AI-Enabled Accurate Non-Invasive Assessment of Pulmonary Hypertension Progression via Multi-Modal Echocardiography

TL;DR

It is shown that MePH significantly outperforms echocardiographers' assessments using echocardiography, reducing the mean absolute error in estimating mean pulmonary arterial pressure and pulmonary vascular resistance by 49.73% and 43.81%, respectively.

Abstract

Echocardiographers can detect pulmonary hypertension using Doppler echocardiography; however, accurately assessing its progression often proves challenging. Right heart catheterization (RHC), the gold standard for precise evaluation, is invasive and unsuitable for routine use, limiting its practicality for timely diagnosis and monitoring of pulmonary hypertension progression. Here, we propose MePH, a multi-view, multi-modal vision-language model to accurately assess pulmonary hypertension progression using non-invasive echocardiography. We constructed a large dataset comprising paired standardized echocardiogram videos, spectral images and RHC data, covering 1,237 patient cases from 12 medical centers. For the first time, MePH precisely models the correlation between non-invasive multi-view, multi-modal echocardiography and the pressure and resistance obtained via RHC. We show that MePH significantly outperforms echocardiographers' assessments using echocardiography, reducing the mean absolute error in estimating mean pulmonary arterial pressure (mPAP) and pulmonary vascular resistance (PVR) by 49.73% and 43.81%, respectively. In eight independent external hospitals, MePH achieved a mean absolute error of 3.147 for PVR assessment. Furthermore, MePH achieved an area under the curve of 0.921, surpassing echocardiographers (area under the curve of 0.842) in accurately predicting the severity of pulmonary hypertension, whether mild or severe. A prospective study demonstrated that MePH can predict treatment efficacy for patients. Our work provides pulmonary hypertension patients with a non-invasive and timely method for monitoring disease progression, improving the accuracy and efficiency of pulmonary hypertension management while enabling earlier interventions and more personalized treatment decisions.

Figures (6)

• Figure 1: Overview of our study.a, Dataset collection shows that each patient underwent two types of examinations: 1. Echocardiography performed by echocardiographers, which includes echocardiogram videos from four views, spectral images from four views, and metadata; 2. RHC, which serves as the "gold standard" for accurately obtaining mPAP and PVR. b, Study cohorts presents the data statistics of our multi-center PH dataset, which includes a total of 1,237 patients from 12 medical centers. c, Model development shows that our MePH model was trained on a collected multi-view, multi-modality dataset to learn dynamic cardiac motion from echocardiogram videos, extract hemodynamic information, and incorporate metadata such as gender and age. d, Real world clinical application of our MePH shows its capability to predict the severity of PH and assess treatment efficacy, ultimately enhancing overall survival for PH patients.
• Figure 2: Comparison of PH recognition and PVR prediction results for MePH, echocardiographers, and RHC in both internal and external tests.a, Results of PH recognition from the internal test set across four medical centers (A, B, C, D) with a test sample size $n=100$. b, Results of PH progression assessment from the external test collected from eight medical centers (E, F, ..., L) with a test sample size of $n=237$. c, Results of PH recognition from the internal test set across four medical centers (A, B, C, D) with a test sample size $n = 100$. d, Results of PH progression assessment from the external test were collected from eight medical centers (E, F, ..., L) with a test sample size of $n=237$. The Bland-Altman plots are presented based on the estimated mPAP and PVR compared to the ground-truth mPAP and PVR measured by RHC, illustrating the limits of agreement (dotted lines) ranging from -1.96 to +1.96 standard deviations.
• Figure 3: The PH progression assessment efficacy, the prospective study of treatment efficacy, and the performance comparison across different subtypes of PH.a, The PH progression assessment efficacy in the internal test set. b, The PH progression assessment efficacy in the external test set c, A prospective study to show the effectiveness of MePH for evaluating treatment efficacy. The predicted ($\Delta\text{PVR}$) is calculated for the same patients by comparing the initial $\text{PVR}^\text{1st}$ accessed by RHC and the follow-up $\text{PVR}^\text{2nd}$ accessed by the echocardiographer and our MePH, respectively ($\Delta\text{PVR}=\text{PVR}^\text{2nd}-\text{PVR}^\text{1st}$). The actual ($\Delta\text{PVR}$) is calculated by RHC over the two stages. d, The overall performance across different PH subtypes. All results in tables of each PH subtype were reported by mean absolute error (MAE) and the coefficient of determination $R^2$ for PH classification based on mPAP (mmHg) and PH progression assessment based on PVR (WU). e, The performance of mPAP assessment in IPAH, CTD-PAH, CHD-PAH and others (HPAH and PoPH). f, The performance of PVR assessment in IPAH, CTD-PAH, CHD-PAH and others (HPAH and PoPH). All results were reported in the external dataset. For the full names of these Pulmonary Hypertension subtypes, please see our Supplementary Table \ref{['tab:ph_etiology']}.
• Figure 4: Illustration of the model decision with activation heatmap on different cases with PH. The saliency map (heat map) was generated by EigenCAM muhammad2020eigen with no class discrimination, which is used to visualize which areas help the model assess the mPAP and PVR. The scale bar spans from 0 to 1, with a value of 1 representing the highest level of influence derived from the normalized EigenCAM value, while 0 indicates the lowest possible influence. a, The activation maps from four cardiac views can be used to observe cardiac motion. The highlighted active regions—the interventricular septum (IVS)—separate the right ventricle (RV) from the left ventricle (LV), as well as the pulmonary artery wall (PAW). The tables at the bottom present mPAP and PVR, measured by RHC, evaluated by MePH, and assessed by echocardiographers through echocardiography. b, Images show the correlation of temporal and spatial information presented by cardiac structures in PSAX views. MePH highlights clinically relevant regions in echocardiogram frames, reflecting structural changes associated with elevated pressure and vascular resistance. These patterns correspond to established indirect signs of PH. In contrast to conventional estimation methods, MePH captures such features without requiring echocardiography signals, enabling parameter-free interpretation and contributing to the model’s superior performance in assessing mPAP and PVR.
• Figure 5: Comparison of PH recognition and PVR prediction with vision-language based method and echocardiography parameter-based methods. a, Results of PH recognition from the internal test set across four medical centers (A, B, C, D) with a test sample size $n = 100$. b, Results of PH progression assessment from the external test collected from eight medical centers (E, F, ..., L) with a test sample size of $n = 237$. c, Results of PH recognition from the internal test set across four medical centers (A, B, C, D) with a test sample size $n = 100$. d, Results of PH progression assessment from the external test collected from eight medical centers (E, F, ..., L) with a test sample size of $n = 237$. All results are reported by metrics MAE and $R^2$, PH classification is assessed by mPAP and PH progression is assessed by PVR.