The Reconstruction of a 12-Lead Electrocardiogram from a Reduced Lead Set Using a Focus Time-Delay Neural Network.

Area of Science:

• Biomedical Engineering
• Cardiology
• Artificial Intelligence in Medicine

Background:

• The 12-lead electrocardiogram (ECG) is a cornerstone diagnostic tool in cardiology.
• Accurate electrode placement for the 12-lead ECG is challenging, time-consuming, and prone to errors, potentially leading to misinterpretations.

Purpose of the Study:

• To develop and validate a method for accurately reconstructing a full 12-lead ECG from a reduced electrode set.
• To enhance the efficiency and reliability of ECG acquisition and interpretation.

Main Methods:

• Utilized a five-electrode placement strategy to derive seven initial ECG leads (I, II, III, aVL, aVR, aVF, V2).
• Employed a focus time-delay neural network (FTDNN) to reconstruct the remaining five precordial leads (V1, V3-V6).
• Trained, validated, and tested the FTDNN model using a database of 549 archived ECG recordings.

Main Results:

• Reconstructed ECG leads demonstrated high quantitative correlations (0.8609–0.9678) and low root mean square errors (123–245 μV).
• The FTDNN method showed favorable performance across healthy and various cardiovascular disease subgroups, excluding bundle branch block.
• Results compared favorably against existing ECG lead reconstruction techniques.

Conclusions:

• A 12-lead ECG can be accurately reconstructed from a reduced five-electrode set, with four electrodes placed on the limbs.
• Minimizing reliance on precordial lead placement significantly reduces errors.
• This approach promises to improve overall ECG accuracy and decrease diagnostic errors.