Non-Invasive Estimation of Haemodynamic Parameters in Pulmonary Hypertension — A Deep Learning Approach Integrating All B-Mode Cine Loops in an Echocardiographic Exam

Abstract

Pulmonary hypertension (PH) is heterogeneous with treatment strategy dependent on the underlying cause and disease severity. Haemodynamic parameters measured through right heart catheterization (RHC) is the gold standard for such diagnosis and desicion making. However, the invasive procedure is associated with a certain level of risk and is not suitable for every patient. Therefore, we seek to investigate whether haemodynamic parameters can be estimated non-invasively using a deep learning approach. The study is based on a retrospective analysis of 833 subjects with suspected PH identified from the ASPIRE research database. Convolutional neural networks were built to integrate B-mode echocardiographic cine loops from multiple views to predict key haemodynamic parameters. The model was able to integrate an arbitrary number of cine loops in the entire exam, unannotated with view names. Additionally, attention weights in feature fusion identify relevant and irrelevant cine loops to the model. The model-predicted mean pulmonary artery pressure (mPAP) correlated to the RHC-ground truth with a Pearson Correlation Coefficient (PCC) of 0.70. Attention weights indicated the apical 4-chamber (A4C) view to be especially relevant for mPAP prediction. Our results demonstrate the feasibility of estimating haemodynamic parameters non-invasively through deep learning models, integrating all B-mode cine loops of a cardiac ultrasound exam, achieving a moderate correlation to RHC measurements.