High-quality echocardiography is essential for accurate and reproducible assessment of cardiac functional indices, which are highly dependent on adequate image quality and proper probe alignment. An artificial intelligence (AI)-based approach may enable automated image quality assessment.
Aims
Our aim was to develop and internally validate an AI-based algorithm to predict image quality from selected echocardiographic frames in candidates for implantable cardioverter-defibrillator and cardiac resynchronization therapy with a defibrillator implantation.
Methods
In this retrospective cross-sectional study, 248 patients (297 echocardiographic examinations) were included. Demographic, electrocardiographic, echocardiographic, and clinical data were collected. Apical 2-, 3-, and 4-chamber views were extracted for image quality analysis, yielding a total of 909 echocardiograms. Image quality was assessed using end-diastolic frames. An internally validated scoring framework was applied, demonstrating high interclass correlation.
Results
Regression models provided more clinically relevant information than classification models. Visual transformer models achieved Pearson correlation coefficients similar to those of convolutional neural networks (up to 0.812 and 0.772, respectively; P = 0.31). Architectures trained on end-diastolic frames achieved comparable Pearson correlation coefficients to those trained on combined end-diastolic and end-systolic frames. Compared with human experts, the models showed significantly higher absolute percentage errors, with values of 11%-12% (median) vs. 7.9% (mean) for the total image quality score and 13%-14% (median) vs. 8.8% (mean) for the border quality score.
Conclusions
Regression models demonstrated the highest performance. An internally validated AI model can predict an echocardiographic image quality score in a small cohort of cardiac resynchronization therapy with a defibrillator/implantable cardioverter-defibrillator candidates. However, external and prospective validation will be required to establish its generalizability, reliability, and utility before clinical application.
Aortic valve calcium scoring by computed tomography (CT) is an established method for assessing aortic stenosis severity but is limited by radiation exposure and availability. Artificial intelligence (AI)-based calcium detection using transthoracic echocardiography has shown promise but depends on acoustic window quality. Transesophageal echocardiography (TEE), particularly 3D TEE, may overcome these limitations by providing improved visualization without radiation. The objective of this study is to evaluate the feasibility of AI-based quantification of aortic valve calcium using 3D TEE. In this prospective pilot study, 23 patients (median age, 76 years; 56.5% male) with moderate or severe aortic stenosis underwent 3D TEE and CT. Multiplanar reconstruction generated 1.5-mm diastolic short-axis slices. A computer vision–based model identified calcium-related speckles. An automated TEE calcium score was derived from the sum of calcium pixels across 11 frames per patient, which was compared with the CT Agatston score. The TEE calcium score showed a significant positive correlation with CT Agatston scores (r = 0.65, P < 0.001). Receiver operating characteristic analysis yielded an area under the curve of 0.87 (95% confidence interval, 0.69–1.00) for identifying severe calcification (cutoff, 68,813 pixels; sensitivity, 89.5%; specificity, 75.0%). AI-based calcium quantification using 3D TEE is feasible and correlates with CT-derived scores. This radiation-free approach may provide a promising alternative for assessing aortic valve calcification.
P. Fazendas, R. Bairros, L. Elvas et al.· Journal of Cardiovascular Im...· 0 citations
Left ventricular diastolic dysfunction (LVDD) is an early precursor to heart failure with preserved ejection fraction (HFpEF) and it is currently diagnosed using echocardiography, a resource-intensive and operator-dependent modality that limits large scale screening. The 12-lead electrocardiogram (ECG) is widely available but lacks sufficient diagnostic accuracy for LVDD. Artificial intelligence (AI)-enhanced ECG analysis has emerged as a potential scalable alternative, although its overall diagnostic performance remains uncertain.
To evaluate the diagnostic accuracy of AI-based algorithms for detecting LVDD in a systematic review and meta-analysis.
We systematically searched eight major databases through August 2025, complemented by forward and backward citation chasing. Studies reporting sensitivity and specificity of AI-ECG models, using echocardiography as the reference standard, were included. Pooled sensitivity, specificity, and area under the summary receiver operating characteristic curve (AUC) were estimated using a bivariate random-effects model.
Five studies including 105,554 participants were analyzed. AI-ECG demonstrated a pooled sensitivity of 0.82 (95% CI: 0.81–0.83) and specificity of 0.77 (95% CI: 0.70–0.82), with an AUC of 0.85 (95% CI: 0.81–0.87). Substantial heterogeneity was observed (I2 = 98.5% for sensitivity and 99.8% for specificity), although results were robust in sensitivity analyses. Predictive values were prevalence-dependent. Negative predictive value was 98.8% at 5% prevalence and 93.7% at 22.2%, but declined to 81.1% at 50% prevalence and 64.7% at 70%.
AI-enhanced electrocardiography demonstrated good diagnostic performance for detecting LVDD and may support future rule-out or risk-enrichment strategies in selected populations. However, current evidence remains insufficient to support routine clinical implementation.
Edjimbi Johann, Nisarg Shah, L. Donisi et al.· European Heart Journal - Dig...· 0 citations
INTRODUCTION
Mitral regurgitation (MR) is one of the most prevalent valvular heart diseases, and its diagnosis traditionally relies on Doppler echocardiography, which is subject to significant variability and technical challenges. We developed and externally validated MitralVision, a deep learning model for automated classification of clinically significant MR using single-view, B-mode echocardiographic loops.
METHODS
MitralVision, a deep neural network, was trained on 28,487 apical four-chamber (A4C) B-mode echocardiographic cine loops from 11,244 studies across 20 U.S. states. The model was designed to differentiate clinically significant (moderate/severe) from non-significant (none/trace/mild) MR using grayscale cine loops without Doppler input. External validation was performed on 629 studies from 26 independent clinical sites using the original clinical interpretation as the reference standard. A separate board-certified Level III echocardiographer independently re-graded all external validation studies to assess interobserver variability.
RESULTS
On external validation, MitralVision achieved an AUROC of 0.91, sensitivity 82.1%, specificity 84.3%, negative predictive value 91.6%, and positive predictive value of 69.3%, compared with the original clinical read. Interobserver agreement between the original clinical read and the additional expert reader was 75.2% for clinically significant MR, with exact agreement across five MR grades of 35.5%. When benchmarking the additional expert reader's interpretation, model AUROC was 0.89. The model demonstrated excellent calibration (Brier score 0.12; expected calibration error 0.03).
CONCLUSIONS
MitralVision reliably distinguishes clinically significant MR using single-view B-mode echocardiography without Doppler input for model inference and may support more standardized MR screening. This streamlined AI-based approach offers reproducible MR assessment and may be compatible with future workflow implementation in high-throughput or resource-limited settings.
R. Sandler, J. Sokol, S.G. Pawar et al.· Journal of the American Soci...· 0 citations
Background Early and accurate detection of coronary artery disease (CAD) remains a challenge in primary care, particularly in low- and middle-income countries where access to advanced diagnostic imaging is limited. A resting electrocardiogram (ECG) is widely available but has low sensitivity for detecting ischemia. Cardisiography (CSG), an artificial intelligence–enhanced vectorcardiography technique, offers a promising non-invasive alternative. Methods This single-center, prospective, double-blinded pilot study enrolled 104 patients aged 40 and above with suspected CAD referred for coronary computerized tomography angiography (CCTA). All participants underwent ECG, CSG, and CCTA as the reference standard. Diagnostic accuracy was assessed using the CAD-RADS classification. Results CCTA identified coronary lesions classified as CAD-RADS 1–3 in 29 patients (12, 11, and 6, respectively). CSG achieved an overall sensitivity of 96.5% (28/29), compared with 6.9% (2/29) for ECG. Sensitivity was 100% for CAD-RADS 1 and 3, and 91% for CAD-RADS 2. Diagnostic accuracy metrics are reported with 95% confidence intervals. Conclusion In this pilot referral cohort, CSG demonstrated higher sensitivity than resting ECG for detecting coronary plaque. These preliminary findings suggest CSG may serve as a triage tool before CCTA or cardiology referral. Larger, multicenter studies are needed to validate its role and determine clinical utility. Clinical Trial Registration https://conabios.gob.do/reglamento, identifier 034-2023.
Rafael A. Guillén-Marmolejos, R. Núñez-Musa, A. Núñez-Selles et al.· Frontiers in Cardiovascular...· 0 citations
Objective To explore the predictive value of combining conventional ECG parameters (P-wave dispersion, Sokolow-Lyon voltage) and echocardiographic parameters (left atrial volume index, LAVI) for major adverse cardiovascular events (MACE) in coronary artery disease (CAD) patients. Methods From January 2024 to February 2026, 306 angiographically confirmed CAD patients were enrolled. All underwent standard 12-lead ECG and transthoracic echocardiography at admission. Over a median follow-up of 15.2 months, 50 patients developed MACE and 256 remained event-free. Baseline ECG (P-wave dispersion, QTc, Sokolow-Lyon voltage) and echocardiographic parameters (LAVI, left ventricular ejection fraction, left ventricular mass index) were compared. Independent predictors were identified by multivariate Cox regression, and a combined model was built. Incremental value was assessed by C-index, net reclassification improvement (NRI), and integrated discrimination improvement (IDI). The study was approved by the institutional review board. Results The MACE group had higher P-wave dispersion, LAVI, and left ventricular mass index, and lower Sokolow-Lyon voltage and LVEF (all p < 0.05). After multivariate adjustment, increased P-wave dispersion and greater LAVI were independent predictors of MACE (both p < 0.05). The combination model (P-wave dispersion + LAVI) significantly outperformed the basic clinical model (age, sex, diabetes, prior MI, eGFR) in risk discrimination (C-index, NRI, IDI all p < 0.05). Sensitivity analyses confirmed robustness. Conclusion Routinely available ECG-derived P-wave dispersion and echocardiographic LAVI are independent, complementary predictors of MACE in CAD patients. Integrating these two parameters into a simple risk model significantly enhances risk discrimination and reclassification, providing a practical, cost-effective tool for individualized management.
Qin Wu, Gang Chen, Jian Chang· Frontiers in Medicine· 0 citations