China-calibration-ready echocardiographic feature modeling for LVEF-derived dysfunction classification: a surrogate-label proof-of-concept

A Nomogram Integrating Clinical and Cardiac Imaging for Predicting Short-Term Left Ventricular Ejection Fraction Decline in Patients with Duchenne Muscular Dystrophy

Purpose This study aimed to develop and validate a nomogram for predicting short-term LVEF decline in patients with DMD. Methods This was a single-center retrospective cohort study enrolling male patients diagnosed with DMD at Sun Yat-sen Memorial Hospital, Sun Yat-sen University, between 2015 and 2025. Data collected included patient age, cardiac troponin I (cTnI) levels, history of steroid therapy, baseline echocardiographic LVEF, and CMR data (including LGE and native T1 values). The primary outcome was the decline in LVEF (ΔLVEF ≤ −10%) during follow-up within 12 months. Least absolute shrinkage and selection operator (LASSO) logistic regression analysis was employed to identify independent risk factors and construct a nomogram-based predictive model. Model performance was assessed by the area under the receiver operating characteristic (ROC) curve and internally validated using bootstrap resampling (1000 repetitions). Results A total of 102 patients were included, of whom 38 (37.3%) exhibited a decline in LVEF. Multivariable analysis identified older age (OR 1.16, 95% CI 1.04–1.30, p = 0.009), abnormal cTnI (cTnI ≥ 0.04 ng/mL) (OR 7.27, 95% CI 1.46–36.28, p = 0.016), longer steroid duration (OR 1.72, 95% CI 1.09–2.71, p = 0.020, likely reflecting disease severity), the presence of LGE (OR 5.45, 95% CI 1.27–23.43, p = 0.023), and higher native T1 values (OR 1.02, 95% CI 1.01–1.04, p = 0.001) as independent risk factors. A higher baseline LVEF was protective (OR 0.81, 95% CI 0.72–0.91, p<0.001). The predictive model demonstrated excellent discrimination, with an AUC of 0.943 (95% CI 0.901–0.985). Internal validation yielded an optimism-corrected C-index of 0.922. Conclusion This study successfully established a comprehensive prediction model incorporating clinical and imaging variables, which can accurately identify DMD patients at risk for short-term LVEF decline. The model demonstrated high discriminative ability (AUC 0.943) in this retrospective, single-center cohort; however, these results are preliminary and require external validation.

Machine learning-based identification of heart failure with reduced ejection fraction using routine laboratory indicators

Background Early recognition of heart failure with reduced ejection fraction (HFrEF) is clinically important because treatment pathways and prognosis differ across left ventricular ejection fraction phenotypes. When echocardiography is delayed, routinely available laboratory data may help prioritize patients for definitive imaging. Methods This single-center retrospective study included 1,480 hospitalized patients with chronic heart failure, comprising 377 with HFrEF and 1,103 with HFmrEF/HFpEF. Baseline laboratory indicators were obtained from the first blood draw within 24 h of admission, and matched echocardiography was typically performed within 48 h. After a stratified 7:3 train-test split, preprocessing, imputation, standardization, feature selection, and model training were conducted within a leakage-free machine-learning pipeline. Cross-validated recursive feature elimination retained 13 routine indicators: proBNP, HCT, TBil, DBil, BUN, β2-MG, CRP, UA, Glb, GGT, MCH, LDL-C, and FIB. Six algorithms were evaluated using discrimination, calibration, decision-curve analysis, threshold analysis, and SHAP interpretation. Results The HFrEF group showed higher proBNP, BUN, TBil, DBil, GGT, HCT, and UA levels than the HFmrEF/HFpEF group. In the independent test set, random forest and XGBoost achieved the highest AUCs (both 0.789), whereas logistic regression showed comparable discrimination (AUC: 0.784) and strong calibration (intercept 0.017; slope 1.069). Random forest had the lowest Brier score (0.152), but pairwise bootstrap comparisons showed no statistically significant AUC differences among models. At the default threshold, sensitivity was limited; however, moving the random forest threshold to 0.15 increased sensitivity to 0.912 and negative predictive value to 0.935, supporting a rule-out triage strategy for prioritized echocardiography. SHAP analysis identified proBNP as the dominant predictor, followed by HCT, DBil, Glb, UA, GGT, TBil, and LDL-C. Conclusions A 13-indicator routine laboratory model provided moderate discrimination for distinguishing HFrEF from HFmrEF/HFpEF. Random forest had the numerically most favorable balance of discrimination and calibration, but its clinical usefulness depends on threshold optimization rather than default classification. The model should be regarded as an adjunctive triage tool pending external and prospective validation, not as a substitute for echocardiographic phenotyping.

Beyond Doppler: Scalable AI Detection of LVOT Obstruction in HCM.

BACKGROUND Accurate assessment of left ventricular outflow tract (LVOT) gradients is critical for hypertrophic cardiomyopathy management, yet Doppler-based measurements are technically demanding and require expertise. The objective of this work was to develop a multi-view deep learning model capable of classifying LVOT obstruction (>20 mm Hg) using routine 2-dimensional echocardiographic windows without reliance on Doppler imaging. METHODS We trained and externally validated a cross-attention-based video-to-video fusion framework that integrated EchoPrime-derived video representations from 3 standard transthoracic echocardiographic views to classify LVOT gradients. RESULTS Training was performed on a derivation cohort (N=1833) from a tertiary care system in the United States, with model performance evaluated on an internally held-out test set (N=275) and a Korean external validation cohort (N=46). Single-view baselines showed limited discrimination (external area under the receiver operating curves, 0.47-0.70). Conversely, the domain-specific foundational model (EchoPrime) achieved superior single-view performance (area under the receiver operating curves, 0.75-0.80 internal; 0.79-0.83 external), highlighting the importance of echo-specific pretraining and temporal modeling. The proposed multi-view fusion further enhanced predictive performance, with the late fusion model reaching an area under the receiver operating curve of 0.84 on the external cohort with significant population-shift. CONCLUSIONS These results suggest LVOT physiology is encoded in routine 2-dimensional imaging and can be leveraged for clinically relevant gradient classification without Doppler input. The proposed artificial intelligence-guided strategy demonstrates substantial cost savings compared with the screen-all approach. By integrating complementary spatial-temporal information across multiple views, our approach generalizes robustly across populations and may enable real-time decision support, extend LVOT assessment to portable or resource-limited settings, and complement Doppler-based evaluation for longitudinal hypertrophic cardiomyopathy management.

Pre-implant echocardiography plus EHR for post-LVAD risk stratification for survival: Single-center exploratory study.

PURPOSE Durable LVAD therapy improves survival for advanced heart failure, yet adverse outcomes remain common. We evaluated whether combining pre-implant echocardiography with routinely available Electronic Health Record (EHR) data yields clinically useful post-LVAD risk predictions to improve patient selection and perioperative management. METHODS In this retrospective study (2015-2022), pre-implant apical four-chamber echocardiograms were processed via raw loops and U-Net segmentation. CNN embeddings were integrated with PCA-reduced EHR variables including demographics, laboratories, and hemodynamics. Survival models, including Cox proportional hazards and random survival forests, were trained on multimodal inputs. Performance was validated using stratified 5-fold cross-validation, targeting a primary endpoint of time to death or missed follow-up. Saliency mapping was utilized to ensure clinical interpretability of the model's features. RESULTS Multimodal models achieved higher discrimination than single-modality models, with segmented-echo inputs outperforming raw videos (mean C-index of 0.711). Saliency mapping identified clinically coherent predictors: right ventricular and septal geometry on imaging, alongside renal, hepatic, and nutritional status from the EHR. CONCLUSIONS Integrating pre-implant echocardiography with EHR data enhances risk stratification survival for LVAD candidates. This multimodal approach identifies high-risk phenotypes, specifically right-heart and systemic frailty, providing a framework for personalized clinical decision support and future multicenter validation.

External validation of machine learning models predicting global longitudinal strain from conventional echocardiography in patients with cancer.

Validation of a deep learning-based workflow for the interpretation of the echocardiogram in a cardiooncology population

Echocardiography is the cornerstone for risk stratification, diagnosis, and monitoring of cancer therapy–related cardiac dysfunction (CTRCD)(1). Artificial intelligence (AI)–guided echocardiography has shown high accuracy and reliability in diverse cardiac populations and may reduce variability while improving workflow efficiency(2, 3). However, this technology has not yet been validated in a dedicated cohort of patients with cancer. To evaluate the accuracy and reliability of AI-guided echocardiography in assessing left ventricular ejection fraction (LVEF) and additional parameters, compared with conventional echocardiography, in a cardio-oncology population. This study included patients identified retrospectively from a cardio-oncology registry. Studies, that had already been analysed manually by expert sonographers and reported using AGFA PACS system, were uploaded to the US2.ai platform for automated analysis. The primary outcome was the level of agreement (LoA) between AI-guided and standard echocardiography for LVEF. Secondary outcomes included LoA for additional echocardiographic parameters and LoA between AI- LVEF and 3D LVEF. The performance of the deep learning (DL) algorithm in identifying LVEF <50% was evaluated using the area under the receiver operating characteristic curve (ROC-AUC). Subgroup analyses were performed in predefined populations clinically relevant in cardio-oncology. A total of 282 patients were included. Mean age was 60 ± 16 years, and 61% were women. Breast cancer was the most frequent malignancy (30.5%), followed by haematological malignancies (16.7%) and gastrointestinal tumours (10.6%). Manual median 2D LVEF was 60% (IQR: 55-64) and AI-derived LVEF was 59.2% (IQR: 53-64) showing good agreement and correlation (bias: −0.138, SD: 5.38, 95% LoA: −10.7 to 10.4, ICC: 0.791, 95% CI: 0.742–0.831, Spearman ρ: 0.718,), Table 1. Comparison between 3D echocardiography LVEF and AI-derived 2D LVEF showed similar agreement with narrower limits (bias: −0.13, 95% LoA: −9.51 to 9.26). The DL algorithm accurately identified LVEF <50% (ROC-AUC: 0.918, 95% CI: 0.875–0.961), Figure 1. Subgroup analyses demonstrated consistent agreement in patients with breast cancer, body mass index >30, prior radiotherapy and pericardial effusion. In a large real-world cardio-oncology cohort, AI-guided echocardiography demonstrated strong agreement with conventional echocardiography for LVEF assessment and high accuracy for detecting clinically relevant LV dysfunction. Performance was consistent across key subgroups, supporting the feasibility, reliability, and potential clinical value of integrating DL-based analysis into routine cardio-oncology echocardiographic workflows.Agreement between manual and AI-echo  AUC-ROC curve for LVEF<50%