Background: The function and myocardial characteristics of the right ventricle (RV) are linked to RV dysfunction and prognosis in pulmonary arterial hypertension (PAH). The prognostic value of RV myocardial features derived from echocardiography remains unclear. Methods: A total of 166 patients (mean age, 35.6±11.7 years; 133 females) were included. The primary endpoint was a composite of cardiovascular hospitalization and all-cause mortality. RV myocardium was manually segmented using ITK-SNAP. DenseNet161 was employed to build the deep transfer learning (DTL) model. Seven machine learning algorithms were used to construct radiomics (Rad) and deep learning-based radiomics (DLR) models. A combined nomogram was developed, with its predictive performance assessed using time-dependent ROC analysis and Kaplan-Meier estimates. Results: The three myocardial models were constructed, and their performance was evaluated using ROC, calibration curves, decision curve analysis (DCA), and other relevant metrics. The DLR model outperformed the others, demonstrating superior AUCs in both the training and test cohorts (AUC = 0.977, 0.875). The combined nomogram demonstrated excellent performance in survival ROC analysis, with AUCs of 0.763, 0.824, and 0.935 for 1-, 3-, and 5-year predictions, respectively. Furthermore, the DLR score added prognostic value to the COMPERA 2.0, as demonstrated by a global chi-square and C-statistic comparison (P < 0.001 for both). Conclusion: Echocardiographic RV myocardial features, analyzed using DLR model, were strongly correlated with RV dysfunction and prognosis in PAH. The combined nomogram demonstrated superior predictive value. These results highlight the potential of myocardial features for improved risk stratification and prognosis in PAH.
Xie et al. (Tue,) studied this question.