Artificial intelligence in pulmonary hypertension: a systematic review

Area of Science:

• Medical Informatics
• Cardiology
• Biomedical Engineering

Background:

• Pulmonary hypertension (PH) involves elevated pulmonary pressures and right ventricular strain.
• Pulmonary arterial hypertension (PAH) has a poor prognosis, particularly with delayed diagnosis.
• Artificial intelligence (AI), machine learning (ML), and deep learning (DL) offer potential for non-invasive PH/PAH prediction and risk stratification.

Purpose of the Study:

• Systematically review ML and DL applications for non-invasive diagnosis, classification, and prognostication in PH and PAH.
• Evaluate the methodological quality and clinical applicability of these AI models.
• Synthesize study characteristics and heterogeneity using the SWiM framework.

Main Methods:

• PRISMA-guided systematic search for studies using ML/DL on non-invasive clinical, imaging, or biomarker data.
• Synthesis of study characteristics and heterogeneity via the SWiM framework.
• Risk of bias assessment using PROBAST+AI, focusing on participant selection, predictors, outcomes, and analysis.

Main Results:

• Fifty-three studies were included, primarily using clinical, echocardiographic, imaging, or molecular data.
• AI models demonstrated high predictive performance (AUC 0.71–1.00), with DL, especially CNNs, increasingly applied.
• Most studies were retrospective, single-center, with limited external validation and heterogeneity in design and outcomes; 83% had low risk of bias but common applicability concerns.

Conclusions:

• ML and DL models show significant promise for PH and PAH diagnosis and prognosis.
• Limitations include subclass differentiation, methodological transparency, and the need for robust external validation.
• Future research should focus on addressing these limitations to enhance clinical applicability.