Optimization framework for patient-specific modeling under uncertainty.

Area of Science:

• Computational modeling
• Uncertainty quantification
• Biomedical engineering

Background:

• Patient-specific computational models require reliable parameter estimation.
• Existing deterministic approaches struggle with unreliable and ill-suited data.
• Uncertainty in data complicates accurate model parameterization.

Purpose of the Study:

• To develop an optimization-based uncertainty quantification framework for probabilistic model tuning.
• To discover model input distributions that generate target output distributions.
• To enhance the reliability of patient-specific computational models.

Main Methods:

• Developed an optimization-based uncertainty quantification framework.
• Utilized probabilistic sampling with a surrogate model for computational efficiency.
• Employed a general distribution parameterization for model inputs.
• Validated the approach on seven patient-specific cardiovascular model examples using CircAdapt.

Main Results:

• Accurately reproduced target output distributions in all tested examples.
• Successfully recreated reference input distributions for synthetic examples.
• Demonstrated the framework's efficacy with both synthetic and real-world patient data.
• Showcased accurate parameter distribution estimation for patient-specific models.

Conclusions:

• The proposed framework effectively handles uncertain data in patient-specific modeling.
• It enables accurate determination of parameter distributions for improved model reliability.
• The approach offers insights into model parameter sensitivity to measured data.