Advancements toward clinical application of AI-assisted CT analysis in pulmonary hypertension: a systematic literature review.
Abstract
Background
Computed tomography (CT) plays an important role in pulmonary hypertension (PH) evaluation but remains largely restricted to qualitative or semi-quantitative imaging features. Artificial intelligence (AI) has potential to enable automated extraction of quantitative CT data, providing additional PH assessment tools. This systematic review evaluated current evidence on clinically actionable AI-enabled CT analysis in PH.
Methods
Systematic literature search of PubMed, EMBASE, Web of Science, and Cochrane Library was performed from database inception to 5th January 2026. Studies were eligible if they applied AI methodologies to CT imaging in adult PH populations for diagnostic detection, aetiological or endophenotype classification, prognostication or risk stratification, or treatment assessment. Risk of bias and applicability were evaluated using the PROBAST tool.
Results
17 studies met the inclusion criteria. AI-assisted CT analysis showed potential for PH detection, with area under the curves of 0.66-0.99, derived from quantitative pulmonary vascular morphology, cardiac chamber dimensions, lung perfusion, and/or parenchymal abnormalities. Several studies demonstrated ability to differentiate CTEPH from acute pulmonary embolism using perfusion and parenchymal features, and to identify pulmonary arterial hypertension endophenotypes through fibrosis quantification. Additionally, CT-based vascular and parenchymal metrics showed potential utility for prognostic risk stratification and evaluation of response to CTEPH interventions. However, most studies involved small, retrospective, or highly selected cohorts, with limited external validation, and methodological heterogeneity precluding meaningful comparability.
Conclusion
AI-assisted CT analysis shows potential for advancing PH assessment through automated quantitative biomarkers. Robust validation, methodological comparability, and clinically intuitive metrics are required before potential clinical integration.