A radiomics nomogram based on CT pulmonary angiographic data for predicting adverse outcomes in non-high-risk acute pulmonary embolism patients

Objective: To develop a radiomics nomogram with cardiovascular data from computed tomography pulmonary angiography (CTPA) to predict adverse outcomes in non-high-risk acute pulmonary embolism (APE) patients. Methods: We retrospectively collected CTPA data from non-high-risk APE patients in first hospital of China medical university (CMU) and Shengjing hospital of CMU between April 2013 and April 2017. We oriented the cardiovascular data by three-dimensional CTPA reconstruction. An interventricular septum curvature (ISC) that leaned toward the left ventricle (LV) or showed straightness indicated deviation (+). A prediction model for adverse outcomes was built by using backward logistic regression from training set (data from Shengjing Hospital) and identified predictors were presented on a radiomics nomogram. The prediction model was evaluated in a test set (data from First Hospital of CMU). Results: 170 patients were enrolled in the training set, the logistics analysis revealed that the RVD4-CH/LVD4-CH ratio and ISC deviation (+) were correlated with adverse outcomes (OR= 7.878 and OR= 44.995, respectively). The AUC values of the logistic model were 0.87 in the training set and 0.784 in the test set, respectively, which showed good discrimination and calibration ability. In the test set (115 cases), 10 fold cross-validation showed sensitivity of 50%, specificity of 96.8%, positive predictive value of 76.9%, negative predictive value of 90.2%, and accuracy of 88.7%. Conclusions: In non-high-risk APE patients, the presented nomogram built with RVD4-CH/LVD4-CH ratio and ISC deviation (+) may facilitate prediction of adverse outcomes.