A 90-second magnetic recording offers a new view of the injured heart

The coronary catheterization lab is full of technology, yet identifying the artery responsible for a non-ST-elevation heart attack before angiography can still be difficult. This group asked whether optically pumped magnetometers could provide useful additional information in a recording lasting only ninety seconds.

A geographically separate testing campus improves credibility over a single retrospective training set. Still, the locked algorithm was correct on about two-thirds of three-vessel classifications and missed many left circumflex culprit vessels. A high ROC AUC for one artery must not obscure that clinical asymmetry.

The genuine breakthrough would not be accuracy alone. It would be a prospective demonstration that the information improves triage, prevents delays, changes management safely, or reduces patient-important harm.

The evidence: 611 hemodynamically stable NSTEMI patients already selected for angiography; XGBoost model on 65 prespecified magnetocardiography features trained in campus 1 (n=493), locked before campus 2 test (n=118). In external campus test, overall vessel identification accuracy was 63.6%; matched ECG-only model 51.7%; integrated standard care 55.9%. The exploratory addition of magnetocardiography probabilities to standard care achieved 67.8% accuracy; sensitivity was unequal across arteries (LAD 76.6%, LCx 34.5%, RCA 69.0%).