A Multimodal and Explainable Machine Learning Approach to Diagnosing Multi-Class Ejection Fraction from Electrocardiograms

2026-04-30 · Source: cs.LG updates on arXiv.org · Field: Health & Wellbeing — Medical Devices & Health Technology, Clinical Care & Medical Practice, Health & Medical Research · Depth: Expert, quick

Summary

A new multimodal machine learning framework has been developed to diagnose multi-class left ventricular ejection fraction (LVEF) directly from 12-lead electrocardiograms (ECGs) and electronic health record (EHR) data. This framework classifies LVEF into four clinical strata: normal (>50%), mildly reduced (40-50%), moderately reduced (30-40%), and severely reduced (<30%). Utilizing retrospective data from Hartford HealthCare, XGBoost models were trained on 36,784 ECG-echocardiogram pairs from 30,952 outpatients. The multimodal model achieved one-vs-rest AUROCs of 0.95 for severe, 0.92 for moderate, 0.82 for mild, and 0.91 for normal LVEF, significantly outperforming ECG-only and EHR-only baselines. The model also demonstrated robust temporal generalizability when evaluated on 19,966 ECGs from a subsequent period, and SHAP attributions were used to identify influential features for explainability.

Key takeaway

For cardiologists and primary care physicians in resource-constrained settings, this multimodal ECG-based LVEF stratification offers a practical screening and triage aid. You can use this approach to prioritize patients for confirmatory echocardiography, improving access to critical cardiac diagnostics and optimizing resource allocation.

Key insights

Multimodal machine learning can accurately stratify LVEF from ECG and EHR data, enhancing accessibility.

Principles

• Combine ECG timeseries with EHR for LVEF classification.
• XGBoost models can achieve high AUROC for LVEF stratification.

Method

The method involves training XGBoost models on engineered 12-lead ECG features and structured EHR variables to classify LVEF into four clinical strata, with SHAP for explainability.

In practice

• Use ECG-based LVEF stratification for screening.
• Prioritize confirmatory imaging in resource-limited settings.

Topics

• Left Ventricular Ejection Fraction
• Electrocardiogram Analysis
• Multimodal Machine Learning
• XGBoost Model
• SHAP Explanations

Best for: AI Scientist, Research Scientist, Data Scientist

Related on AIssential

• Echo2ECG: Enhancing ECG Representations with Cardiac Morphology from Multi-View Echos — Aligning global ECGs with multi-view Echos resolves representational mismatch, yielding powerful, lightweight cardiac morphology features.
• Beyond the Blood Draw: Explainable Machine Learning for Non-Invasive Dysglycemia Risk Screening — Explainable ML models can non-invasively screen for dysglycemia risk using readily available data, outperforming existing clinical scores.
• Can AI help predict which heart-failure patients will worsen within a year? — PULSE-HF is a deep-learning model forecasting heart failure worsening via ECGs, achieving high accuracy in LVEF decline prediction.
• End-to-end pipeline for automated heart failure diagnosis with clinical notes using SNOMED-CT — An end-to-end pipeline automates heart failure diagnosis from German clinical notes using zero-shot learning and SNOMED-CT.
• ECG-Lens: Benchmarking ML & DL Models on PTB-XL Dataset — Complex CNNs significantly outperform traditional ML for automated 12-lead ECG classification on raw data.
• Stress Estimation in Elderly Oncology Patients Using Visual Wearable Representations and Multi-Instance Learning — Wearable data and MIL can estimate perceived stress in oncology patients, offering continuous cardiotoxicity surveillance.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.LG updates on arXiv.org.