The study introduces a novel approach, the frequency-enhanced geometric-constrained iterative network (FGIN), to improve the localization of myocardial infarction using 12-lead electrocardiograms (ECG). FGIN addresses the challenge of ill-posed surface potential reconstruction by analyzing ECG data across time and frequency domains, enhancing data dimensionality, and incorporating ventricular geometry as a constraint. This method stands out by assigning variable weights to different edges in the reconstruction process.

FGIN’s performance was validated through experiments on both synthetic and clinical datasets. It surpassed seven existing methods in the synthetic dataset with a Pearson Correlation Coefficient of 0.8624, a Root Mean Square Error of 0.1548, and a Structural Similarity Index Measure of 0.7988. In the clinical setting, using the 2007 PhysioNet/Computers in Cardiology Challenge dataset, FGIN achieved superior localization accuracy according to the 17-segment model with an average Segment Overlap of 87.2%. Furthermore, clinical trials on 50 patients confirmed FGIN’s effectiveness, showing an average accuracy of 91.6% and an average Segment Overlap of 88.2%. This advancement is significant for clinical management and therapeutic strategies for myocardial infarction, offering a more precise and reliable diagnostic tool.