Machine learning-based hemodynamics quantitative assessment of pulmonary circulation using computed tomographic pulmonary angiography.
Authors
Affiliations (8)
Affiliations (8)
- Department of Radiology, Beijing Anzhen Hospital, Capital Medical University, No. 2 Anzhen Rd, Chaoyang District, Beijing, China.
- School of Automation, Beijing Institute of Technology, No. 5 Zhongguancun South Street, Haidian District, Beijing, China.
- Bioengineering Department and Imperial-X, Imperial College London, London W12 7SL, UK; National Heart and Lung Institute, Imperial College London, London SW7 2AZ, UK; Cardiovascular Research Centre, Royal Brompton Hospital, London SW3 6NP, UK; School of Biomedical Engineering & Imaging Sciences, King's College London, London WC2R 2LS, UK.
- Department of Cardiology, Gansu Provincial Hospital, Lanzhou, Gansu, China; Heart, Lung and Vessels Center, Sichuan Provincial People's Hospital, University of Electronic Science and Technology of China, Chengdu, Sichuan, China.
- E204, Wu Yee Sun college, Chinese University of Hong Kong, Shatin, NT, HK.
- Discipline of Medical Radiation Science, Curtin Medical School, Curtin University, Perth, Australia.
- School of Automation, Beijing Institute of Technology, No. 5 Zhongguancun South Street, Haidian District, Beijing, China. Electronic address: [email protected].
- Department of Radiology, Beijing Anzhen Hospital, Capital Medical University, No. 2 Anzhen Rd, Chaoyang District, Beijing, China. Electronic address: [email protected].
Abstract
Pulmonary hypertension (pH) is a malignant pulmonary circulation disease. Right heart catheterization (RHC) is the gold standard procedure for quantitative evaluation of pulmonary hemodynamics. Accurate and noninvasive quantitative evaluation of pulmonary hemodynamics is challenging due to the limitations of currently available assessment methods. Patients who underwent computed tomographic pulmonary angiography (CTPA) and RHC examinations within 2 weeks were included. The dataset was randomly divided into a training set and a test set at an 8:2 ratio. A radiomic feature model and another two-dimensional (2D) feature model aimed to quantitatively evaluate of pulmonary hemodynamics were constructed. The performance of models was determined by calculating the mean squared error, the intraclass correlation coefficient (ICC) and the area under the precision-recall curve (AUC-PR) and performing Bland-Altman analyses. 345 patients: 271 patients with PH (mean age 50 ± 17 years, 93 men) and 74 without PH (mean age 55 ± 16 years, 26 men) were identified. The predictive results of pulmonary hemodynamics of radiomic feature model integrating 5 2D features and other 30 radiomic features were consistent with the results from RHC, and outperformed another 2D feature model. The radiomic feature model exhibited moderate to good reproducibility to predict pulmonary hemodynamic parameters (ICC reached 0.87). In addition, pH can be accurately identified based on a classification model (AUC-PR =0.99). This study provides a noninvasive method for comprehensively and quantitatively evaluating pulmonary hemodynamics using CTPA images, which has the potential to serve as an alternative to RHC, pending further validation.