Detecting low left ventricular ejection fraction from ECG using an interpretable and scalable predictor-driven framework

Dimensions overall score 7.0

• The ECGPD-LEF framework achieved an internal AUROC of 88.4% and an external AUROC of 86.8% for detecting moderate low left ventricular ejection fraction (LEF).result 95%
  Evidencepartial

  our framework achieved robust discrimination for moderate LEF (internal AUROC 88.4%, F1 64.5%; external AUROC 86.8%, F1 53.6%)

  Implicationpartial

  Explicitly stated in the abstract with specific performance metrics for both internal and external validation cohorts.

  Verificationpartial

  partial

• The ECGPD-LEF framework consistently outperformed the official end-to-end baseline provided with the EchoNext benchmark across demographic and clinical subgroups.result 90%
  Evidencepartial

  consistently outperforming the official end-to-end baseline provided with the benchmark across demographic and clinical subgroups.

  Implicationpartial

  Directly stated in the abstract as a key comparative result.

  Verificationpartial

  partial

• The framework's high-impact predictors for LEF risk include normal ECG, incomplete left bundle branch block, and subendocardial injury in anterolateral leads.result 90%
  Evidencepartial

  Interpretability analyses identified high-impact predictors, including normal ECG, incomplete left bundle branch block, and subendocardial injury in anterolateral leads, driving LEF risk estimation.

  Implicationpartial

  Explicitly listed in the abstract as findings from interpretability analyses.

  Verificationpartial

  partial

• The framework's predictors enabled zero-shot-like inference for LEF detection without task-specific retraining, achieving AUROCs of 71.6-81.0%.technical 90%
  Evidencepartial

  Notably, these predictors independently enabled zero-shot-like inference without task-specific retraining (internal AUROC 75.3-81.0%; external AUROC 71.6-78.6%)

  Implicationpartial

  Explicitly stated in the abstract with performance ranges for internal and external validation.

  Verificationpartial

  partial

• The ECGPD-LEF framework integrates foundation model-derived diagnostic probabilities with interpretable modeling, moving beyond end-to-end black-box models.method 85%
  Evidencepartial

  We introduced ECG-based Predictor-Driven LEF (ECGPD-LEF), a structured framework that integrates foundation model-derived diagnostic probabilities with interpretable modeling for detecting LEF from ECG.

  Implicationpartial

  Directly stated in the abstract as the core methodological innovation.

  Verificationpartial

  partial

• The framework was trained on the EchoNext dataset comprising 72,475 ECG-echocardiogram pairs.method 95%
  Evidencepartial

  Trained on the benchmark EchoNext dataset comprising 72,475 ECG-echocardiogram pairs

  Implicationpartial

  Explicitly stated in the abstract with the exact dataset size.

  Verificationpartial

  partial

• Existing AI-ECG approaches for LEF detection have limited interpretability or depend on commercial algorithms with suboptimal performance.limitation 80%
  Evidencepartial

  existing approaches rely solely on end-to-end black-box models with limited interpretability or on tabular systems dependent on commercial ECG measurement algorithms with suboptimal performance.

  Implicationpartial

  Directly stated in the abstract as a limitation of prior work that motivates the new framework.

  Verificationpartial

  partial

• The zero-shot inference capability indicates that ventricular dysfunction is intrinsically encoded within structured diagnostic probability representations from ECG.technical 85%
  Evidencepartial

  indicating that ventricular dysfunction is intrinsically encoded within structured diagnostic probability representations.

  Implicationpartial

  Directly stated in the abstract as an interpretation of the zero-shot result.

  Verificationpartial

  partial

Competitive landscape

A scalable and interpretable AI framework for early detection of low left ventricular ejection fraction from ECGs, outperforming existing black-box models.