Artificial Intelligence-Based Electrocardiography for Aortic Stenosis ― Moving Beyond Voltage Criteria Toward Real-World Screening ―

Kenichi Ishizu; Hirokuni Akahori; Yasuko K. Bando

doi:10.1253/circrep.cr-26-0294

November 2, 2026Circulation ReportsOpen Access

Artificial Intelligence-Based Electrocardiography for Aortic Stenosis ― Moving Beyond Voltage Criteria Toward Real-World Screening ―

View Full Paper

Ask AI

Bookmark

Share

Why the study?

Does a deep learning AI-ECG model improve the detection of aortic stenosis beyond conventional LVH voltage criteria in hospitalized patients?

Key Points

To examine how deep learning models analyze 12-lead electrocardiograms to screen for aortic stenosis and identify the biological signals detected beyond conventional voltage criteria.
Editorial appraisal of a deep learning model trained on 244,816 ECGs from 51,713 patients across 6 Japanese academic institutions within the CLIDAS database.
Evaluated model mechanics using gradient-weighted class activation mapping compared against Sokolow–Lyon and Cornell voltage criteria for left ventricular hypertrophy.
The evaluated deep learning model achieved an area under the receiver operating characteristic curve (AUC) of 0.849 for aortic stenosis detection.
The algorithm demonstrated a negative predictive value of 98.0% and a positive predictive value of 18.0%.
Gradient-weighted class activation mapping revealed model attention concentrated on limb lead QRS morphology rather than standard precordial voltage criteria, with performance modestly attenuated in bundle branch block.

Structured PICO

Does a deep learning AI-ECG model improve the detection of aortic stenosis beyond conventional LVH voltage criteria in hospitalized patients?

P

Population

Hospitalized patients across 6 Japanese academic institutions within the CLIDAS database (n=51,713 patients, 244,816 ECGs), with an aortic stenosis prevalence of 6.3%

I

Intervention

Deep learning artificial intelligence model applied to 12-lead electrocardiography (AI-ECG)

C

Comparator

Conventional left ventricular hypertrophy (LVH) voltage criteria (Sokolow–Lyon and Cornell voltage criteria)

O

Outcome

Detection of aortic stenosis (AS) defined by inpatient Diagnosis Procedure Combination (DPC) codessurrogate

Limitations

AS labels derived from inpatient diagnosis codes cannot distinguish mild from severe disease, and hospitalization-related factors may have influenced ECG morphology independent of valvular pathology.
The cohort was drawn exclusively from hospitalized patients at academic centers where AS prevalence (6.3%) far exceeds primary care or community populations, meaning the PPV of 18.0% would likely fall further in lower-prevalence settings.
Grad-CAM findings were not quantified across the full cohort and should be regarded as hypothesis-generating rather than mechanistic proof.
Discriminative performance was modestly attenuated in patients with bundle branch block.

Ask about this paper

Answers are grounded in this paper and cite their sources.

Implication

This editorial highlights a large-scale Japanese study demonstrating that an AI-ECG model detects aortic stenosis with high negative predictive value (98.0%) by extracting QRS morphology features in the limb leads beyond conventional LVH voltage criteria.

Abstract

No abstract yet. The journal hasn't published one we can show.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Ishizu et al. (2026) studied this question.

synapsesocial.com/papers/6ae9d3d1422f9cdec5f560a8 https://doi.org/10.1253/circrep.cr-26-0294

View Full Paper

Ask AI

Bookmark

Share

Artificial Intelligence-Based Electrocardiography for Aortic Stenosis ― Moving Beyond Voltage Criteria Toward Real-World Screening ―

Key Points

Structured PICO

Limitations

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

Artificial Intelligence-Based Electrocardiography for Aortic Stenosis ― Moving Beyond Voltage Criteria Toward Real-World Screening ―

Key Points

Structured PICO

Limitations

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading