Machine learning in CTEPH: predicting the efficacy of BPA based on clinical and echocardiographic features
Abstract Background This study aims to develop a machine learning (ML)-based predictive model for evaluating the efficacy of percutaneous pulmonary balloon angioplasty (BPA) in patients with chronic thromboembolic pulmonary hypertension (CTEPH) by integrating clinical and echocardiographic parameter...
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
BMC
2025-08-01
|
| Series: | BMC Medical Imaging |
| Subjects: | |
| Online Access: | https://doi.org/10.1186/s12880-025-01870-3 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Background This study aims to develop a machine learning (ML)-based predictive model for evaluating the efficacy of percutaneous pulmonary balloon angioplasty (BPA) in patients with chronic thromboembolic pulmonary hypertension (CTEPH) by integrating clinical and echocardiographic parameters. By comparing the predictive performance of different algorithms, we aimed to establish a robust tool to identify patients most likely to benefit from BPA. Methods We retrospectively included 135 inoperable CTEPH patients who underwent BPA between January 2017 and September 2024. Clinical and echocardiographic data prior to the first BPA procedure were collected. The cohort was temporally split into a training set (2017–2021) and a test set (2022–2024). Key variables were identified using univariate logistic regression followed by Least Absolute Shrinkage and Selection Operator (LASSO) feature selection. Five ML models were trained to predict BPA response, defined as either a mean pulmonary artery pressure (mPAP) ≤ 30 mmHg or a ≥ 30% reduction in pulmonary vascular resistance (PVR) from baseline. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, F1 score, and Brier score. SHapley Additive exPlanations (SHAP) values were applied to interpret feature importance of the predictive model. Results A total of 135 patients were included to construct models. 6 features were selected from 49 variables for model training. The Logistic Regression with L2 regularisation method demonstrated the most optimal predictive efficacy, as evidenced by the highest AUC of 0.865 (95% CI: 0.710–0.985), with an accuracy of 0.848, sensitivity of 0.950, specificity of 0.692, an F1 score of 0.884, and a Brier score of 0.162 in the test set. According to SHAP, the most influential predictors for BPA response in patients with CTEPH were the proportion of occlusive lesions, tricuspid annular plane systolic excursion to pulmonary artery systolic pressure ratio(TAPSE/PASP), six-minute walk distance (6MWD), right ventricular end-systolic area (RVESA), the severity of tricuspid regurgitation (TR) and PVR. Conclusions By integrating clinical characteristics and echocardiographic parameters, an ML-based BPA efficacy prediction model was developed. Among all of the models, the Logistic Regression with L2 regularisation model exhibited the best overall performance. Graphical Abstract |
|---|---|
| ISSN: | 1471-2342 |