Generalization of ML Models Between ECG and VCG Representation

Integrating heterogeneous data sources is vital for developing and validating robust medical machine learning models. Although the 12-lead format is standard in clinical electrocardiography (ECG), many datasets include only single-lead or vectorcardiogram (VCG) recordings. This study examines the transferability of 12-lead ECG data to VCG representations. Results show that ECG-VCG transfer depends on the acquisition system, whether clinical or Holter-based, and that the V6-X lead configuration achieves the most stable cross-lead performance.