# Optimization of the Parameters of a Minimal Coagulation Model

## TL;DR

This paper introduces a method to optimize a simplified blood clotting model to match detailed models while reducing computational costs.

## Contribution

A novel approach using machine learning to adapt minimal coagulation model parameters for accurate thrombin generation.

## Key findings

- Machine learning improves agreement between minimal and comprehensive coagulation models.
- Optimized minimal model achieves accurate thrombin generation with reduced computational cost.
- Results enable future applications in complex simulations and personalized medical interventions.

## Abstract

The formation of a blood clot within a vessel can result in its complete blockage. This phenomenon, known as thrombosis, can have severe consequences. In contrary, thrombosis can be sometimes desirable. Intra-aneurysmal thrombosis is the primary objective of an endovascular treatment aimed at occluding the aneurysm sac. The proper modeling of the coagulation system is, therefore, important for the prediction, early recognition, and prevention of these tendencies. In silico investigations based on computational fluid dynamics (CFD) extended by thrombosis models provide a valuable tool for a detailed analysis. Minimal models are particularly useful for practical purposes to reduce computational efforts. This work proposes an approach to adapt the parameters of a minimal model to reproduce the behavior obtained with a comprehensive description of the coagulation cascade. The objective is to obtain the same thrombin generation curves while reducing strongly computational costs. For this purpose, machine learning—based here on an evolutionary algorithm—is used to optimize the obtained agreement. By adapting the reaction rate coefficients, a significant improvement can be achieved. The obtained results pave the way for future applications of the improved model in complex configurations such as for planning personalized interventions. Notably, the minimal model will be used for CFD in future studies to take advantage of its low computational cost.