benefit with an area under the receiver operator curve (AUROC) of 0.897 ±0.05 during internal validation (Figure 1, solid line). In the external cohort, the AUROC was 0.793 (95% CI 0.75-0.84) (Figure 1, dashed line). Survival analysis for low vs. high predicted risk indicated ICD non-benefit rates of 6.0% versus 30.3% at 3-year follow-up, respectively (Figure 2). Conclusion A ML model that combined clinical with ECG features better predicted ICD non-benefit at 3-years in a primary prevention population than currently available risk scores such as the MADIT-ICD score. This approach may provide new tools to support personalized decision making for ICD therapy. Prospective validation is needed to assess the real-world clinical performance of the model.

Background Atrial fibrillation (AF) can have organized regions, in the form of consistent dominant frequency sites, focal or reentrant sites, but it is unclear how these overlap with or differ from focal atrial tachycardias (AT) or potential drivers. We set out to develop an intuitive method based on fundamental electrogram shape and timing to separate types of AF. Objective To test the hypothesis that spatial regions of electrogram (EGM) in AF that show similar shapes over time based on cross-correlation analysis may separate patients with differing response to ablation. Methods We recruited N=133 patients (63.8±12.1 Y, 32% women), (i) N=10 had AT, (ii) N=122 AF that was or was not terminated by ablation, and (iii) N=1 pacing. All patients had left atrial mapping by 64 pole baskets. We applied repetitive activity (REACT) mapping that correlates EGMs in contiguous 2x2 regions (Fig. 1A) over 4sec. To calibrate REACT, we introduced simulated variations in shape (gaussian noise) and timing (gaussian delay) to pacing EGMs and computed nomograph over 100 random trials (Fig. 1C). Results Fig. 1B shows that REACT in a 71-year-old man with AT is more organized than in a 65 YO man with AF (100% vs 40% mapped field). Overall, REACT was higher in AT than AF (0.63±0.15 vs 0.36±0.22, p Conclusion Basic electrogram properties in AF of similar shapes in spatial areas over time can separate response to ablation and may represent “islands” of AT. Future studies should investigate the mechanisms for such islands and whether they may be targeted for therapy. Funding Acknowledgement Type of funding sources: Public grant(s) – National budget only. Main funding source(s): US National Institutes of Health

Background/Introduction Signals in Electrophysiology cases are often noisy despite laboratory shielding and filtering, and current noise-reduction methods are suboptimal. Template matching can identify a “nearest type” of electrogram, but libraries of signal shapes may be unavailable. Beat averaging can reduce noise but obscures beat-to-beat variations and is not optimal to analyze dynamically changing signals, such as when moving a catheter in the heart. Smoothing reduces noise yet blurs high frequency components. Purpose We set out to test if machine learned autoencoders could reduce noise in single beats without requiring massive training data or beat libraries. Specifically, we hypothesised that noisy electrograms in small datasets of atrial signals could be de-noised using an encoder-decoder neural network (NN) using transfer learning of machines trained to recognize key features in larger datasets of related signals. Methods We applied NN to monophasic action potentials (MAPs), because they have visually verifiable shapes. The NN was first trained to reconstruct 5706 left and right ventricular MAPs in 42 patients (67±13y; Fig. 1A). Transfer learning was then used to apply the NN to a much smaller dataset of 641 atrial MAPs in 21 patients (67±5y, 13 women; Fig. 1B, D, F). Results NN reconstructed atrial MAPs with a Pearson correlation of 0.87±0.11. After fine-tuning, NN reconstruction accuracy improved dramatically (Pearson 0.99±0.01; p Conclusion Machine learned autoencoders are a novel and powerful approach to de-noise electrophysiological signals in a dynamic, beat-to-beat fashion. The ability to learn fundamental signal features from models trained in large datasets, and apply them via transfer learning to small datasets in different heart chambers may have wide ranging applications for automated signal annotation, mapping and ablation. Funding Acknowledgement Type of funding sources: Public grant(s) – National budget only. Main funding source(s): NIH

Background/Introduction Mapping AF is challenging. Monophasic action potentials (MAPs) show that most of the recorded signals are not representing the atrial activity, and arise from far-field or other artifacts. Thus, reducing noise in electrophysiological signals is essential, yet it can be difficult for cross-talk from other chambers and pacing. Strategies to reduce noise include template matching, averaging, and smoothing, but all of them have major limitations. Furthermore, expert interpretation requires knowledge to discriminate signals from noise, but is subjective. Purpose We hypothesised a) that atrial and ventricular electrograms with varying artifact and noise can be denoised using autoencoder neural networks (NNs) without requiring manual annotation and in a reproducible manner, and b) we could train these NNs on a large available dataset ventricular signals, then apply transfer learning to the original smaller atrial dataset. We applied this approach to MAPs, which have interpretable shapes and would help identifying local from far-field signals helping in diagnosis, mapping and ablation. Methods We first trained with 5706 left and right ventricular MAPs from 42 patients with ischemic cardiomyopathy (age 65±13y; Fig. 1A), with 60% for training, 20% (validation) and 20% (testing). Transfer learning and parameter-tuning were then used to apply this NN to a smaller sample of atrial MAPs (N=641 from 21 patients, 67±5y, 13 women; Fig. 2B, D, F). The autoencoder was used to eliminate pacing artifacts in ventricular MAPs (Fig. 1B, C) and denoise atrial MAPs (Fig. 2C, E, G) by reconstructing key learned features. The accuracy of the reconstruction was evaluated using Pearson Correlation Coefficient (PCC) and a novel similarity coefficient (SC). No manual annotation was required to identify noisy signals. Results The trained NN encoder learned key features of ventricular MAPs and reconstructed these clean signals with a SC=0.91±0.16 and PCC=0.99±0.01 (Fig. 1A). With this training, the NN was able to denoise ventricular MAPs with pacing artifact (Fig. 1B, C). After fine-tuning, the NN learned key signal features (upstroke, triangular descent, terminus) and thus reduced diverse noise without specific training or manual annotation. Namely, it was able to reconstruct atrial MAPs eliminating ventricular noise, high frequency noise and truncated signals (Fig. 2). Conclusions Machine learned encoder-decoders are powerful tools that can learn essential features of atrial and ventricular signals and hence isolate noise. Transfer learning is effective when large datasets are unavailable for training. This approach can separate atrial beats from far-field ventricular beats and other sources of noise. The ability to eliminate a diverse range of noise improves this approach over existing techniques and may have far-reaching applications in electrophysiology, such as mapping and ablation. Funding Acknowledgement Type of funding sources: Public grant(s) – National budget only. Main funding source(s): NIH

Introduction Segmenting left atrial (LA) substructures, including the LA body, appendage (LAA), and pulmonary veins (PVs), from computed tomography (CT) is central to electroanatomic mapping for ablation and functional studies in patients with atrial fibrillation (AF). However, this process requires manual outlining which needs special training, is subjective, and is difficult to scale. Computer graphics imaging (CGI) has been applied in media, film, and computer-aided design to reliably segment complex structures using their basic geometric representations. Purpose We hypothesized that LA substructures can be “virtually” dissected using CGI to separate geometric contours of the “convex ellipsoid” LA, “tubular” PVs, and “conical” LAA. We further hypothesized that the results of virtual dissection can be used to train a deep learning (DL) model to segment raw CT scans. Methods First, a mathematical method based on CGI techniques – erosion and dilation – was developed to “virtually dissect” the convex LA body from the original concave shell in publicly available digital atria with diverse simulated morphologies (Fig. 1A). The PVs and LAA were then automatically revealed and labeled by a 3D subtraction approach. Second, we refined precise LA/PV/LAA boundaries by tuning hyper-parameters from N=5 patient shells (Fig. 1B). Third, we used virtual dissection to train a DL model to segment CTs in N=20 patient atria (Fig. 1C). Finally, we applied this pipeline to segment raw CTs in a validation cohort of N=105 patients (23.8% women, 63.8±10.3Y; Fig. 1D). Results Virtual dissection accurately identified LA/PV/LAA boundaries in the training set (Dice coefficients 89–98%). In the independent test cohort (N=105), this automated pipeline accurately segmented raw CTs with Dice 81–95% (Fig. 1D) compared to a panel of experts (p Conclusion CGI of basic cardiac geometry combined with deep learning in small datasets can accurately segment raw CT scans in large populations. This computational pipeline may automate and simplify cardiac image processing and ablation procedures, and could be applied to the ventricle or other organ systems for diverse therapeutic strategies or to train machine learning. Funding Acknowledgement Type of funding sources: Public grant(s) – National budget only. Main funding source(s): National Institutes of Health

Background Although atrial tachycardia (AT) may interconvert with fibrillation (AF) in many patients, it is undefined if this represents a pathophysiological spectrum of organization, or whether it indicates that AF will respond better to ablation. Objective To test the hypothesis that the spatial area within which electrograms (EGMs) repeat in synchronized fashion over time indicates a spectrum from AT, in which areas span the entire atria, to AF, in which areas are limited. We further hypothesized that repetitive areas would be larger in AF patients with acute termination than in those with poor response to ablation. Methods We studied N=234 patients (47% women, 64±10Y), of whom (i) N=10 had AT, (ii) N=120 had AF that terminated with ablation (“Term”), (ii) N=104 had AF that did not terminate (“Non-term”). All patients had global left atrial mapping by 64 pole baskets (Abbott, IL). Spatial areas of repetitive activity (REACT) were calculated by correlating unipolar EGMs in 2x2 grids for 4 sec, repeated for the entire atria (Figure 1A, B). We quantified global organization by averaging the REACT map for each patient. Results Figure 1C shows progressively decreasing areas of repetitive EGM from AT to AF Term to AF Non-term (p0.90, a case of AF REACT 0.45 in a 65 YO male with termination by ablation, and a case of AF with REACT 0.19 in an 85 YO male that did not terminate. Further, ROC analysis of REACT analysis in AF cases predicted termination with an AUC of 0.71. Conclusion Spatial areas of repeating electrogram shapes indicates a spectrum from AT to AF with good and AF with poor acute response to ablation. Future studies should investigate whether REACT areas can be identified non-invasively, such as by body surface ECG, to guide ablation or prognosis. Funding Acknowledgement Type of funding sources: Public grant(s) – National budget only. Main funding source(s): US National Institutes of Health

Background Several mapping systems are being introduced to guide atrial fibrillation (AF) ablation to patient-specific regions of interest. However, results have been extremely heterogeneous between studies, ranging from very poor to very promising. It is unknown if this reflects specific patient characteristics or procedural factors because most prior series were middle sized (N∼30–100 patients). Purpose To study 1 year and 3 year very long-term outcomes from map guided AF-driver ablation in a large patient registry with multiple operators, to identify clinical and procedural features influencing outcomes. In real-world AF patients with diverse comorbidities, we applied a consistent patient-tailored AF mapping and ablation strategy, monitored outcomes carefully and applied statistical and unsupervised machine learning approaches to identify features of success and failure. Method We studied 632 consecutive patients (65±10 y, 178 F) undergoing ablation for drug-refractory AF. 59.7% had persistent AF, and 29.9% had prior unsuccessful ablation (median 1 procedure). All patients underwent pulmonary vein isolation (PVI), followed by ablation of AF regions of interest mapped from 64 pole baskets (RhythmView, Abbott, IL), by 11 operators. Patients were followed using ambulatory ECG monitors quarterly for one year, and at the time of symptoms for 3 years. Results Fig. 1A shows overall freedom from AF at 1-year of 77.5% (95% CI: 74.2%, 80.9%) and at 3 years of 55.5% (95% CI: 51.2%, 60.1%). Freedom from AF/AT at 1-year was 70.1% (95% CI: 66.5%, 73.8%), and at 3 years was 48.6% (95% CI: 44.3%, 53.3%). Success was higher in patients with procedural termination, first ablation versus prior unsuccessful procedures, for paroxysmal AF than non-paroxysmal AF (1 year: AT/AF freedom 74.9% versus 66.7%, p=0.006), and smaller left atrium. Three clusters (Fig 1B) were identified comprising CHA2DS2VASc score, enlarged LA, prior failed case, presenting rhythm and termination during the procedure (Table 2). At 1 year, freedom from AT/AF was 77.8% (95% CI: 72.2%, 82.1%) for cluster 3 and 56.2% (95% CI: 48.3%, 65.4%) for cluster 1 (Fig. 1B). Conclusion In our large registry of N=632 patients undergoing AF-map guided ablations, machine learned clusters identified cohorts with success of 56.2 to 77.8% at 1 year. Future studies should identify if lower success represents technical challenges, such as difficulties in mapping very large atria, or more difficult to treat mechanisms. These results may inform patient inclusion and ablation strategy in upcoming AF treatment trials. Funding Acknowledgement Type of funding sources: Public grant(s) – National budget only. Main funding source(s): National budget only - NIH, R01 HL149134, R01HL83359

Background/Introduction Over the past decade, various studies have demonstrated an association between remote device monitoring (RM) in implantable cardioverter-defibrillator (ICD) patients and all-cause mortality. However, it is unclear which clinical phenotypes account for this survival benefit. Purpose We aimed to identify clinical phenotypes within an ICD population on RM and conventional follow-up (NRM), and evaluate differences in long-term survival per clinical phenotype. Methods This is a single-center, retrospective, observational study of de novo ICD implantations (single- and dual chamber, biventricular and subcutaneous) between 2010 and 2021 in patients ≥18 years old. Data was extracted from electronic health records. Unsupervised two-step cluster analysis (TSC) was performed to identify distinct clinical phenotypes in the RM and NRM cohorts. Multinomial logistic regression analysis was performed to identify cluster characteristics, differences among the different clusters were analysed by Chi-squared test. Variable importance was evaluated for each of the 18 included clinical variables. Survival analysis was performed using the Kaplan-Meier method, log-rank was used to compare survival between groups. Results A total of 1872 ICD patients were analysed using TCS, comprising 1265 RM and 607 NRM patients, respectively. TCS for a mean follow-up duration of 5.4±3.1 years partitioned the RM and NRM groups into six clusters (Table 1). Clusters RM1 (n=444) and NRM3 (n=220) represented a predominantly primary prevention and DCM phenotype. Clusters RM2 (n=300) and NRM2 (n=173) indicated young and secondary prevention phenotypes. Clusters RM3 (n=220) and NRM1 (n=214) comprised an older, male and ischaemic cardiomyopathy (ICM) phenotype. In survival analysis 5-year cumulative incidence of mortality for the subsequent clusters was 7.4%, 4.5%, 11.6%, 41.4%, 29.8%, and 25.1%, respectively (log rank p-value Conclusion We identify novel clinical phenotypes (clusters) of ICD patients on RM and having conventional follow-up with large differences in long-term survival. Future studies should further explore if these differences reflect benefit from RM, or the underlying natural history of each phenotype. These results may accordingly guide monitoring strategies tailored to specific ICD patient profiles. Funding Acknowledgement Type of funding sources: Public hospital(s). Main funding source(s): Public funding

Sudden cardiac death (SCD) from ventricular tachycardia (VT) or ventricular fibrillation (VF) remains a major cause of mortality in the U.S. and Western Europe, and claims 2–3 lives per 100,000 per year [1, 2]. The incidence of SCD has not declined in tandem with significant advances in the management of acute coronary syndromes, congestive heart failure (CHF) and cardiovascular disease in general, further highlighting the importance of identifying and treating patients at high risk for VT/VF.

Line of reasoning (LOR) is offered as an alternative representation of clinical decision making for studies using protocol analysis. A LOR is defined as an argument or set of arguments leading to a conclusion. Because LOR combines both knowledge and cognitive processes, it provides a more complete representation of how a person uses knowledge to make a decision in a particular situation than do other representations. Operationalization of LOR in the form of templates and narratives enhances systematic data interpretation and coding. The use of LOR as a representation is illustrated in a study of critical care nurses' clinical decision making, specifically the determination of a patient's readiness to wean from mechanical ventilation.