Study on parameter optimization for support vector regression in solving the inverse ECG problem.

Area of Science:

• Biomedical Engineering
• Computational Electrophysiology
• Medical Imaging

Background:

• The inverse electrocardiography (ECG) problem involves reconstructing transmembrane potentials (TMPs) from body surface potentials (BSPs).
• Support Vector Regression (SVR) is a viable method for solving this multi-input, multi-output regression problem.

Purpose of the Study:

• To compare the effectiveness of three optimization algorithms—Genetic Algorithm (GA), Differential Evolution (DE), and Particle Swarm Optimization (PSO)—in determining optimal hyperparameters for SVR in inverse ECG problems.
• To identify the most efficient optimization method for accurate cardiac TMP reconstruction from BSPs.

Main Methods:

• Treating the inverse ECG problem as a regression task using SVR.
• Employing GA, DE, and PSO algorithms to optimize SVR hyperparameters.
• Evaluating and comparing the performance of GA, DE, and PSO in reconstructing cardiac TMPs.

Main Results:

• All three optimization methods (GA, DE, PSO) successfully identified appropriate SVR parameters, demonstrating good generalization performance.
• Particle Swarm Optimization (PSO) exhibited superior efficiency in hyperparameter optimization compared to GA and DE.
• PSO achieved better performance in solving the inverse ECG problem, resulting in more accurate TMP reconstructions.

Conclusions:

• PSO is the most effective optimization algorithm for SVR in the context of the inverse ECG problem.
• The study validates the utility of GA, DE, and PSO for improving the accuracy of noninvasive cardiac potential mapping.