Multimodal explainable artificial intelligence identifies patients with non-ischaemic cardiomyopathy at risk of lethal ventricular arrhythmias.

The efficacy of an implantable cardioverter-defibrillator (ICD) in patients with a non-ischaemic cardiomyopathy for primary prevention of sudden cardiac death is increasingly debated. We developed a multimodal deep learning model for arrhythmic risk prediction that integrated late gadolinium enhanced (LGE) cardiac magnetic resonance imaging (MRI), electrocardiography (ECG) and clinical data. Short-axis LGE-MRI scans and 12-lead ECGs were retrospectively collected from a cohort of 289 patients prior to ICD implantation, across two tertiary hospitals. A residual variational autoencoder was developed to extract physiological features from LGE-MRI and ECG, and used as inputs for a machine learning model (DEEP RISK) to predict malignant ventricular arrhythmia onset. In the validation cohort, the multimodal DEEP RISK model predicted malignant ventricular arrhythmias with an area under the receiver operating characteristic curve (AUROC) of 0.84 (95% confidence interval (CI) 0.71-0.96), a sensitivity of 0.98 (95% CI 0.75-1.00) and a specificity of 0.73 (95% CI 0.58-0.97). The models trained on individual modalities exhibited lower AUROC values compared to DEEP RISK [MRI branch: 0.80 (95% CI 0.65-0.94), ECG branch: 0.54 (95% CI 0.26-0.82), Clinical branch: 0.64 (95% CI 0.39-0.87)]. These results suggest that a multimodal model achieves high prognostic accuracy in predicting ventricular arrhythmias in a cohort of patients with non-ischaemic systolic heart failure, using data collected prior to ICD implantation.

Dynamic prediction of malignant ventricular arrhythmias using neural networks in patients with an implantable cardioverter-defibrillator.

BACKGROUND: Risk stratification for ventricular arrhythmias currently relies on static measurements that fail to adequately capture dynamic interactions between arrhythmic substrate and triggers over time. We trained and internally validated a dynamic machine learning (ML) model and neural network that extracted features from longitudinally collected electrocardiograms (ECG), and used these to predict the risk of malignant ventricular arrhythmias. METHODS: A multicentre study in patients implanted with an implantable cardioverter-defibrillator (ICD) between 2007 and 2021 in two academic hospitals was performed. Variational autoencoders (VAEs), which combine neural networks with variational inference principles, and can learn patterns and structure in data without explicit labelling, were trained to encode the mean ECG waveforms from the limb leads into 16 variables. Supervised dynamic ML models using these latent ECG representations and clinical baseline information were trained to predict malignant ventricular arrhythmias treated by the ICD. Model performance was evaluated on a hold-out set, using time-dependent receiver operating characteristic (ROC) and calibration curves. FINDINGS: 2942 patients (61.7 ± 13.9 years, 25.5% female) were included, with a total of 32,129 ECG recordings during a mean follow-up of 43.9 ± 35.9 months. The mean time-varying area under the ROC curve for the dynamic model was 0.738 ± 0.07, compared to 0.639 ± 0.03 for a static (i.e. baseline-only model). Feature analyses indicated dynamic changes in latent ECG representations, particularly those affecting the T-wave morphology, were of highest importance for model predictions. INTERPRETATION: Dynamic ML models and neural networks effectively leverage routinely collected longitudinal ECG recordings for personalised and updated predictions of malignant ventricular arrhythmias, outperforming static models. FUNDING: This publication is part of the project DEEP RISK ICD (with project number 452019308) of the research programme Rubicon which is (partly) financed by the Dutch Research Council (NWO). This research is partly funded by the Amsterdam Cardiovascular Sciences (personal grant F.V.Y.T).

Reversible Cause of Cardiac Arrest and Secondary Prevention Implantable Cardioverter Defibrillators in Patients With Coronary Artery Disease: Value of Complete Revascularization and LGE-CMR.

Background In survivors of sudden cardiac arrest with obstructive coronary artery disease, it remains challenging to distinguish ischemia as a reversible cause from irreversible scar-related ventricular arrhythmias. We aimed to evaluate the value of implantable cardioverter-defibrillator (ICD) implantation in sudden cardiac arrest survivors with presumably reversible ischemia and complete revascularization. Methods and Results This multicenter retrospective cohort study included 276 patients (80% men, age 67±10 years) receiving ICD implantation for secondary prevention. Angiography was performed before ICD implantation. A subgroup of 166 (60%) patients underwent cardiac magnetic resonance imaging with late gadolinium enhancement before implantation. Patients were divided in 2 groups, (1) ICD-per-guideline, including 228 patients with incomplete revascularization or left ventricular ejection fraction ≤35%, and (2) ICD-off-label, including 48 patients with complete revascularization and left ventricular ejection fraction >35%. The primary outcome was time to appropriate device therapy (ADT). During 4.0 years (interquartile range, 3.5-4.6) of follow-up, ADT developed in 15% of the ICD-off-label group versus 43% of the ICD-per-guideline group. Time to ADT was comparable in the ICD-off-label and ICD-per-guideline groups (hazard ratio (HR), 0.46; P=0.08). No difference in mortality was observed (HR, 0.95; P=0.93). Independent predictors of ADT included age (HR, 1.03; P=0.01), left ventricular end-diastolic volume HR, (1.05 per 10 mL increase; P<0.01) and extent of transmural late gadolinium enhancement (HR, 1.12; P=0.04). Conclusions This study demonstrates that sudden cardiac arrest survivors with coronary artery disease remain at high risk of recurrent ventricular arrhythmia, even after complete revascularization and with preserved left ventricular function. Late gadolinium enhancement-cardiac magnetic resonance imaging derived left ventricular volumes and extent of myocardial scar were independently associated with.

Differences between gap-related persistent conduction and carina-related persistent conduction during radiofrequency pulmonary vein isolation.

BACKGROUND: During pulmonary vein isolation (PVI), nonisolation after initial encircling of the pulmonary veins (PVs) may be due to gaps in the initial ablation line, or alternatively, earliest PV activation may occur on the intervenous carina and ablation within the wide-area circumferential ablation (WACA) circle is needed to eliminate residual conduction. This study investigated prognostic implications and predictors of gap-related persistent conduction (gap-RPC) and carina-related persistent conduction (carina-RPC) during PVI. METHODS AND RESULTS: Two hundred fourteen atrial fibrillation (AF) patients (57% paroxysmal, 61% male, mean age 62 ± 9 years) undergoing first contact force-guided radiofrequency PVI were studied. Preprocedural cardiac computed tomography imaging was used to assess left atrial and PV anatomy. PVI was assessed directly after initial WACA circle creation, after a minimum waiting period of 30 minutes, and after adenosine infusion. Persistent conduction was targeted for additional ablation and classified as gap-RPC or carina-RPC, depending on the earliest activation site. The 1-year AF recurrence rate was higher in patients with gap-RPC (47%) compared to patients without gap-RPC (28%; P = .003). No significant difference in 1-year recurrence rate was found between patients with carina-RPC (37%) and patients without carina-RPC (31%; P = .379). Multivariate analyses identified paroxysmal AF and WACA circumference as independent predictors of gap-RPC, whereas carina width and WACA circumference correlated with carina-RPC. CONCLUSIONS: Gap-RPC is associated with increased AF recurrence risk after PVI, whereas carina-RPC does not predict AF recurrence. Moreover, gap-RPC and carina-RPC have different correlates and may thus have different underlying mechanisms.

Correlation between septal midwall late gadolinium enhancement on CMR and conduction delay on ECG in patients with nonischemic dilated cardiomyopathy.

BACKGROUND: Septal midwall late gadolinium enhancement (LGE) on cardiac magnetic resonance imaging (CMR) is a characteristic finding in nonischemic dilated cardiomyopathy (DCM) and is associated with adverse cardiac events. QRS-prolongation in DCM is also frequently present and a predictor of arrhythmic events and mortality. Since the His-Purkinje fibres are located in the interventricular septum, QRS-prolongation may directly result from septal fibrosis, visualized by LGE. Our aim was to study the correlation of the presence and extent of septal midwall LGE and QRS-duration. METHODS: DCM-patients with left ventricular (LV) dysfunction (LVEF < 50%) were included. LV volumes, systolic function and nonischemic septal midwall LGE, defined as patchy or stripe-like LGE in the septal segments, were quantified. QRS-duration on standard 12-lead ECG was measured. RESULTS: 165 DCM-patients were included (62% male, mean age 59 ± 15 years) with a median LVEF of 36% [24-44]. Fifty-one patients (31%) demonstrated septal midwall LGE with a median extent of 8.1 gram [4.3-16.8]. Patients with midwall LGE had increased LV end-diastolic volumes (EDV) 248 mL [193-301] vs. 193 mL [160-239], p < 0.001) and lower LVEF (26% [18-35] vs. 40% [32-45], p < 0.001). Median QRS-duration was 110 ms [95-146] without a correlation to the presence nor extent of midwall LGE. QRS-duration was moderately correlated with LV-dilation and mass (respectively r = 0.35, p < 0.001 and r = 0.30, p < 0.001). CONCLUSION: In DCM-patients, QRS-prolongation and septal midwall LGE are frequently present and often co-exist. However, they are not correlated. This suggests that the assessment of LGE-CMR has complementary value to ECG evaluation in the clinical assessment and risk stratification of DCM-patients.

Left atrial sphericity as a marker of atrial remodeling: Comparison of atrial fibrillation patients and controls.

BACKGROUND: Left atrial (LA) sphericity has been proposed as a more sensitive marker of atrial fibrillation (AF)-associated atrial remodeling compared to traditional markers such as LA size. However, mechanisms that underlie changes in LA sphericity are not fully understood and studies investigating the predictive value of LA sphericity for AF ablation outcome have yielded conflicting results. The present study aimed to assess correlates of LA sphericity and to compare LA sphericity in subjects with and without AF. METHODS: Measures of LA size (LA diameter, LA volume, LA volume index), LA sphericity and thoracic anteroposterior diameter (APd) at the level of the LA were determined using computed tomography (CT) imaging data in 293 AF patients (62% paroxysmal AF) and 110 controls. RESULTS: LA diameter (40.1 ± 6.8 mm vs. 35.2 ± 5.1 mm; p < 0.001), LA volume (116.0 ± 33.0 ml vs. 80.3 ± 22.6 ml; p < 0.001) and LA volume index (56.1 ± 15.3 ml/m2 vs. 41.6 ± 11.1 ml/m2; p < 0.001) were significantly larger in AF patients compared to controls, also after adjustment for covariates. LA sphericity did not differ between AF patients and controls (83.7 ± 2.9 vs. 83.9 ± 2.4; p = 0.642). Multivariable linear regression analysis demonstrated that LA diameter, LA volume, female sex, body length and thoracic APd were independently associated with LA sphericity. CONCLUSIONS: The present study suggests that thoracic constraints rather than the presence of AF determine LA sphericity, implying LA sphericity to be unsuitable as a marker of AF-related atrial remodeling.

Long-Term Prognostic Implications of Previous Silent Myocardial Infarction in Patients Presenting With Acute Myocardial Infarction.

OBJECTIVES: This study investigated the prevalence of silent myocardial infarction (MI) in patients presenting with first acute myocardial infarction (AMI), and its relation with mortality and major adverse cardiovascular events (MACE) at long-term follow-up. BACKGROUND: Up to 54% of MI occurs without apparent symptoms. The prevalence and long-term prognostic implications of previous silent MI in patients presenting with seemingly first AMI are unclear. METHODS: A 2-center observational longitudinal study was performed in 392 patients presenting with first AMI between 2003 and 2013, who underwent late gadolinium enhancement cardiac magnetic resonance (LGE-CMR) examination within 14 days post-AMI. Silent MI was assessed on LGE-CMR images by identifying regions of hyperenhancement with an ischemic distribution pattern in other territories than the AMI. Mortality and MACE (all-cause death, reinfarction, coronary artery bypass grafting, and ischemic stroke) were assessed at 6.8 ± 2.9 years follow-up. RESULTS: Thirty-two patients (8.2%) showed silent MI on LGE-CMR. Compared with patients without silent MI, mortality risk was higher in patients with silent MI (hazard ratio: 3.87; 95% confidence interval: 1.21 to 12.38; p = 0.023), as was risk of MACE (hazard ratio: 3.10; 95% confidence interval: 1.22 to 7.86; p = 0.017), both independent from clinical and infarction-related characteristics. CONCLUSIONS: Silent MI occurred in 8.2% of patients presenting with first AMI and was independently related to poorer long-term clinical outcome, with a more than 3-fold risk of mortality and MACE. Silent MI holds prognostic value over important traditional prognosticators in the setting of AMI, indicating that these patients represent a high-risk subgroup warranting clinical awareness.

Scar tissue-guided left ventricular lead placement for cardiac resynchronization therapy in patients with ischemic cardiomyopathy: an acute pressure-volume loop study.

BACKGROUND: Response to cardiac resynchronization therapy (CRT) is hampered by the extent and location of left ventricular (LV) scar tissue. It is commonly advised to avoid scar tissue while placing the LV lead. However, whether individual patients benefit from this strategy remains unclear. METHODS: Thirty-two CRT candidates with ischemic cardiomyopathy were enrolled from 2 successive clinical trials (TBS and E-pot study). Magnetic resonance imaging with late contrast enhancement was performed to assess location, degree and transmurality of LV scar tissue. Patients underwent invasive pressure-volume loop measurements to assess acute LV pump function changes during pacing at posterolateral (PL) and anterolateral LV sites. RESULTS: In the study population (26 [81%] men, ejection fraction [EF] 22% ± 8%, QRS 149 ± 20 milliseconds), baseline mean stroke work (SW) and dP/dtmax were 4.4 ± 2.2 L∙mmHg and 849 ± 212 mmHg/s, respectively. The extent of scar tissue was inversely related to the acute increase in SW during pacing (R = -0.53, P = .002). Stimulating PL scar tissue resulted in deterioration of pump function (∆SW -17% ± 17%, P = .018), whereas pacing PL viable tissue led to an increase in pump function (∆SW +62% ± 51%, P < .001). Switching from pacing at the location of scar tissue, irrespective of the scar location, to viable tissue showed a significant increase in SW (-8% ± 20% vs +20 ± 40, P = .004). CONCLUSIONS: The extent of LV scar tissue is inversely related to acute pump function improvement during CRT. Pacing at the location of (transmural) scar tissue at any site of the LV will generally deteriorate LV pump function. Placing the LV lead over viable myocardium significantly improves pump function as compared with pacing at the location of scar tissue in patients with ischemic cardiomyopathy.

Coronary risk factors and myocardial blood flow in patients evaluated for coronary artery disease: a quantitative [15O]H2O PET/CT study.

BACKGROUND: There has been increasing interest in quantitative myocardial blood flow (MBF) imaging over the last years and it is expected to become a routinely used technique in clinical practice. Positron emission tomography (PET) using [(15)O]H(2)O is the established gold standard for quantification of MBF in vivo. A fundamental issue when performing quantitative MBF imaging is to define the limits of MBF in a clinically suitable population. The aims of the present study were to determine the limits of MBF and to determine the relationship among coronary artery disease (CAD) risk factors, gender and MBF in a predominantly symptomatic patient cohort without significant CAD. METHODS: A total of 128 patients (mean age 54 ± 10 years, 50 men) with a low to intermediate pretest likelihood of CAD were referred for noninvasive evaluation of CAD using a hybrid PET/computed tomography (PET/CT) scanner. MBF was quantified with [(15)O]H(2)O at rest and during adenosine-induced hyperaemia. Obstructive CAD was excluded in these patients by means of invasive or CT-based coronary angiography. RESULTS: Global average baseline MBF values were 0.91 ± 0.34 and 1.09 ± 0.30 ml·min(-1)·g(-1) (range 0.54-2.35 and 0.59-2.75 ml·min(-1)·g(-1)) in men and women, respectively (p < 0.01). However, no gender-dependent difference in baseline MBF was seen following correction for rate-pressure product (0.98 ± 0.45 and 1.09 ± 0.30 ml·min(-1)·g(-1) in men and women, respectively; p = 0.08). Global average hyperaemic MBF values were 3.44 ± 1.20 ml·min(-1)·g(-1) in the whole study population, and 2.90 ± 0.85 and 3.78 ± 1.27 ml·min(-1)·g(-1) (range 1.52-5.22 and 1.72-8.15 ml·min(-1)·g(-1)) in men and women, respectively (p < 0.001). Multivariate analysis identified male gender, age and body mass index as having an independently negative impact on hyperaemic MBF. CONCLUSION: Gender, age and body mass index substantially influence reference values and should be corrected for when interpreting hyperaemic MBF values.

Quantitative relationship between coronary artery calcium score and hyperemic myocardial blood flow as assessed by hybrid 15O-water PET/CT imaging in patients evaluated for coronary artery disease.

BACKGROUND: The incremental value of CAC over traditional risk factors to predict coronary vasodilator dysfunction and inherent myocardial blood flow (MBF) impairment is only scarcely documented (MBF). The aim of this study was therefore to evaluate the relationship between CAC content, hyperemic MBF, and coronary flow reserve (CFR) in patients undergoing hybrid (15)O-water PET/CT imaging. METHODS: We evaluated 173 (mean age 56 ± 10, 78 men) patients with a low to intermediate likelihood for coronary artery disease (CAD), without a documented history of CAD, undergoing vasodilator stress (15)O-water PET/CT and CAC scoring. Obstructive coronary artery disease was excluded by means of invasive (n = 44) or CT-based coronary angiography (n = 129). RESULTS: 91 of 173 patients (52%) had a CAC score of zero. Of those with CAC, the CAC score was 0.1-99.9, 100-399.9, and ≥400 in 31%, 12%, and 5% of patients, respectively. Global CAC score showed significant inverse correlation with hyperemic MBF (r = -0.32, P < .001). With increasing CAC score, there was a decline in hyperemic MBF on a per-patient basis [3.70, 3.30, 2.68, and 2.53 mL · min(-1) · g(-1), with total CAC score of 0, 0.1-99.9, 100-399.9, and ≥400, respectively (P < .001)]. CFR showed a stepwise decline with increasing levels of CAC (3.70, 3.32, 2.94, and 2.93, P < .05). Multivariate analysis, including age, BMI, and CAD risk factors, revealed that only age, male gender, BMI, and hypercholesterolemia were associated with reduced stress perfusion. Furthermore, only diabetes and age were independently associated with CFR. CONCLUSION: In patients without significant obstructive CAD, a greater CAC burden is associated with a decreased hyperemic MBF and CFR. However, this association disappeared after adjustment for traditional CAD risk factors. These results suggest that CAC does not add incremental value regarding hyperemic MBF and CFR over established CAD risk factors in patients without obstructive CAD.