AI-based prediction of pulmonary hypertension in COPD patients with cor pulmonale using clinical and CT features.

Background and objectives: This study aimed to characterize COPD with clinically defined cor pulmonale and to develop clinical and contrast-enhanced CT-based AI-assisted models for its identification. Methods: We retrospectively enrolled 179 patients with COPD (90 COPD alone and 89 COPD with cor pulmonale). Clinical, laboratory, pulmonary function, electrocardiographic, and echocardiographic data were compared. To reduce incorporation bias, right ventricular, right atrial, and pulmonary artery measurements and electrocardiographic variables used in the operational case definition were excluded from candidate predictors. Variables associated with cor pulmonale in univariable analysis were entered simultaneously into multivariable logistic regression. A conservative sensitivity analysis additionally excluded mMRC score because dyspnea contributed to clinical case ascertainment. In 63 patients with diagnostic-quality CT angiography, pulmonary artery volumes were quantified using a U-Net model. Results: In the revised multivariable clinical model ( Conclusion: A revised clinical model that excluded variables incorporated into the diagnostic definition retained good discrimination for clinically defined cor pulmonale. AI-assisted quantification of small pulmonary vessels may provide complementary non-invasive information.