Noninvasive Prediction of Pulmonary Hypertension Based on Finite Element Analysis and Machine Learning

Geometry morphological features (GMF) of the ventricles were often used to predict the pulmonary hypertension (PH) in the clinical process. This study analyzed the relationship between right ventricular pressure (RVP) and ventricular morphological changes in simplified statistical shape models which were analyzed with finite element analysis method, and the RVP higher than 40 mmHg was adopted as a criterion to determine the presence of PH. Ten GMF features were utilized and three classifiers (decision tree, SVM and random forest) were performed to predict PH and achieved recognition accuracy of 90.7%, 91.4% and 93.5%, respectively. Through this study, the PH can be identified by morphometric features measured from ventricular images.