Optimization of Patient-specific Mitral Valve Model Using CFD Simulation and Machine Learning

Yingyi Geng1, yue wang2, zhenyin fu3, Zhaokai Kong1, ruiqing dong4, Dongdong Deng5, Jucheng Zhang6, Ling Xia1
1zhejiang university, 2Sir Run Run Shaw Hospital, Cardiac Surgery Department, 3浙江大学, 4The Fourth Affiliated Hospital of Soochow University, 5Dalian University of Technology, 6The Second Affiliated Hospital Zhejiang University School of Medicine

Abstract

Background: Surgical mitral valve replacement is a common approach for treating regurgitation or stenosis, but it highly requires professional surgeons. Additionally, predicting postoperative hemodynamics and clinical outcomes remains challenging due to the complexity of mitral valve geometry. Therefore, designing patient-specific valve models and optimizing them based on hemodynamic performance become the forefront of current research. Methods: In this study, we propose an approach for designing patient-specific mitral valves and predict the most efficient geometry. Firstly, the mitral valve model was reconstructed from epicardial echocardiography data, including mitral annulus, leaflets, papillary muscles, and a network of chordae. Based on the patient-specific model, we modified the mitral annulus size and leaflet material parameters to generate various geometries. Each model was simulated using ABAQUS. We evaluated stresses, orifice area, and leaflet coaptation. The obtained data was divided into training and testing sets. In the training set, we use two parameters (annulus size and leaflet material) as features to train the Decision Tree model with corresponding hemodynamic parameters, used as targets. In the testing set, we used the Decision Tree model to predict hemodynamic performance and examined differences between simulated and predicted. Finally, this method is applied to the design and prediction of optimal model. Results: Based on the patient-specific model, we designed 10 different annulus sizes and 5 different material models. Different annulus sizes had an impact of 1.27±0.27 MPa on stresses and resulted in an average change of 5.6±4.6% in orifice area. The optimization of material was also directly related to leaflet coaptation. The predictive performance was evaluated by comparing actual hemodynamic values, showing no significant difference (p>0.05), which confirmed the efficiency of our method. Conclusion: Our study provides promising insights into the design, prediction and optimization of patient-specific mitral valve models, assisting in the procedure planning of mitral valve surgery.