Echocardiographic artificial intelligence for pulmonary hypertension classification.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate classification of pulmonary hypertension (PH) is essential for guiding treatment strategies.
• Current guidelines recommend integrating multiple echocardiographic parameters for PH assessment.
• Machine learning (ML) offers a potential avenue for improving PH prediction accuracy.

Purpose of the Study:

• To investigate the efficacy of ML algorithms in predicting PH using echocardiographic data.
• To compare the predictive accuracy of an ML model against current guideline-based echocardiographic assessments.

Main Methods:

• Echocardiographic and physical data from 885 patients were analyzed.
• Patients were categorized into non-PH, precapillary PH, and postcapillary PH groups based on right heart catheterisation (RHC) data.
• A logistic regression model with elastic net regularization was developed and validated.

Main Results:

• The ML model achieved high areas under the curve (0.789 for normal, 0.766 for precapillary PH, 0.742 for postcapillary PH).
• The ML model showed significantly better predictive accuracy than guideline-based assessment in the derivation cohort (59.4% vs 51.6%).
• In the validation cohort, ML model accuracy was comparable to guideline-based assessment (59.4% vs 57.8%).

Conclusions:

• The developed ML model shows promising potential for predicting echocardiographic PH.
• Further research is required to validate the clinical utility of ML in PH diagnosis and treatment decisions.