Optimization of energy efficiency factor for high-intensity focused ultrasound ablation using machine learning and genetic algorithm

High-Intensity Focused Ultrasound (HIFU) is a potent treatment for solid tumors. The Energy Efficiency Factor (EEF), defined as the acoustic energy needed to ablate a unit volume of tissue, is pivotal for assessing HIFU's efficiency. However, determining the optimal transducer configuration and irradiation parameters remains challenging. A hybrid approach combining Machine Learning (ML) and Genetic Algorithm (GA) was utilized to minimize the EEF. Multi-Layer Perceptron (MLP), Radial Basis Function (RBF), and Extreme Gradient Boosting (XGB) models were employed to predict EEF across various HIFU transducer settings and irradiation conditions. The most accurate model was chosen for further refinement using GAs to identify minimal EEF values. The MLP model demonstrated superior performance in EEF prediction over RBF and XGB. The MLP-GA model effectively determined the optimal parameters for HIFU ablation, achieving minimal EEF values of 0.68 J/mm³ for continuous wave and 0.70 J/mm³ for pulsed wave modes. The differences between the MLP-GA model's findings and those of a traditional grid search were approximately 3.0%. The integrated ML-GA approach is effective for optimizing HIFU transducer design and irradiation settings to achieve maximal efficiency, defined as the minimal EEF, in simulated coagulative ablation scenarios.