Artificial intelligence in detecting and quantifying thrombotic burden compared to manual analysis using CTPA

Abstract

<bold>Aim:</bold> Evaluate the performance of artificial intelligence (AI) in quantifying chronic thromboembolic pulmonary hypertension disease (CTEPH) in comparison to manual scoring, and its prognostic value. <bold>Methods:</bold> A deep learning model for thromboembolic disease segmentation was developed using ASPIRE Registry data. A clinical test set of 179 CTEPH patients (77 pulmonary endarterectomy and 102 medical management) was used to assess the clinical utility. AI-derived thromboembolic severity was evaluated for survival impact. A subcohort of 120 patients, AI thromboembolic volume was compared to manual scoring for accuracy and reliability. <bold>Results:</bold> AI-derived total thromboembolic disease volume showed a significant correlation with pulmonary vascular resistance (PVR) (r=0.29,p=0.002). Survival analysis demonstrated that AI-derived thromboembolic volume was associated with mortality in all patients (central: p=0.01; total clot volume: p=0.03), and in those who did not undergo endarterectomy (central: p<0.001; total clot volume: p=0.01). In the endarterectomy group, thromboembolic disease severity was not significantly associated with survival. Following adjustment for demographic and clinical variables, central thromboembolism remained an independent predictor of outcomes in non-endarterectomy patients (p<0.001). In 120 patients, AI-derived thromboembolic volume Correlated significantly with manual scoring (r=0.72, p<0.001). <bold>conclusion:</bold> AI demonstrates comparable accuracy to manual analysis in assessing the extent of chronic thromboembolic disease. AI-derived thromboembolism burden may play a role in prognostication outcomes in patients with CTEPH. Further validation in external cohorts is required.

Ähnliche Arbeiten