Optimizing pulsed field ablation for cardiac arrhythmias integrating Taguchi method, machine learning and genetic algorithms

• Novel PFA framework integrating FEM, Taguchi method, and machine learning. • Effect of critical factors on the efficacy of PFA is investigated. • Machine learning reduces simulation time by 99.9%, enabling seamless clinical use. • Developed framework advances personalized PFA treatment for cardiac arrhythmia. Pulse field ablation (PFA) is a promising irreversible electroporation-based treatment modality for cardiac arrhythmias. Yet, optimal pulse train parameters and electrode configurations remain undefined. This study presents a novel approach integrating finite-element modeling (FEM), the Taguchi method, machine learning (ML), and multi-objective optimization to enhance PFA outcomes. A comprehensive FEM model simulating electrical, thermal, and fluid dynamics in cardiac tissue evaluates critical factors, viz., pulse amplitude, inter-pulse delay, pulse number, blood flow velocity, and electrode contact depth. Taguchi’s L27 orthogonal array is utilized to quantify the effect of selected factors on PFA outcomes. Simulation data is used to train a range of ML models, such as support vector machines, decision trees, Gaussian processes, and neural networks, incorporating data-augmentation techniques. The Gaussian process model exhibited strong predictive performance for ablation volume and maximum tissue temperature, achieving mean absolute errors (MAE) of 0.2 mm³ and 0.1 °C during training, and 1.1 mm³ and 0.5 °C during testing, and is used for multi-objective optimization via the non-dominated sorting genetic algorithm (NSGA-II). Pulse amplitude emerged as the most statistically significant factor affecting both ablation volume and temperature rise. Electrode insertion depth positively correlated with the ablation volume, whereas blood flow velocity showed minimal impact. The ML-integrated framework allowed rapid prediction of treatment outcomes, reducing computation time from hours to milliseconds and enabling real-time optimization. The generalized framework presented in this study, integrating computational modeling, statistical design, ML, and genetic algorithms, paves the way forward for tailoring PFA protocols to patient-specific needs, thus advancing personalized treatment strategies for arrhythmia.