Cardiac electrophysiology is rapidly evolving with advancements and applications in artificial intelligence (AI), given the rich level of diverse and high-dimensional data relevant to electrophysiology. AI has been useful to leverage data spanning the spectrum of basic science to clinical medicine including genetic information, atomic and molecular physicochemical features, signal data from action potential recordings and electrocardiograms, computational modeling data, cardiac imaging, and the electronic medical record. In this chapter, we review the current state of the art of AI in electrophysiology, first with its applications in basic science and computational modeling, then within clinical medicine across select disease focuses (atrial fibrillation, ventricular arrhythmia, and cardiac resynchronization therapy). We then outline future directions for AI in electrophysiology. A substantial component of AI in electrophysiology—AI interpretation of the electrocardiogram (ECG)—is discussed in a separate dedicated chapter.

Background: We hypothesized that computerized morphological analysis of the left atrium (LA) and pulmonary veins (PVs) via fractal measurements of shape and texture features of the LA myocardial wall could predict atrial fibrillation (AF) recurrence after ablation. Methods: Preablation contrast computed tomography scans were collected for 203 patients who underwent AF ablation. The LA body, PVs, and myocardial wall were segmented using a semi-automated region growing method. Twenty-eight fractal-based shape and texture-based features were extracted from resulting segments. The top features most associated with postablation recurrence were identified using feature selection and subsequently evaluated with a Random Forest classifier. Feature selection and classifier construction were performed on a discovery cohort (D-1) of 137 patients; classifiers were subsequently validated on an independent set (D-2) of 66 patients. Dedicated classifiers to capture the fractal and morphological properties of LA body (C-LA), PVs (C-PV), and LA myocardial (C-LAM) tissue were constructed, as well as a model (C-All) capturing properties of all segmented compartments. Fractal-based models were also compared against a model employing machine estimation of LA volume. To assess the effect of clinical parameters, such as AF type and catheter technique, a clinical model (C-clin) was also compared against C-All. Results: Statistically significant differences were observed for fractal features of C-LA, C-LAM, and C-All in distinguishing AF recurrence (P<0.001) on D-1. Using the 5 top features, C-All had the best prediction performance (area under the receiver operating characteristic curve [AUROC], 0.81 [95% CI, 0.78-0.85]), followed by C-PV (AUROC, 0.78 [95% CI, 0.74-0.80]), and C-LA (AUROC, 0.70 [95% CI, 0.63-0.78]) on D-2. The clinical parameter model C-clin yielded an AUROC, 0.70 (95% CI, 0.65-0.77), while the atrial volume model yielded an AUROC, 0.59. Combining C-All and C-clin on D-2 improved the AUROC to 0.87 (95% CI, 0.82-0.93). Conclusions: Fractal measurements of the LA, PVs, and atrial myocardium on computed tomography scans were associated with likelihood of postablation AF recurrence.

Objective: To identify radiomic and clinical features associated with post-ablation recurrence of AF, given that cardiac morphologic changes are associated with persistent atrial fibrillation (AF), and initiating triggers of AF often arise from the pulmonary veins which are targeted in ablation. Methods: Subjects with pre-ablation contrast CT scans prior to first-time catheter ablation for AF between 2014–2016 were retrospectively identified. A training dataset (D1) was constructed from left atrial and pulmonary vein morphometric features extracted from equal numbers of consecutively included subjects with and without AF recurrence determined at 1 year. The top-performing combination of feature selection and classifier methods based on C-statistic was evaluated on a validation dataset (D2), composed of subjects retrospectively identified between 2005–2010. Clinical models ( $\text{M}_{\mathrm {C}}$ ) were similarly evaluated and compared to radiomic ( $\text{M}_{\mathrm {R}}$ ) and radiomic-clinical models ( $\text{M}_{\mathrm {RC}}$ ), each independently validated on D2. Results: Of 150 subjects in D1, 108 received radiofrequency ablation and 42 received cryoballoon. Radiomic features of recurrence included greater right carina angle, reduced anterior-posterior atrial diameter, greater atrial volume normalized to height, and steeper right inferior pulmonary vein angle. Clinical features predicting recurrence included older age, greater BMI, hypertension, and warfarin use; apixaban use was associated with reduced recurrence. AF recurrence was predicted with radio-frequency ablation models on D2 subjects with C-statistics of 0.68, 0.63, and 0.70 for radiomic, clinical, and combined feature models, though these were not prognostic in patients treated with cryoballoon. Conclusions: Pulmonary vein morphology associated with increased likelihood of AF recurrence within 1 year of catheter ablation was identified on cardiac CT. Significance: Radiomic and clinical features-based predictive models may assist in identifying atrial fibrillation ablation candidates with greatest likelihood of successful outcome.

Artificial intelligence (AI) and machine learning (ML) in medicine are currently areas of intense exploration, showing potential to automate human tasks and even perform tasks beyond human capabilities. Literacy and understanding of AI/ML methods are becoming increasingly important to researchers and clinicians. The first objective of this review is to provide the novice reader with literacy of AI/ML methods and provide a foundation for how one might conduct an ML study. We provide a technical overview of some of the most commonly used terms, techniques, and challenges in AI/ML studies, with reference to recent studies in cardiac electrophysiology to illustrate key points. The second objective of this review is to use examples from recent literature to discuss how AI and ML are changing clinical practice and research in cardiac electrophysiology, with emphasis on disease detection and diagnosis, prediction of patient outcomes, and novel characterization of disease. The final objective is to highlight important considerations and challenges for appropriate validation, adoption, and deployment of AI technologies into clinical practice.

BACKGROUND:Cardiac resynchronization therapy (CRT) improves heart failure outcomes but has significant nonresponse rates, highlighting limitations in ECG selection criteria: QRS duration (QRSd) ≥150 ms and subjective labeling of left bundle branch block (LBBB). We explored unsupervised machine learning of ECG waveforms to identify CRT subgroups that may differentiate outcomes beyond QRSd and LBBB. METHODS:We retrospectively analyzed 946 CRT patients with conduction delay. Principal component analysis (PCA) dimensionality reduction obtained a 2-dimensional representation of preCRT 12-lead QRS waveforms. k-means clustering of the 2-dimensional PCA representation of 12-lead QRS waveforms identified 2 patient subgroups (QRS PCA groups). Vectorcardiographic QRS area was also calculated. We examined following 2 primary outcomes: (1) composite end point of death, left ventricular assist device, or heart transplant, and (2) degree of echocardiographic left ventricular ejection fraction (LVEF) change after CRT. RESULTS:Compared with QRS PCA Group 2 (n=425), Group 1 (n=521) had lower risk for reaching the composite end point (HR, 0.44 [95% CI, 0.38-0.53]; P<0.001) and experienced greater mean LVEF improvement (11.1±11.7% versus 4.8±9.7%; P<0.001), even among patients with LBBB with QRSd ≥150 ms (HR, 0.42 [95% CI, 0.30-0.57]; P<0.001; mean LVEF change 12.5±11.8% versus 7.3±8.1%; P=0.001). QRS area also stratified outcomes but had significant differences from QRS PCA groups. A stratification scheme combining QRS area and QRS PCA group identified patients with LBBB with similar outcomes to non-LBBB patients (HR, 1.32 [95% CI, 0.93-1.62]; difference in mean LVEF change: 0.8% [95% CI, -2.1% to 3.7%]). The stratification scheme also identified patients with LBBB with QRSd <150 ms with comparable outcomes to patients with LBBB with QRSd ≥150 ms (HR, 0.93 [95% CI, 0.67-1.29]; difference in mean LVEF change: -0.2% [95% CI, -2.7% to 3.0%]). CONCLUSIONS:Unsupervised machine learning of ECG waveforms identified CRT subgroups with relevance beyond LBBB and QRSd. This method may assist in objective classification of bundle branch block morphology in CRT.

Background: Cardiac resynchronization therapy (CRT) has significant nonresponse rates. We assessed whether machine learning (ML) could predict CRT response beyond current guidelines. Methods: We analyzed CRT patients from Cleveland Clinic and Johns Hopkins. A training cohort was created from all Johns Hopkins patients and an equal number of randomly sampled Cleveland Clinic patients. All remaining patients comprised the testing cohort. Response was defined as ≥10% increase in left ventricular ejection fraction. ML models were developed to predict CRT response using different combinations of classification algorithms and clinical variable sets on the training cohort. The model with the highest area under the curve was evaluated on the testing cohort. Probability of response was used to predict survival free from a composite end point of death, heart transplant, or placement of left ventricular assist device. Predictions were compared with current guidelines. Results: Nine hundred twenty-five patients were included. On the training cohort (n=470: 235, Johns Hopkins; 235, Cleveland Clinic), the best ML model was a naive Bayes classifier including 9 variables (QRS morphology, QRS duration, New York Heart Association classification, left ventricular ejection fraction and end-diastolic diameter, sex, ischemic cardiomyopathy, atrial fibrillation, and epicardial left ventricular lead). On the testing cohort (n=455, Cleveland Clinic), ML demonstrated better response prediction than guidelines (area under the curve, 0.70 versus 0.65; P =0.012) and greater discrimination of event-free survival (concordance index, 0.61 versus 0.56; P <0.001). The fourth quartile of the ML model had the greatest risk of reaching the composite end point, whereas the first quartile had the least (hazard ratio, 0.34; P <0.001). Conclusions: ML with 9 variables incrementally improved prediction of echocardiographic CRT response and survival beyond guidelines. Performance was not improved by incorporating more variables. The model offers potential for improved shared decision-making in CRT (online calculator: http://riskcalc.org:3838/CRTResponseScore ). Significant remaining limitations confirm the need to identify better variables to predict CRT response.

We hypothesized that the patient-specific time-varying changes in the spatial and temporal variability in cardiac repolarization (quantified by spatial TT' angle), and in myocardial injury (measured by high sensitivity troponin I, hsTnI), are independently associated with each other. Spatial TT' angle on resting 12-lead ECG (transformed to vectorcardiogram) and hsTnI were measured simultaneously every 3 hours during a 12-hour observation period in a prospective cohort of emergency department patients (n=379; age 57.8±13.2y; 54% female, 64% black), diagnosed with acute coronary syndrome (ACS; n=28), acute decompensated heart failure (ADHF; n=35), or an acute non-cardiac condition (n=316). High (above median) HsTnI in ACS was characterized by significantly larger TT' angle (12±8 vs 5±2 degrees; P=0.01) 12 hours after admission, but not earlier. In adjusted multinomial logit model, spatial TT' angle was associated with ADHF (Relative Risk Ratio 6.24 (95%CI 1.32-29.57; P=0.021), but not ACS. After full adjustment for confounders in random-effect linear regression, a 10-fold increase in hsTnI in a specific study participant was associated with 1.05 (95%CI 0.19 - 1.92) degrees increase in spatial TT' angle. Longitudinal association of hsTnI and TT' angle was especially prominent in patients with acute non-cardiac conditions, but not in ACS or ADHF.

BackgroundA three‐dimensional electrocardiographic (ECG) metric, the sum absolute QRST integral (SAI QRST), predicts ventricular arrhythmias in heart failure (HF) patients with implantable cardioverter defibrillator and mechanical response to cardiac resynchronization therapy. We hypothesized that there is an association between patient‐specific changes in SAI QRST and myocardial injury as measured by high‐sensitivity troponin I (hsTnI).MethodsSum absolute integral QRST on resting 12‐lead ECG and hsTnI were measured simultaneously, every 3 hours, and during 12‐hour observation period in a prospective cohort of emergency department patients (n = 398; mean age 57.8 ± 13.2 years; 54% female, 64% black), diagnosed with acute coronary syndrome (ACS, n = 28), acutely decompensated HF (acute decompensated heart failure, n = 35), cardiac non‐ACS (n = 19), or noncardiac condition (n = 316). Random‐effects linear regression analysis assessed the association of SAI QRST and myocardial injury, with adjustment for demographics (age, sex, race), prevalent cardiovascular disease (myocardial infarction, history of revascularization, stroke, and HF), risk factors (diabetes, smoking, hypercholesterolemia, hypertension, and cocaine use), and left bundle branch block.ResultsWithin the entire cohort, SAI QRST decreased by 3 (95%CI −5 to −1) mV*ms every 3 hours. A 10‐fold increase in hsTnI was associated with a 7.7 (0.6–14.9) mV*ms increase in SAI QRST. In the subgroup of acutely decompensated HF patients (n = 35), a 10‐fold increase in hsTnI was associated with a 61.0 (5.9–116.1) mV*ms increase in SAI QRST.ConclusionPatient‐specific time‐varying changes in the surface ECG scalar measure of global electrical heterogeneity, as measured by SAI QRST, and in myocardial injury as measured by hsTnI, are independently and directly associated with each other, likely reflecting a common underlying mechanism.

Background: The goal of this study was to compare the time series predictors of beat-to-beat variability in repolarization in healthy individuals.Methods: Spatial QRS- and T-vector amplitudes, spatial QRS-T, RR' and TT' angles, RR' and QT intervals, and QRS- and T-loop roundness were measured on 453 consecutive sinus beats in 168 healthy subjects (mean age 39.8 +/- 15.6 years; 50% men; 93% white). Panel time-series regression models were adjusted by age, sex, and race. Appropriate time series of ECG metrics served as predictors and outcomes.Results: Increase in T-loop roundness index by 0.1 was associated with 1.1 degrees (95%CI 0.9-2.2; P = 0.048) increase in corresponding TT' angle. One unit increase in a respiration index was associated with 4 ms (95%CI 0.6-7.0; P = 0.021) increase in QT interval.Conclusions: Spatial TT' angle and beat-to-beat variability in T-loop roundness represent intrinsic measures of repolarization variability. QT interval variability characterizes the effect of respiration and heart rate variability. (C) 2016 Elsevier Inc. All rights reserved.

BACKGROUND:Age-related macular degeneration (AMD), left untreated, is the leading cause of vision loss in people older than 55. Severe central vision loss occurs in the advanced stage of the disease, characterized by either the in growth of choroidal neovascularization (CNV), termed the "wet" form, or by geographic atrophy (GA) of the retinal pigment epithelium (RPE) involving the center of the macula, termed the "dry" form. Tracking the change in GA area over time is important since it allows for the characterization of the effectiveness of GA treatments. Tracking GA evolution can be achieved by physicians performing manual delineation of GA area on retinal fundus images. However, manual GA delineation is time-consuming and subject to inter-and intra-observer variability.METHODS:We have developed a fully automated GA segmentation algorithm in color fundus images that uses a supervised machine learning approach employing a random forest classifier. This algorithm is developed and tested using a dataset of images from the NIH-sponsored Age Related Eye Disease Study (AREDS). GA segmentation output was compared against a manual delineation by a retina specialist.RESULTS:Using 143 color fundus images from 55 different patient eyes, our algorithm achieved PPV of 0.82±0.19, and NPV of 0:95±0.07.DISCUSSION:This is the first study, to our knowledge, applying machine learning methods to GA segmentation on color fundus images and using AREDS imagery for testing. These preliminary results show promising evidence that machine learning methods may have utility in automated characterization of GA from color fundus images.

Timely detection of myocardial injury is essential for appropriate management of patients in emergency department (ED) evaluated for acute myocardial infarction. A novel electrocardiogram (ECG) metric, the Cardiac Electrical Biomarker (CEB), uses eigenvalue modeling of the 12-lead ECG and quantifies dipolar vs. multipolar forces. The goal of this project was to study association between the CEB and high-sensitivity troponin I (HsTnI). We conducted a retrospective study of patients, evaluated in the ED for acute myocardial infarction [n = 411; 57.6 ± 13.2 years; 186 (45%) men; 266 (64%) African-Americans]. Resting 12-lead ECG and HsTnI were measured at presentation and at 3, 6, and 9 hours after the initial measurement. The CEB was measured by the VectraplexECG System (VectraCor, Totowa, NJ). Patient-specific longitudinal analysis was performed to study association between the CEB with HsTnI changes over time. The CEB indicated myocardial injury in 116 (28.2%) study participants. HsTnI was significantly elevated during ED observation period in patients with myocardial injury, diagnosed by the CEB [median (interquartile range), 10.3 (5.2-31.4) vs. 6.3 (3.5-16.5) ng/L; P = 0.002]. In a mixed-effects linear regression adjusted for age, race, and sex, increasing HsTnI was associated with the CEB elevation [β-coefficient, 0.071 (95% confidence interval, 0.008-0.134); P = 0.027]. In conclusion, in patients in ED evaluated for acute myocardial injury, increasing values of HsTnI were associated with increasing values of the CEB, suggesting that myocardial injury is the mechanism that underlines acute changes in the CEB.

Background: Reproducibility of spatial TT' angle on the 10-second ECG and its agreement with QT variability has not been previously studied.Methods: We analyzed 2 randomly selected 10-second segments within 3-minute resting orthogonal ECG in 172 healthy IDEAL study participants (age 38.1 +/- 15.2 years, 50% male, 94% white). Repolarization lability was measured by the QT variance (QTV), short-term QT variability (STY (QT)), and spatial TT' angle. Bland-Altman analysis was used to assess the agreement between different log-transformed metrics of repolarization lability, and to assess the reproducibility.Results: The heart rate showed a very high reproducibility (bias 0.14%, Lin's rho_c = 0.99). As expected, noise suppression by averaging improves reproducibility. Agreement between two 10-second LogQTV was poor (bias -0.04; 95% limits of agreement [-1.89; 1.81]), while LogSTV(QT) (0.04 [-1.01; 1.10]), and especially LogTT' angle (-0.009 [-0.84; 0.82]) was better.Conclusion: TT' angle is a satisfactory reproducible metric of repolarization lability on the 10-second ECG. (C) 2014 Elsevier Inc. All rights reserved

Timely detection of myocardial injury is essential for appropriate management of patients in emergency department (ED) evaluated for acute myocardial infarction. A novel electrocardiogram (ECG) metric, the Cardiac Electrical Biomarker (CEB), uses eigenvalue modeling of the 12-lead ECG and quantifies dipolar vs. multipolar forces. The goal of this project was to study association between the CEB and high-sensitivity troponin I (HsTnI). We conducted a retrospective study of patients, evaluated in the ED for acute myocardial infarction [n = 411; 57.6 ± 13.2 years; 186 (45%) men; 266 (64%) African-Americans]. Resting 12-lead ECG and HsTnI were measured at presentation and at 3, 6, and 9 hours after the initial measurement. The CEB was measured by the VectraplexECG System (VectraCor, Totowa, NJ). Patient-specific longitudinal analysis was performed to study association between the CEB with HsTnI changes over time. The CEB indicated myocardial injury in 116 (28.2%) study participants. HsTnI was significantly elevated during ED observation period in patients with myocardial injury, diagnosed by the CEB [median (interquartile range), 10.3 (5.2-31.4) vs. 6.3 (3.5-16.5) ng/L; P = 0.002]. In a mixed-effects linear regression adjusted for age, race, and sex, increasing HsTnI was associated with the CEB elevation [β-coefficient, 0.071 (95% confidence interval, 0.008-0.134); P = 0.027]. In conclusion, in patients in ED evaluated for acute myocardial injury, increasing values of HsTnI were associated with increasing values of the CEB, suggesting that myocardial injury is the mechanism that underlines acute changes in the CEB.

The kinematics of the transverse motion of a swimming fish are analyzed using a complex modal decomposition. Cinematographic images of a swimming whiting (Gadus merlangus) were obtained from the work of Sir James Gray (Journal of Experimental Biology, 1933). The position of the midline for each image was determined, and used to produce planar positions of virtual markers distributed along the midline of the fish. Transverse deflections of each virtual marker were used for the complex orthogonal decomposition of modes. This method was applied to a normal whiting and an amputated whiting, both of Gray’s paper. The fish motions were well represented by a single complex mode, which was used as a modal filter. The modal coordinate was also extracted. The mode and modal coordinate were used to estimate the frequency, wavelength, and wave speed. The amputated fish was compared to the non-amputated fish, and the different amount of traveling in the respective waveforms was quantified.