Electrode placement based on multiobjective optimization for electrocardiography t-shirt

Abstract

Although electrocardiography (ECG) t-shirts have some advantages, obtaining the signal-to-noise ratio (SNR) of the captured ECG signal as high as traditional ECGs remains challenging. Reducing the number of electrodes by employing limited-lead systems has been an approach to minimize artifacts. However, the accuracy and correlation of the derived 12-lead ECG remain a problem. Electrode placement for ECG t-shirts should consider two aspects to maximize the SNR including the electrophysiological and practical aspects. These aspects should be quantified for computing purposes. Unfortunately, the existing studies have not quantified the practical aspects. Additionally, the previous research formulated them in a single objective function for optimization, whereas both aspects are independent. This study is aimed to maximize the SNR of ECG t-shirts using limited-lead systems by trading-off between the two aspects. It has three objectives: to improve accuracy and correlation of the synthesized 12-lead ECG by segmenting the ECG waveform, to quantify some factors in electrode placement (including ECG signal amplitude, skin-shirt gap, relative shirt movement, and regional sweat rate) for optimization purposes, and to improve SNR by compromising electrophysiological and practical aspects in the electrode placement. In this study, one cycle of ECG is divided into three segments: P, QRS, and ST. Each segment is transformed to obtain a derived 12-lead ECG signal. This proposed segment-specific (SS) approach is then compared to conventional full-cycle (FC) by using six existing methods: Dower's method with generic coefficients, Dower's method with individual (patient-specific) coefficients, linear regression (LR), 2nd-degree polynomial regression (PR), 3rd-degree PR, and artificial neural network (ANN). Simulations using 3DS Max® and MATLAB® were carried out to quantify the ECG signal amplitude, skin-shirt gap, relative shirt movement, and regional sweat rate into variables in the range of [0,1], called satisfaction degrees. These variables represent the likelihood of the placement of electrodes. Multi-objective optimization (MOO) is employed to find the optimal electrode placement, i.e., high SNR, by compromising electrophysiological and practical aspects. As a result, the new SS approach outperformed the conventional method (FC). It has significantly reduced the transformation error up to 30.94% and improved the transformation correlation as high as 4.89%. The simulations have successfully quantified the electrophysiological and practical aspects of electrode placement into satisfaction degrees. The MOO yielded Pareto optimal solutions to assist decision-makers in selecting the final solution subjectively. Based on the experiment results, this new approach improved the SNR as high as 29.44%. This study provides a comprehensive method for determining the location of the electrodes to support ECG t-shirt manufacturers.