Exploring the capabilities of machine learning (ML) for 1D blood flow: Application to coronary flow
MetadataVis full innførsel
The aim of this thesis is to explore the capabilities of deep neural networks to reproduce 1D computational models for the pressure in a coronary tree. A machine learning algorithm was implemented. The algorithm was trained with a synthetically generated database of coronary trees, where the anatomical data was retrieved from published literature. A grid search was performed to optimize the hyper-parameters in the machine learning model. Two different models were trained to solve a steady state; coronary blood flow model and Young and Tsai's stenosis model. Correlation between the predicted values was excellent for both models with $r^2= 1 $ for the steady state coronary blood flow model, and $r^2=0.997$ for Young and Tsai's stenosis model. The established ML models were tested with patient specific data. The prediction on the patient specific data showed that the synthetically generated database did not represent the pathological variation of coronary arteries. For a reduced patient specific database, the model predicted the pressure drop along the healthy vessels with a coefficient of determination of 0.799 and for the reduced database with the patient specific stenoses the coefficient of determination was 0.997.