Reconstruction of electrocardiogram (ECG) is the synthesizing of ECG leads from a subset of leads. The need for this can arise when the recording system has a limited number of leads, there is noise in some recorded leads, or there is a need to compress and decompress ECG signals for efficient transmission and interpretation. Although numerous studies have explored optimal protocols for selecting lead subsets that minimize reconstruction error, no standardized approach and lead set has been established.
This study evaluates six well-known protocols for ECG reconstruction via linear transformation. The protocols analysed include the cross-correlation method, the 3 approaches by Finlay et al. (2006), the Lux et al. (1978) technique, and the Barr et al. (1971) technique. According to Schreck et al (1998), three lead accounts for about 99% of the 12-lead ECG information content. Based on this finding, the present analysis employs only the top three leads selected by each protocol for re-construction. The study was conducted using 1,000 unique 10-second nor-mal ECG recordings from the CODE 15% database, which had been de-noised and resampled to 500 Hz. A 10-fold patient-wise cross-validation was then performed on the dataset using the selected three leads for reconstruction.
The results indicate that the precordial lead V3 and limb leads performed best in reconstructing the 12 lead ECG. This gave an overall correlation (r) of 0.93 ± 0.12 and RMSE of 0.10 ± 0.13mV. These findings align with previous studies suggesting that one precordial lead and two limb leads are optimal for ECG reconstruction. However, based on the dataset analysed, this study conclusively recommends using any two limb leads in combination with pre-cordial lead V3, as also suggested by Butchy et al. (2023).