Developing a novel deep learning—based model for automatic right ventricular parameters assessment on ctpa in pulmonary embolism

To develop an automated method for quantifying right ventricular parameters from CTPA in pulmonary embolism (PE). In this study, a U-Net based deep learning model was employed to perform automatic segmentation of the right ventricular, left ventricular, ascending aorta, and pulmonary arteries on CTPA images of patients with PE. The right ventricular to left ventricular diameter ratio (RV/LV), main pulmonary artery to ascending aorta diameter ratio (PA/AA), and ventricular septal angle (SA) were automatically calculated using a deep learning model based on anatomical structures. Additionally, two senior radiologists manually annotated the three sets of features to serve as the gold standard. Intra-class Correlation Coefficients (ICCs) were calculated to assess the inter-observer reliability of the proposed method by comparing the three sets of automatically derived measurements with the manual annotations. The calculated ICCs between automatically derived and manually annotated results demonstrated overall high consistency across the three parameters: RV/LV: 0.76 (95% CI: 0.638–0.877); SA: 0.79 (95% CI: 0.602–0.897); PA/AA: 0.77 (95% CI: 0.672–0.885). This study developed a deep learning approach for automated and accurate quantification of right ventricular dysfunction in pulmonary embolism (PE), which was shown to be highly consistent with manually annotated results. This provides an automated approach for quantifying right ventricular parameters in PE.