Atrial lead system for enhanced P-wave recording: A comparative study on optimal leads using gradient boosting and deep learning algorithms

Abstract

Morphological characteristics of the P-wave in an Electrocardiogram (ECG) provide insights into interatrial and atrioventricular (AV) conduction pathologies. However, accurate diagnosis through visual inspection is hindered by the P-wave’s diminutive size and susceptibility to noise. Furthermore, conventional 12-lead systems often fail to capture these elusive P-waves. Consequently, our study describes a new Atrial Lead System (ALS) aimed at enhancing P-wave signal strength and ranking leads using advanced Gradient-Boosting (GB) and Deep Learning (DL) algorithms. ECG data from 75 healthy volunteers (mean age: 45 ± 17.2 years; 40 % women) were recorded using ALS and specially designed optimal bipolar leads to improve atrial activity. Employing Gradient Boosting Classifier (GBC), Extreme Gradient Boosting Classifier (XGBoost), Light Gradient Boosting Machine (LightGBM), and a synergistic One-dimensional Convolutional Neural Network (1D-CNN) with Bidirectional Long Short-Term Memory (Bi-LSTM) layers, we assessed and ranked optimal bipolar leads based on P-wave parameters, indices, and AV ratios. Notably, AL-I and AL-II exhibited significantly higher median amplitudes, RMS, and area for the recorded P-wave compared to other leads (p < 0.001). Evaluations using LightGBM, GBC, XGBoost, and 1D-CNN+Bi-LSTM models identified P-lead, AL-II, and AL-I as the top three leads, with accuracies of 0.96, 0.95, and 0.94; 0.96, 0.94, and 0.94; 0.96, 0.95, and 0.94; and 0.96, 0.94, and 0.93, respectively. These findings underscore the efficacy of ALS in elevating P-wave signal strength and AV ratios, suggesting its potential as a valuable tool for improved clinical screening and diagnosis of atrial arrhythmias.