Multifidelity Uncertainty Quantification for Focused Ultrasound Breast Cancer Therapies Using Reduced Order Models

Area of Science:

• Medical physics
• Computational modeling
• Biomedical engineering

Background:

• Focused ultrasound (FUS) is a noninvasive thermal therapy for destroying diseased tissue.
• Computational simulations aid in planning FUS treatments, but require uncertainty quantification (UQ) for reliability.
• Traditional UQ methods are computationally expensive due to repeated simulations.

Purpose of the Study:

• To evaluate a multifidelity UQ technique using reduced order models (ROMs) for FUS thermal simulations.
• To compare the accuracy and efficiency of multifidelity UQ against traditional Monte Carlo UQ.
• To assess the impact of multifidelity UQ on predicting treatment outcome uncertainties.

Main Methods:

• Implemented a multifidelity UQ approach with projection-based ROMs as low-fidelity models.
• Applied the technique to thermal simulations of FUS treatments for breast cancer.
• Compared error in mean response estimates and distribution predictions (mean, std, skewness) with Monte Carlo UQ.

Main Results:

• Multifidelity UQ reduced mean response estimation error by up to 50% compared to Monte Carlo UQ.
• Both methods showed similar accuracy in predicting the distribution of quantities of interest.
• Pearson Type III distributions were used to analyze the predicted outcome distributions.

Conclusions:

• Multifidelity UQ offers a more computationally efficient approach for uncertainty quantification in FUS simulations.
• This technique enhances the reliability of predictive models for FUS treatments.
• Multifidelity UQ can lead to faster, safer, and more effective FUS treatment planning.