Toward Uncertainty-Aware Hemolysis Modeling: A Universal Approach to Address Experimental Variance.

Area of Science:

• Biomedical Engineering
• Computational Fluid Dynamics
• Medical Device Design

Background:

• Numerical hemolysis models are crucial for evaluating blood damage in medical devices.
• Existing models often lack robust uncertainty quantification, limiting their predictive accuracy.
• Experimental data variability presents a significant challenge in model development.

Purpose of the Study:

• To develop a probabilistic hemolysis model incorporating experimental variability.
• To enhance the predictive accuracy and robustness of hemolysis predictions.
• To address the limitations of deterministic models in capturing experimental uncertainty.

Main Methods:

• Applied a grid search to analyze the objective function landscape of a Power Law hemolysis model.
• Utilized Markov Chain Monte Carlo (MCMC) to derive stochastic distributions for model parameters (C, α, β).
• Propagated parameter distributions through a reduced-order model of the FDA benchmark pump.

Main Results:

• Identified a global flat minimum in the objective function landscape, indicating mathematical fitting limitations.
• Converged to optimal parameters: C = 3.515 × 10⁻⁵, with log-normal distributions for α (mean 0.614) and β (mean 1.795).
• The probabilistic model successfully captured both mean and variance in experimental FDA benchmark pump data.

Conclusions:

• Uncertainty quantification via MCMC significantly improves hemolysis model robustness and predictive power.
• The probabilistic model offers better comparison between simulated and in vitro hemolysis experiments.
• This approach has the potential to set a new standard for hemolysis modeling in medical device evaluations.