Aims Atrial flutter (AFlut) is a common re-entrant atrial tachycardia driven by self-sustainable mechanisms that cause excitations to propagate along pathways different from sinus rhythm. Intra-cardiac electrophysiological mapping and catheter ablation are often performed without detailed prior knowledge of the mechanism perpetuating AFlut, likely prolonging the procedure time of these invasive interventions. We sought to discriminate the AFlut location [cavotricuspid isthmus-dependent (CTI), peri-mitral, and other left atrium (LA) AFlut classes] with a machine learning-based algorithm using only the non-invasive signals from the 12-lead electrocardiogram (ECG). Methods and results Hybrid 12-lead ECG dataset of 1769 signals was used (1424 in silico ECGs, and 345 clinical ECGs from 115 patients—three different ECG segments over time were extracted from each patient corresponding to single AFlut cycles). Seventy-seven features were extracted. A decision tree classifier with a hold-out classification approach was trained, validated, and tested on the dataset randomly split after selecting the most informative features. The clinical test set comprised 38 patients (114 clinical ECGs). The classifier yielded 76.3% accuracy on the clinical test set with a sensitivity of 89.7%, 75.0%, and 64.1% and a positive predictive value of 71.4%, 75.0%, and 86.2% for CTI, peri-mitral, and other LA class, respectively. Considering majority vote of the three segments taken from each patient, the CTI class was correctly classified at 92%. Conclusion Our results show that a machine learning classifier relying only on non-invasive signals can potentially identify the location of AFlut mechanisms. This method could aid in planning and tailoring patient-specific AFlut treatments.
OBJECTIVE: Atrial flutter (AFl) is a common arrhythmia that can be categorized according to different self-sustained electrophysiological mechanisms. The non-invasive discrimination of such mechanisms would greatly benefit ablative methods for AFl therapy as the driving mechanisms would be described prior to the invasive procedure, helping to guide ablation. In the present work, we sought to implement recurrence quantification analysis (RQA) on 12-lead ECG signals from a computational framework to discriminate different electrophysiological mechanisms sustaining AFl. METHODS: 20 different AFl mechanisms were generated in 8 atrial models and were propagated into 8 torso models via forward solution, resulting in 1,256 sets of 12-lead ECG signals. Principal component analysis was applied on the 12-lead ECGs, and six RQA-based features were extracted from the most significant principal component scores in two different approaches: individual component RQA and spatial reduced RQA. RESULTS: In both approaches, RQA-based features were significantly sensitive to the dynamic structures underlying different AFl mechanisms. Hit rate as high as 67.7% was achieved when discriminating the 20 AFl mechanisms. RQA-based features estimated for a clinical sample suggested high agreement with the results found in the computational framework. CONCLUSION: RQA has been shown an effective method to distinguish different AFl electrophysiological mechanisms in a non-invasive computational framework. A clinical 12-lead ECG used as proof of concept showed the value of both the simulations and the methods. SIGNIFICANCE: The non-invasive discrimination of AFl mechanisms helps to delineate the ablation strategy, reducing time and resources required to conduct invasive cardiac mapping and ablation procedures.
The arrhythmogenesis of atrial fibrillation is associated with the presence of fibrotic atrial tissue. Not only fibrosis but also physiological anatomical variability of the atria and the thorax reflect in altered morphology of the P wave in the 12-lead electrocardiogram (ECG). Distinguishing between the effects on the P wave induced by local atrial substrate changes and those caused by healthy anatomical variations is important to gauge the potential of the 12-lead ECG as a non-invasive and cost-effective tool for the early detection of fibrotic atrial cardiomyopathy to stratify atrial fibrillation propensity. In this work, we realized 54,000 combinations of different atria and thorax geometries from statistical shape models capturing anatomical variability in the general population. For each atrial model, 10 different volume fractions (0-45%) were defined as fibrotic. Electrophysiological simulations in sinus rhythm were conducted for each model combination and the respective 12-lead ECGs were computed. P wave features (duration, amplitude, dispersion, terminal force in V1) were extracted and compared between the healthy and the diseased model cohorts. All investigated feature values systematically in- or decreased with the left atrial volume fraction covered by fibrotic tissue, however value ranges overlapped between the healthy and the diseased cohort. Using all extracted P wave features as input values, the amount of the fibrotic left atrial volume fraction was estimated by a neural network with an absolute root mean square error of 8.78%. Our simulation results suggest that although all investigated P wave features highly vary for different anatomical properties, the combination of these features can contribute to non-invasively estimate the volume fraction of atrial fibrosis using ECG-based machine learning approaches.
In patients with atrial fibrillation, intracardiac electrogram signal amplitude is known to decrease with increased structural tissue remodeling, referred to as fibrosis. In addition to the isolation of the pulmonary veins, fibrotic sites are considered a suitable target for catheter ablation. However, it remains an open challenge to find fibrotic areas and to differentiate their density and transmurality. This study aims to identify the volume fraction and transmurality of fibrosis in the atrial substrate. Simulated cardiac electrograms, combined with a generalized model of clinical noise, reproduce clinically measured signals. Our hybrid dataset approach combines and clinical electrograms to train a decision tree classifier to characterize the fibrotic atrial substrate. This approach captures different dynamics of the electrical propagation reflected on healthy electrogram morphology and synergistically combines it with synthetic fibrotic electrograms from experiments. The machine learning algorithm was tested on five patients and compared against clinical voltage maps as a proof of concept, distinguishing non-fibrotic from fibrotic tissue and characterizing the patient's fibrotic tissue in terms of density and transmurality. The proposed approach can be used to overcome a single voltage cut-off value to identify fibrotic tissue and guide ablation targeting fibrotic areas.
Atrial flutter (AFL) is a common atrial arrhythmia typically characterized by electrical activity propagating around specific anatomical regions. It is usually treated with catheter ablation. However, the identification of rotational activities is not straightforward, and requires an intense effort during the first phase of the electrophysiological (EP) study, i.e., the mapping phase, in which an anatomical 3D model is built and electrograms (EGMs) are recorded. In this study, we modeled the electrical propagation pattern of AFL (measured during mapping) using network theory (NT), a well-known field of research from the computer science domain. The main advantage of NT is the large number of available algorithms that can efficiently analyze the network. Using directed network mapping, we employed a cycle-finding algorithm to detect all cycles in the network, resembling the main propagation pattern of AFL. The method was tested on two subjects in sinus rhythm, six in an experimental model of in-silico simulations, and 10 subjects diagnosed with AFL who underwent a catheter ablation. The algorithm correctly detected the electrical propagation of both sinus rhythm cases and in-silico simulations. Regarding the AFL cases, arrhythmia mechanisms were either totally or partially identified in most of the cases (8 out of 10), i.e., cycles around the mitral valve, tricuspid valve and figure-of-eight reentries. The other two cases presented a poor mapping quality or a major complexity related to previous ablations, large areas of fibrotic tissue, etc. Directed network mapping represents an innovative tool that showed promising results in identifying AFL mechanisms in an automatic fashion. Further investigations are needed to assess the reliability of the method in different clinical scenarios.
Computer modeling of the electrophysiology of the heart has undergone significant progress. A healthy heart can be modeled starting from the ion channels via the spread of a depolarization wave on a realistic geometry of the human heart up to the potentials on the body surface and the ECG. Research is advancing regarding modeling diseases of the heart. This article reviews progress in calculating and analyzing the corresponding electrocardiogram (ECG) from simulated depolarization and repolarization waves. First, we describe modeling of the P-wave, the QRS complex and the T-wave of a healthy heart. Then, both the modeling and the corresponding ECGs of several important diseases and arrhythmias are delineated: ischemia and infarction, ectopic beats and extrasystoles, ventricular tachycardia, bundle branch blocks, atrial tachycardia, flutter and fibrillation, genetic diseases and channelopathies, imbalance of electrolytes and drug-induced changes. Finally, we outline the potential impact of computer modeling on ECG interpretation. Computer modeling can contribute to a better comprehension of the relation between features in the ECG and the underlying cardiac condition and disease. It can pave the way for a quantitative analysis of the ECG and can support the cardiologist in identifying events or non-invasively localizing diseased areas. Finally, it can deliver very large databases of reliably labeled ECGs as training data for machine learning.
Atrial fibrillation (AF) is an irregular heart rhythm due to disorganized atrial electrical activity, often sustained by rotational drivers called rotors. In the present work, we sought to characterize and discriminate whether simulated single stable rotors are located in the pulmonary veins (PVs) or not, only by using non-invasive signals (i.e., the 12-lead ECG). Several features have been extracted from the signals, such as Hjort descriptors, recurrence quantification analysis (RQA), and principal component analysis. All the extracted features have shown significant discriminatory power, with particular emphasis to the RQA parameters. A decision tree classifier achieved 98.48% accuracy, 83.33% sensitivity, and 100% specificity on simulated data. Clinical relevance— This study might guide ablation proce- dures, suggesting doctors to proceed directly in some patients with a pulmonary veins isolation, and avoiding the prior use of an invasive atrial mapping system.
Atrial fibrillation (AF) is the most frequent irregular heart rhythm due to disorganized atrial electrical activity, often sustained by rotational drivers called rotors. The non-invasive localization of AF drivers can lead to improved personalized ablation strategy, suggesting pulmonary vein (PV) isolation or more complex extra- PV ablation procedures in case the driver is on other atrial regions. We used a Machine Learning approach to characterize and discriminate simulated single stable rotors (1R) location: PVs, left atrium (LA) excluding the PVs, and right atrium (RA), utilizing solely non-invasive signals (i.e., the 12-lead ECG). 1R episodes sustaining AF were simulated. 128 features were extracted from the signals. Greedy forward algorithm was implemented to select the best feature set which was fed to a decision tree classifier with hold-out cross-validation technique. All tested features showed significant discriminatory power, especially those based on recurrence quantification analysis (up to 80.9% accuracy with single feature classification). The decision tree classifier achieved 89.4% test accuracy with 18 features on simulated data, with sensitivities of 93.0%, 82.4%, and 83.3% for RA, LA, and PV classes, respectively. Our results show that a machine learning approach can potentially identify the location of 1R sustaining AF using the 12-lead ECG.
Atrial flutter (AFl) is a common heart rhythm disor- der driven by different self-sustaining electrophysiological atrial mechanisms. In the present work, we sought to dis- criminate which mechanism is sustaining the arrhythmia in an individual patient using non-invasive 12-lead elec- trocardiogram (ECG) signals. Specifically, we analyse the influence of atrial and torso geometries for the success of such discrimination. 2,512 ECG were simulated and 151 features were extracted from the signals. Three clas- sification scenarios were investigated: random set clas- sification; leave-one-atrium-out (LOAO); and leave-one- torso-out (LOTO). A radial basis neural network classifier achieved test accuracies of 89.84%, 88.98%, and 59.82% for the random set classification, LOTO, and LOAO, re- spectively. The most discriminative single feature was the F-wave duration (74% test accuracy). Our results show that a machine learning approach can potentially identify a high number of different AFl mechanisms using the 12- lead ECG. More than the 8 atrial models used in this work should be included during training due to the significant influence that the atrial geometry has on the ECG signals and thus on the resulting classification. This non-invasive classification can help to identify the optimal ablation strategy, reducing time and resources required to conduct invasive cardiac mapping and ablation procedures.
G. Luongo, S. Schuler, O. Dössel, and A. Loewe. 12-Lead ECG Feature Identification to Discriminate Different Types of Atrial Flutter. In 41 Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), 2019
This work uses a highly detailed computational model of human atria to investigate the effect of spatial gradient and smoothing of atrial wall thickness on inducibility and maintenance of atrial fibrillation (AF) episodes. An atrial model with homogeneous thickness (HO) was used as baseline for the generation of different atrial models including either a low (LG) or high thickness gradient between left/right atrial free wall and the other regions. Since the model with high spatial gradient presented non-natural sharp edges between regions, either 1 (HG1) or 2 (HG2) Laplacian smoothing iterations were applied. Arrhythmic episodes were initiated using a rapid pacing protocol and long-living rotors were detected and tracked over time. Thresholds optimised with receiver operating characteristic analysis were used to define high gradient/curvature regions. Greater spatial gradients increased the atrial model inducibility and unveiled additional regions vulnerable to maintain AF drivers. In the models with heterogeneous wall thickness (LG, HG2 and HG1), 73.5 ± 8.7% of the long living rotors were found in areas within 1.5 mm from nodes with high thickness gradient, and 85.0 ± 3.4% in areas around high endocardial curvature. These findings promote wall thickness gradient and endocardial curvature as measures of AF vulnerability.
Changes in atrial fibrillation cycle length (AF-CL) are broadly used as a ‘ground truth’ to assess the effect of substrate modification during AF ablation. This work sought to optimize thresholds for changes in coronary sinus CL (CS-CL) after local ablation using different atrial electrogram (AEG)-derived markers. 834 AEGs were collected from 11 patients undergoing persAF ablation. CS-CL was measured before and after each ablation point. Five AEG-derived markers were tested as classifiers for CS-CL changes: ICL (Biosense Webster), CFE-Mean (St. Jude Medical), Wave Similarity, Shannon Entropy and AEG-CL. The area under the receiver operating characteristic (AUROC) curve was used to assess the quality of classification for each marker. Maximum AUROC was found at threshold values between 9 and 14 ms in all markers, except for Shannon Entropy. The average AUROC of the five markers reached a maximum of 0.60 at a threshold value of 10 ms. The 10 ms threshold is suggested as a starting setpoint for future studies seeking to identify AF ablation targets based on an objective 'ground truth'.
G. Luongo. Atrial Arrhythmia Diagnosis Using the 12-Lead ECG: Machine learning leveraging in silico and clinical signals. Karlsruher Institut für Technologie (KIT). Dissertation. 2021
The electrocardiogram (ECG) in general, and the 12-lead ECG in particular, is one of the most common and widely available digital device that can be found in clinical facilities to measure the electrical activity of the heart. Therefore, it is considered the gold standard tool for this purpose. It is an inexpensive and non-invasive monitoring device that allows for rapid diagnosis of cardiovascular diseases (CVD). Among the most common CVDs there are atrial fibrillation (AFib or AF) and atrial flutter (AFlut or AFl). These two arrhythmias play a central role in the world’s healthcare systems, being among the main reasons for hospitalization, and responsible for very high costs in all countries. Moreover, even if they are not a direct cause of death, they can lead to multiple complications up to heart failure. For the reasons mentioned above, AFib, and AFlut are the focus of this thesis. The content of this thesis is divided into two projects. The overall goal is to develop methods with the help of biosignal processing, electrophysiological simulations, and machine learning to characterize the arrhythmia, support diagnosis, and predict complications or therapy outcomes. In the first project, in silico 12-lead ECGs produced from simulations at multiscale level are used to develop two signal processing algorithms for several AFlut mechanisms char- acterization: individual component, and spatial reduced recurrence quantification analysis (icRQA, and srRQA, respectively). Moreover, an analysis of the influence that the atrial and torso models have on the cardiac simulation results, thus on the resulting ECGs, is described. The findings from these two previous analyses are incorporated into the final study of the project: hybrid (in silico plus clinical data) and feature-based machine learning discrimination of three main AFlut categories (cavotricuspid isthmus-dependent, peri-mitral, and other left atrium AFlut classes). The two RQA algorithms allowed us to extract relevant features for AFlut differentiation. Analysis of models’ influence suggested that many atrial geometries should be used in the computational framework to avoid overfitting and thus leading to the incapacity of such in silico data in the clinical practice use. The final hybrid classifier demonstrated how an automatic and non-invasive discrimination of different AFlut mechanisms is possible using appropriate features, computational simulations, and taking into account the findings of the previous studies. The second project aims at estimating the location of AFib drivers with the surface ECG. Rotors and focal sources are simulated and considered as AFib drivers. A machine learning approach only trained on in silico 12-lead ECGs is implemented to discriminate between AFib drivers located near the pulmonary veins (PVs) vs. extra-PVs atrial areas. Moreover, the success of acute AFib termination by ablation procedure is studied and linked to the clinical relevance that such classifier may have in clinical practice. The last study of this second project aims at the prediction of one of the AFib complications (i.e., heart failure) using clinical single-lead ECG signals. Machine learning enabled the identification of AFib drivers located near PVs, also suggesting that PV isolation (PVI) is the most suitable therapy to terminate the arrhythmia in such cases. On the contrary, when proceeding with PVI for AFib drivers located outside the PV areas, the arrhythmia did not terminate. In these cases, physicians should plan further ablation procedures. Moreover, the use of a classifier trained only on simulated data and demonstrating to be effective on clinical test data may open the door to the use of in silico data for machine learning. To conclude, the successful prediction of AFib-induced heart failure has proven the existence of a link between some AFib cases and this serious complication, thus providing physicians with a tool to recognize when urgent action is needed to reduce patient safety risks. In all studies in which in silico ECGs are used to develop and tune the machine learning algorithms, tests on clinical data are performed to demonstrate the real applicability of these methods in healthcare. Advantages compared to existing approaches are discussed and all the studies have been published, or are under review, in peer-reviewed journals or in a conference proceeding. The results from the two projects demonstrate how simulated data can be used to develop, and improve adequate ECG signal processing methods, and how diagnosis, and therapy planning can be supported. Furthermore, the potential of the combination of simulations and machine learning for overcoming the problem of clinical data not available in large scale is demonstrated. The proposed methods can be used to support ablation procedure planning, arrhythmia diagnosis, complication prediction, and invasive procedure time reduction, and it is therefore likely to improve the outcome of the patients.