A standard ECG can now yield synthetic heart-muscle motion: synthetic e′ predicts post-discharge death on its own and lifts discrimination from 0.67 to 0.70

Left ventricular mechanical function after myocardial infarction carries prognostic information that has mostly come from echocardiography; the same longitudinal mechanical indices can now be synthesized from a routine 12-lead ECG and used for risk stratification. Echocardiographic assessment is episodic, capturing isolated snapshots, and is often unavailable when critical decisions are made, whereas a 12-lead ECG is obtained in essentially every acute MI patient. Among 1,065 hospitalized acute MI patients (109 deaths, median follow-up 1.44 years), synthetic e′ independently predicted all-cause mortality after adjustment for GRACE 2.0, with a hazard ratio of 0.90 per 1 cm/s (95% CI 0.83–0.97, P=0.006).

In the post-discharge setting the authors pre-specified as the clinical point of interest, the prognostic information in synthetic e′ is largely independent of ejection fraction, so it adds a line of evidence beyond systolic function. Earlier AI-ECG models mostly output binary probabilities against ejection-fraction thresholds, which do not capture diastolic dysfunction and are awkward to track over time. Among 1,017 patients discharged alive (61 deaths, median follow-up 536 days), synthetic e′ raised the GRACE C-index from 0.67 to 0.70; beyond the AiTiA and Pan et al. AI-ECG models and measured LVEF below 50%, the hazard ratio was 0.84–0.86 per 1 cm/s (P≤0.004), contributing 28%–46% of each model's likelihood-ratio χ².

Methodologically, a generative adversarial network decodes the ECG into myocardial velocity waveforms, returning continuous curves shaped like ultrasound tissue Doppler rather than an isolated risk score. A generative adversarial network pits a generator against a discriminator so that generated waveforms approach real tissue Doppler recordings; the network was applied to all cohorts without retraining and blinded to outcome. Raw ECG tracings were segmented into cardiac cycles, resampled to 512 timesteps and synchronized with paired tissue Doppler recordings acquired within one hour; in 219,564 outpatients (817 MI deaths, median follow-up 3.42 years), after adjustment for age, sex and conventional ECG abnormalities, synthetic MAPSE had a subdistribution hazard ratio of 0.49 (95% CI 0.37–0.65), s′ 0.78 (0.72–0.84) and e′ 0.87 (0.82–0.93).

The demonstrated scope is prognostic association in two observational cohorts: 1,065 hospitalized acute MI patients at two US hospitals, with all-cause mortality as the outcome, and 219,564 outpatients in the Brazilian CODE-15 registry, with MI-specific mortality as the outcome. Synthetic parameters came from a generative adversarial network applied without retraining and blinded to outcome, and all survival analyses were performed in the MI cohort alone; the non-MI controls served only for waveform characterization and risk-factor association analyses.

These analyses establish observational association and do not show that decisions guided by the synthetic parameters improve outcomes; the dynamic mechanical fidelity of synthetic tissue Doppler during acute ischemia is addressed separately and still requires head-to-head comparison against simultaneously measured tissue Doppler. Because the signal derives from the ECG, part of it may reflect a broad ECG-derived risk marker, and dedicated comparative and prospective study is needed to establish mechanism specificity. The CODE-15 registry does not capture comorbidity burden, prior cardiovascular disease or echocardiographic data, so those confounders could not be adjusted for directly.