Modeling Focused-Ultrasound Response for Non-Invasive Treatment Using Machine Learning.

Area of Science:

• Biomedical Engineering
• Acoustics
• Computational Modeling

Background:

• Accurate simulation of focused ultrasound (FUS) interactions with biological tissues is crucial for therapeutic applications.
• Traditional numerical models like Rayleigh-Sommerfeld and angular spectrum methods are accurate but computationally intensive, requiring hours to days for complex scenarios.
• This computational burden limits the widespread adoption of FUS in clinical settings, particularly for real-time treatment planning.

Purpose of the Study:

• To develop a rapid estimation model for focused ultrasound therapy using machine learning.
• To address the computational expense of traditional numerical models for simulating FUS-tissue interactions.
• To improve the efficiency and practicality of therapeutic planning in focused ultrasound applications.

Main Methods:

• Trained multiple machine learning models on a large dataset of 19,227 FUS simulations.
• Utilized a random forest algorithm as the primary machine learning approach.
• Evaluated model performance using metrics including Root Mean Squared Error (RMSE), R-squared (R²), Akaike Information Criterion (AIC), and Bayesian Information Criterion (BIC).

Main Results:

• The developed random forest model achieved high accuracy on an external test dataset.
• Achieved an R-squared value of 0.997, indicating excellent model fit.
• Reported an RMSE of 0.0123, with AIC of -82.56 and BIC of -81.65, demonstrating superior predictive performance.

Conclusions:

• The random forest-based model effectively estimates focused ultrasound responses in biological tissues.
• This machine learning approach significantly minimizes simulation time compared to traditional methods.
• Practical adoption of this rapid estimation model can enhance therapeutic planning and optimize focused ultrasound treatments.