Coronary disease & ACSDiagnostic or prognostic study
Ultrasonic Texture Analysis for Predicting Acute Myocardial Infarction
- Diagnostic accuracy study
- multisource data
- leave-one-source-out cross-validation
- external validation
Authors' conclusion
This study demonstrates the potential of cardiac ultrasomics to distinguish healthy from infarcted myocardium
Study context
- Population
- Participants from a retrospective single-centre matched case-control dataset, a prospective multicentre matched clinical trial dataset and an open-source international multivendor datasetmultisource data
- Participants
- 684 participants
- Comparison
- Machine learning (ML) model vs Deep features
- Outcome
- Identification of MI
Performance
As reported in the abstract
Identifying MI
| Test or model | AUC |
|---|---|
| ML model | 0.87 (95% CI: 0.84-0.89) |
| Deep features | 0.74 (95% CI: 0.70-0.77) |
- ML model vs Deep features: P < 0.0001 — 0.87 vs 0.74
Further findings
| Outcome | adjusted OR (95% CI) | P value | |
|---|---|---|---|
| ML probability as independent predictor of MI, adjusted for conventional echocardiographic parameters | 1.03 (1.01–1.05) | P < 0.0001 | |
| 0.512Adjusted odds ratio (log scale) |
| Outcome | ML model | Deep features | Estimate | P value |
|---|---|---|---|---|
| Acute MI with myocardial dysfunction (averaged LS <16%) (subgroup) | ||||
| AUC | 0.84 (95% CI: 0.77-0.89) | Not reported in the abstract | ||
| Averaged LS + ML probability vs LS alone · 0.86 vs 0.80 | P = 0.02 |
- Averaged LS + ML probability vs LS alone: LS + ML probability AUC 0.86 (95% CI: 0.80-0.91)
- Averaged LS + ML probability vs LS alone: LS alone AUC 0.80 (95% CI: 0.72-0.87)
Manhattan plot of ultrasomics features
- Discriminated infarcted and noninfarcted segments
- P < 0.001
Original publication
Ultrasonic Texture Analysis for Predicting Acute Myocardial Infarction. JACC Cardiovasc Imaging. Published 15 August 2025. doi:10.1016/j.jcmg.2025.06.018
Summarised from the published abstract; every number is checked against it