An Algorithm for the Reconstruction of 4 ECG Lead Signals Based on the Attention Mechanism

This work proposes an algorithm to reconstruct 4 precordial electrocardiogram (ECG) lead signals. Standard cardiovascular disease (CVD) monitoring and detection uses all 12 available ECG leads. However, this number of leads implies a certain complexity of the equipment in terms of size, weight, and power consumption. Computational algorithms aimed at reducing the number of required leads for CVD detection help lower the time consumption and errors due to needing many signal acquisition cables. In this work, an LSTM sequence-to-sequence (Seq2Seq) neural network model with attention takes only 4 ECG leads (I, II, V2, and V5) and outputs the mentioned precordial leads. This proposal contributes to making ECG signal acquisitions easier and more accessible by requiring fewer cables and thus facilitating its use by people with little training. The model achieved a maximum average Pearson correlation coefficient of 0.9707 for all leads. It was validated using the PTB Diagnostic ECG Database.