Reconstruction of the 12-lead electrocardiogram from reduced lead sets

In clinical practice, continuous recording of all leads of the 12-lead electrocardiogram (ECG) is often not possible. We wanted to assess how well absent, noisy, or defective leads can be reconstructed from different lead subsets and how well lead reconstruction performs over time. A data set of 234 24-hour ECG recordings was divided into an equally sized training and test set. Precordial leads were systematically removed, and for all lead subsets including both limb leads and at least one precordial lead, the absent leads were reconstructed using general and patient-specific reconstruction templates. Reconstruction performance was measured by correlation between the original and reconstructed leads over the QRS and T waves, by average and maximum absolute ST differences, and by agreement when a clinical decision rule was applied. Reconstruction performance over time was evaluated at baseline, at 20 minutes, and 1, 6, 12 and 24 hours after the start of each recording. Reconstruction accuracy was high (correlation ≥0.932, average ST difference ≤30 μV, agreement ≥94.9%) with general reconstruction for lead sets with 1 or 2 precordial leads removed but was less satisfactory when more leads were missing. Patient-specific reconstruction performed well when up to 4 precordial leads were removed (correlation ≥0.967, average ST difference ≤26 μV, agreement ≥95.7%). Patient-specific reconstruction performance initially slightly decreased and then stabilized over time but remained much better than general reconstruction after 24 hours. Accurate reconstruction of the 12-lead ECG from lead subsets is possible over time. General reconstruction allows reconstruction of 1 or 2 precordial leads, whereas up to 4 leads can be reconstructed well using patient-specific reconstruction.