Development of patient-specific apparent blood viscosity predictive models for computational fluid dynamics analysis of intracranial aneurysms with machine learning approaches.

A model to predict patient-specific apparent viscosity as a computational condition in computational fluid dynamics (CFD) analysis, which is used in research on intracranial aneurysms, is important. The purpose of this study was to develop a model to predict patient-specific apparent viscosity from clinical blood test results. The data were from 15 patients with intracranial aneurysms in whom blood viscosity and density were measured and blood tests were performed on the same day. The dataset was divided into two, a training dataset and a test dataset at a ratio of 4:1. The training dataset was used in constructing regression models with shear rate and 12 blood test items (the flexible model) or hematocrit (the simple model) as input, and the measured apparent viscosity as output. CFD analysis was implemented with and without coil geometries, and the viscosity models were evaluated. The root mean squared error (RMSE) of viscosity predicted with the flexible model and the simple model was 0.136 mPa·s and 0.226 mPa·s, respectively. The RMSE of time-averaged and space-averaged velocity and time-averaged and space-averaged wall shear stress computed in CFD analysis were <0.01 m/s and <0.21 Pa, respectively. Regression models to predict patient-specific apparent blood viscosity from shear rate and blood test items were constructed with machine learning. There is a possibility that, using this predictive model, patient-specific blood apparent viscosity can be predicted with high accuracy from the blood test results of individual patients.