Does computer-assisted detection of pulmonary emboli enhance severity assessment and risk stratification in acute pulmonary embolism?

Aim: To prospectively assess the value of computer-aided detection (CAD) for the computed tomography (CT) severity assessment of acute pulmonary embolism (PE). Materials and methods: CT angiographic scans of 58 PE-positive patients (34-89 years, mean 66 years) were analysed by four observers for PE severity using the Mastora index, and by CAD. Patients were stratified to three PE risk groups and results compared to an independent reference standard. Interobserver agreement was tested by Bland and Altman and extended kappa (Ke) statistics. Mastora index changes after CAD data review were tested by Wilcoxon signed ranks. Results: CAD detected 343 out of 1118 emboli within given arterial segments and a total of 155 out of 218 polysegmental emboli (segmental vessel-based sensitivity = 30.7%, embolus-based sensitivity = 71.2% false-positive rate = 4.1/scan). Interobserver agreement on PE severity [95% limits of agreement (LOA) = -19.7-7.5% and-5.5-3% for reader pairs 1 versus 2 and 3 versus 4, respectively was enhanced by consensus with CAD data (LOA = -6.5-5.4% and-3.7-2% for reader pairs 1 versus 2 and 3 versus 4, respectively). Simultaneously, the percentual scoring errors (PSE) were significantly decreased (PSE = 35.4 ± 31.8% and 5.1 ± 8.9% for readers1/2 and 2/3, respectively, and PSE = 27.6 ± 31% and 3.8 ± 6.2%, respectively, after CAD consensus; p ≤ 0.005). Misclassifications to PE risk groups occurred in 27.6, 24.1, 5.2, and 5.2% of patients for readers 1-4, respectively, (Ke = 0.74) and were corrected by CAD consensus in 56.3, 36, 33.3, and 33.3% of misclassified patients, respectively (Ke = 0.83; p < 0.05). Conclusion: Radiologists may benefit from consensus with CAD data that improve PE severity scores and stratification to PE risk groups.