Background
Left ventricular systolic dysfunction (LVSD) is a precursor to heart failure arising from diverse cardiac conditions. Although echocardiography remains the reference standard for LVSD diagnosis, its routine use in the emergency department (ED) may be constrained by cost, time, equipment availability, and the need for specialized expertise. We evaluated the diagnostic performance of an artificial intelligence-based electrocardiogram analysis model (AI-ECG model) for detecting LVSD in patients presenting to the ED.

Methods
This retrospective observational study included patients treated at a single tertiary hospital between 2020 and 2022 who underwent 12-lead electrocardiography within 24 h of ED admission and echocardiography within 30 days. Electrocardiographic data were analyzed using AiTiALVSD version 1.00.00, with a predefined cutoff score of 9.7 used to classify patients as being at high or low risk of LVSD. Diagnostic performance was assessed using standard discrimination and classification metrics.

Results
Among 4529 included patients, 531 had LVSD. The AI-ECG model demonstrated high discrimination, with an area under the receiver operating characteristic curve (AUROC) of 0.934 (95% confidence interval [CI]: 0.923–0.945). Performance remained robust in patients with a shock index ≥ 0.9 (n = 453; AUROC, 0.895; 95% CI: 0.854–0.937) and in those with hypotension (n = 75; AUROC, 0.885; 95% CI: 0.786–0.984).

Conclusions
The AI-ECG model accurately identified LVSD in ED patients in this cohort despite heterogeneous acquisition conditions and retained good discrimination in hemodynamically unstable subgroups, although findings in the smaller hypotensive subgroup should be interpreted as exploratory.

Background
Artificial intelligence–enabled electrocardiogram (AI-ECG) detects left ventricular systolic and diastolic dysfunction at single time points, but longitudinal behavior during heart failure treatment remains unexplored. We examined whether paired serial AI-ECG scores track echocardiographic and biochemical trajectories.

Methods
In this retrospective 4-patient case series, previously validated deep learning AI-ECG models generated left ventricular systolic dysfunction and left ventricular diastolic dysfunction scores from serial 12-lead ECGs, paired with ejection fraction, E/e′, and N-terminal pro–B-type natriuretic peptide across treatment courses spanning revascularization, valve replacement with resynchronization therapy, percutaneous coronary intervention, and rhythm control.

Results
Serial score changes were broadly directionally concordant with echocardiographic and biochemical changes. Case-level and within-case discordances occurred, including persistent left ventricular systolic dysfunction elevation after resynchronization and bidirectional movement coinciding with rhythm transitions—consistent with rhythm- and pacing-related effects on ECG morphology.

Conclusions
These hypothesis-generating observations support the feasibility of longitudinal paired AI-ECG analysis and highlight the need for rhythm- and confounder-aware design in future studies.

Background
Structural heart disease (SHD) drives heart failure and cardiovascular mortality but remains underdiagnosed, and echocardiography is limited as a population-level screening tool.

Objectives
We evaluated whether a composite artificial intelligence–enabled electrocardiogram (AI-ECG), combining independently developed models for left ventricular systolic (LVSD) and diastolic dysfunction (LVDD), identifies prevalent and predicts incident SHD across diverse populations.

Methods
In this multinational cohort study, detection was assessed cross-sectionally in a Korean clinical cohort (Incheon Sejong Hospital) and a US dataset (Columbia University Irving Medical Center), and incident risk was assessed in the Korean cohort and the UK Biobank among individuals without baseline SHD or heart failure. Adults with paired ECG and echocardiography were analyzed for detection, with the composite defined as positive on either model. SHD comprised reduced left ventricular ejection fraction, moderate or severe valvular disease, left ventricular hypertrophy, or pulmonary hypertension. Detection was assessed by sensitivity and specificity, and incident risk by Cox models and the C statistic.

Results
Among 46,082 and 36,286 participants in the two detection cohorts, the composite detected SHD with sensitivity of 71.8% and 76.1% and specificity of 88.3% and 70.1%, with positivity across all phenotypes. Among at-risk individuals, composite positivity was associated with incident SHD (hazard ratios, 3.75 and 2.75), with C statistics of 0.69 to 0.78.

Conclusions
A composite AI-ECG identified prevalent and predicted incident SHD across multinational cohorts, capturing signals beyond its training targets and supporting its potential as a scalable cardiovascular screening tool; whether ECG-based risk stratification improves outcomes requires prospective evaluation.

Abstract. Wearable and portable devices enable convenient electrocar diography (ECG) acquisition, yet they typically measure only a single limb lead, limiting diagnostic utility compared to the standard 12-lead ECG used in hospitals. We study the problem of generating the full standard 12-lead ECG from a single measured lead by leveraging inherent ECG principles. Specifically, we emphasize two claims: (i) limb lead reconstruction benefits from a hybrid strategy that generates only one additional limb lead via learning and computes the remaining limb leads through vector operations; and (ii) precordial lead reconstruction is more reliable when conditioned on the full set of limb leads. Based on these claims, we propose AURORA, a 12-lead ECG reconstruction framework that combines learning models and deterministic vector operations to generate limb leads, followed by the reconstruction of precordial leads. To instantiate the learning components, we propose IGUANA, a lead-pattern learning model comprising a lead representation learner and a lead generator for precise lead-to-lead transformation. Experiments on two benchmark datasets demonstrate that AURORA consistently outperforms state-of-the-art methods and validate the effectiveness of incorporating vector operations in lead reconstruction. Our code is available at https://github.com/DHSeo11/AURORA.git.

Background
Women with suspected acute myocardial infarction (AMI) are at increased risk of delayed or missed diagnosis. Artificial intelligence–enabled electrocardiogram (AI-ECG) may support earlier AMI detection, but prospective evidence for sex-consistent performance and rule-out safety is limited.

Objectives
The purpose of this study was to evaluate sex-stratified diagnostic performance, rule-out safety, and phenotype robustness of an AI-ECG for AMI in a prospective, multicenter emergency department cohort.

Methods
ROMIAE (Rule-Out Acute Myocardial Infarction Using Artificial Intelligence Electrocardiogram Analysis) was a prospective external validation study at 18 centers in South Korea. Adults with suspected AMI were enrolled. AI-ECG analyzed initial ECGs using prespecified risk thresholds. Diagnostic performance was assessed using the area under the receiver-operating characteristic curve. Rule-out safety was evaluated by sensitivity, negative predictive value, and missed AMI rate at the low-risk cutoff. All analyses were sex-stratified, with prespecified subgroup analyses.

Results
Among 8,493 patients, 3,186 were women. AI-ECG demonstrated comparable discrimination in women and men (area under the receiver-operating characteristic curve 0.875 [95% CI: 0.853-0.896] vs 0.871 [95% CI: 0.858-0.883]). At the prespecified low-risk cutoff, rule-out sensitivity was 98.8% (95% CI: 97.0-99.5) in women and 99.8% (95% CI: 99.4-100.0) in men, with high negative predictive value (99.2% [95% CI: 97.9-99.7] in women and 99.1% [95% CI: 96.7-99.7] in men) and low missed AMI rates (<1% in both sexes). Performance remained stable across ST-segment elevation and non–ST-segment elevation myocardial infarction, without degradation in women.

Conclusions
In a prospective, multicenter emergency department cohort, AI-ECG demonstrated sex-consistent diagnostic performance and preserved rule-out safety for AMI. Further validation and implementation studies are warranted. (ROMIAE [Rule-Out Acute Myocardial Infarction Using Artificial Intelligence Electrocardiogram Analysis] Trial; NCT05435391)

Background and Objectives
The diagnostic accuracy of an artificial intelligence (AI)-derived initial 12-lead electrocardiogram (ECG) analysis was evaluated for early carbon monoxide-induced cardiomyopathy (CO-CMP) risk detection. Materials and Methods: Retrospective medical data of carbon monoxide poisoning (COP) cases between 1 January 2015 and 31 December 2024 were screened for the primary outcome: odds ratio (OR) for echocardiographically confirmed CO-CMP among those with high-risk probability score per the AI-derived model. Secondary outcomes included left ventricular ejection fraction (LVEF) and AI-derived probability score, critical care requirements, including intubation and intensive care unit (ICU) admission, and cardiac arrest events. Results: A total of 51 patients with acute COP were included in the final analysis, with 13 (25.5%) being diagnosed with CO-CMP. The LVEF in the CO-CMP group was lower than that in the non-CO-CMP group (40.00 ± 13.80% vs. 63.76 ± 6.24%, p < 0.001). The AI-derived probability score was higher in the CO-CMP group (11.3 [3.8–32.7] vs. 0.5 [0.2–2.2], p < 0.001). Among cardiac biomarkers, troponin I (2.37 [0.32–7.88] vs. 0.06 [0.06–0.95] ng/mL, p = 0.002) was higher in the CO-CMP group. Patients with CO-CMP required recurrent ventilator support (76.9% vs. 21.1%, p < 0.001) and ICU admission (92.3% vs. 42.1%, p = 0.003). In multivariable regression analysis, the AI-derived prediction model was independently associated with CO-CMP (OR 1.14; 95% confidence interval (CI) 1.02–1.27; p = 0.017; Firth-penalized OR 1.11; 95% CI 1.03–1.25; p < 0.001). Receiver operating characteristic analysis of the AI-derived model showed an area under the curve of 0.85 (95% CI 0.70–0.96) for the AI score alone and 0.92 (95% CI 0.83–0.99) for the Combined AI–cardiac marker model, with a sensitivity of 92.3% and specificity of 81.6%. Pairwise DeLong comparisons between the Combined AI model and comparator models did not reach statistical significance (Combined vs. AI-only, p = 0.092; Combined vs. cardiac markers, p = 0.052); however, the likelihood-ratio test for adding the AI probability score to the cardiac marker-only model demonstrated significant incremental information (χ2 = 13.68, p < 0.001). Conclusions: AI-based ECG analysis showed exploratory diagnostic association with LV systolic dysfunction observed in suspected CO-CMP patients. Given the limited sample size, low events-per-variable ratio, and lack of external validation, these findings suggest that AI-ECG analysis may provide incremental information for early cardiac risk stratification in selected patients.

Personalizing biosignal models is challenging due to temporal drift, where physiological signals evolve gradually or fluctuate abruptly over time. Recent studies have applied Online Test-Time Adaptation (OTTA), which continuously updates model parameters with streaming inputs, and demonstrated its promise for real-world deployment. This naturally raises the question of whether meta-learning, a related paradigm originally developed for rapid task adaptation, can also serve as an effective strategy for biosignal personalization. To address this, we systematically compare OTTA and meta-learning under identical streaming conditions for blood pressure prediction. Our analysis shows that OTTA achieves strong personalization by rapidly following distributional changes, while conventional meta-learning exhibits conservative adaptation in streaming regression tasks—avoiding large deviations in stable regimes but failing to capture gradual long-term trends. These findings suggest that meta-learning should not be treated as a direct alternative to OTTA, but rather redesigned to support continuous, long-term personalization in biosignals better.

Question
How well does an artificial intelligence electrocardiogram (AI-ECG) algorithm detect left ventricular systolic dysfunction (LVSD)?

Findings
In this cross-sectional study of 1444 participants with paired ECG and echocardiography across 8 health care facilities in Kenya, a high burden of LVSD was identified with high sensitivity and negative predictive value of the AI-ECG algorithm.

Meaning
The findings suggest that AI-ECG screening may be an effective strategy for identifying LVSD in resource-limited settings.

Aims
Vasovagal syncope (VVS) can cause injury and impaired quality of life, and effective prevention requires timely warning before loss of consciousness. To evaluate whether smartwatch photoplethysmography (PPG)-derived heart rate variability (HRV) can predict VVS before symptom onset, and to identify an optimal observation window and lead time.

Methods and results
We prospectively enrolled 132 patients with suspected neurally mediated syncope who underwent head-up tilt (HUT) testing while wearing a wrist-worn Samsung Galaxy Watch 6 for continuous multiwavelength PPG acquisition (25 Hz). The HRV features (n = 107) were extracted. An Extra Trees classifier (600 trees) was trained using an 80/20 subject-level split and evaluated on a hold-out test set. Model performance was assessed using AUROC and threshold metrics, including specificity, at a fixed sensitivity of 0.90. Sixty-three participants were HUT-positive, and 69 were HUT-negative. The 5-min presyncope window achieved the highest discrimination (AUROC, 0.91; 95% CI 0.77–1.00). At 90% sensitivity, specificity was 0.64 (95% CI 0.40–0.85). Using a fixed 5-min window, early prediction remained robust at a 5-min lead time (AUROC 0.91; 95% CI 0.76–1.00; accuracy 84.6%; 95% CI 0.65–0.92). The most informative predictors included nonlinear complexity metrics (approximate entropy and composite multiscale entropy) and autonomic balance indices (normalized low-frequency, log-transformed high-frequency, and the cardiac vagal index).

Conclusion
Artificial intelligence-enabled analysis of smartwatch PPG–derived HRV can prospectively predict VVS during HUT using a short 5-min observation window while maintaining clinically meaningful performance at a 5-min lead time, supporting the feasibility of wearable, real-time warning systems.

Background
Left ventricular filling pressure is associated with heart failure symptoms and a key prognostic marker and therapeutic target, but a scalable, accessible, and affordable tool for its noninvasive, serial estimation remains lacking. We developed an artificial intelligence (AI) model using a standard 12‐lead ECG to detect increased E/e', a general surrogate of elevated left ventricular filling pressure on echocardiography.

Methods
The AI model was built upon a foundation model trained with >1 million multiethnic ECGs and fine‐tuned through a development cohort of 225 737 ECGs and 115 982 echocardiogram records from 92 775 unique patients across 2 tertiary hospitals. The model performance was assessed in a separate internal test set (n=9278) and an independent external cohort from another tertiary hospital (n=17 926). Prognostic significance of the AI‐ECG output was evaluated in these hospital cohorts, as well as the UK Biobank (n=43 347). Finally, we validated the model output against invasively measured left ventricular filling pressure through cardiac catheterization (n=60).

ResultsThe AI‐ECG model detected increased E/e' with an area under the curve of 0.868 (95% CI, 0.859–0.877) and 0.850 (95% CI, 0.841–0.858) in the internal and external test cohorts, respectively. The AI‐ECG output value demonstrated a strong correlation with invasively measured left ventricular end‐diastolic pressure (Pearson's r=0.655) and was significantly associated with incident heart failure and mortality.

Conclusions
The AI‐ECG may enable identification of patients with increased left ventricular filling pressure and provide powerful prognostic information. Further prospective studies are warranted to evaluate its clinical utility.