A geometric signal-processing framework called the ECG time-frequency "eyeball" successfully differentiated acute myocardial infarction (AMI) from healthy cardiac rhythms using just 30 seconds of single-lead ECG data. Applied to 170 healthy subjects and 80 AMI patients, the method — combining empirical mode decomposition with Hilbert-based analytic signal mapping — produced rotational and geometric features that achieved individual AUCs up to 0.78. AMI cases showed higher rotational frequency metrics, lower envelope metrics, and displaced centroid positions compared to controls, with stable outputs across windows from 30 seconds to 5 minutes.
The broader significance lies in what this framework is NOT doing: unlike dominant deep-learning ECG models, it generates interpretable geometric representations clinicians can actually examine and reason about. Explainability has been a persistent bottleneck in AI-assisted cardiology, and this approach offers a middle path between simple rule-based algorithms and opaque neural networks. An AUC of 0.78 from a single feature is modest but meaningful as a screening signal, especially for consumer wearables like Apple Watch or similar single-lead devices where conventional 12-lead interpretation is impossible. Critical limitations include the relatively small AMI cohort (n=80), lack of multi-site validation, and no comparison against existing wearable-optimized algorithms. The cohort size also raises concerns about demographic generalizability. As a preprint not yet peer-reviewed, these findings should be considered preliminary — independent validation in larger, diverse populations will be essential before clinical deployment.
