Chest Computed Tomography-Based Radiomics for the Diagnosis and Prognosis of Pulmonary Hypertension.

Summary

A multicenter chest CT radiomics model integrating pulmonary vascular features with clinical data achieved AUC ~0.98 for PH diagnosis and outperformed the ESC 4-strata risk tool for 2‑year prognosis (AUC 0.866 vs 0.709). External validation supported generalizability despite smaller cohorts.

Key Findings

• Diagnostic radiomic-clinical model achieved AUC 0.984 (derivation) and 0.980 (external validation) for PH detection.
• Prognostic model’s 2-year AUC (0.866) exceeded ESC 4-strata risk assessment (0.709).
• Exploratory pulmonary arterial hypertension subtyping reached internal AUC 0.898 and external AUC 0.877.

Clinical Implications

Radiomics could complement echocardiography and right heart catheterization by flagging high-risk PH patients from standard CT scans, guiding earlier referral and more tailored therapy.

Limitations

• External validation sample sizes were modest, potentially inflating uncertainty bounds
• Radiomics reproducibility depends on CT acquisition/segmentation standardization

Future Directions

Harmonize CT protocols and validate across broader scanners and populations; integrate with hemodynamics for hybrid risk scores; assess impact on clinical decision-making and outcomes.