Derived per-participant measurements for: Accuracy of ECG-to-VCG transformations for 3D atrial vectorcardiography

Derived per-participant measurements supporting the manuscript "Accuracy of ECG-to-VCG transformations for 3D atrial vectorcardiography: comparison with simultaneously recorded Frank lead VCG." The study compares three established 12-lead-to-vectorcardiogram transformations — inverse Dower, Kors regression, and Guillem PLSV — against the simultaneously recorded Frank vectorcardiogram, using the atrial P wave as the target signal. The file AtrioMarkV_derived_dataset.csv contains one row for each of 129 analyzed recordings: 44 participants classified as normal and 85 patients with myocardial infarction, one recording per person. Each row carries the original PTB patient and record identifiers, the operator-defined P-wave and QRS landmarks, the recorded-Frank-lead quality-control correlations, and, for each of the three transformations, the per-axis Pearson correlation with the recorded Frank lead, the per-axis root-mean-square error, the three-dimensional peak-vector angular error, and the peak and area vector magnitudes. README.md gives the full data dictionary, the processing steps, and instructions for reproducing each published table. Source recordings come from the PTB Diagnostic ECG Database version 1.0.0, PhysioNet (https://doi.org/10.13026/C28C71), distributed under the Open Data Commons Attribution License (ODC-By) version 1.0. This work contains information from the PTB Diagnostic ECG Database, which is made available under the ODC Attribution License. Because the original record identifiers are retained, every value here can be traced back to its source signal. The dataset contains no personally identifiable information; the source recordings are fully anonymized and publicly available. Two scripts are included so that the published results can be regenerated without proprietary software. reproduce_tables.py(Python 3, requires numpy, pandas and scipy) recomputes every statistic in Tables 1 to 4, and figure3.R (R, requires ggplot2 and patchwork) regenerates Figure 3. Both read only the deposited CSV. Note on precision: the numeric columns are given at full precision and should not be rounded. Several paired differences on the Vy axis are smaller than 0.01, and rounding to three decimals materially alters the repeated-measures statistics.