A Real-World Clinical Study on Computed Tomography Pulmonary Angiography-Based Artificial Intelligence-Driven Risk Stratification in Pulmonary Embolism

PURPOSE: To evaluate the efficacy of a computed tomography pulmonary angiography (CTPA)-based artificial intelligence system for use in the clinical setting to stratify the risk of pulmonary embolism (PE). MATERIALS AND METHODS: This study enrolled 115 patients diagnosed with PE based on positive CTPA reports at our institution between January 1 and October 10, 2023. Cases were classified as relatively high-risk or low-risk according to the 2019 ESC Guidelines. Two radiologists independently and blindly measured risk-related imaging parameters, including main pulmonary artery diameter (MPAD), ascending aortic diameter (AAD), right and left ventricular diameters (RVD, LVD), the right-to-left ventricular diameter ratio (RVD/LVD ratio), pulmonary artery obstruction index (PAOI), and pulmonary embolism volume (PEV). These results were compared with those from the automated artificial intelligence (AI) system to assess consistency. RESULTS: The agreement (intraclass correlation coefficients, ICCs) of MPAD, AAD, RVD, PAOI, LVD, and RVD/LVD ratio between AI system and manual measurements were 0.94, 0.79, 0.78, 0.80, 0.54, and 0.48, respectively. Among the AI-extracted covariates, an RVD/LVD ratio >1.2 showed good performance in identifying relatively high-risk PE cases, with an area under the curve (AUC) of 0.771. The AI model incorporating all covariates achieved an AUC of 0.848 for classifying high-risk PE. CONCLUSIONS: The CTPA-based AI system can rapidly extract the key imaging parameters relevant to PE risk stratification in real-world clinical settings, significantly reducing the CTPA interpretation time and improving evaluation efficiency.