Semi-automated cerebral aneurysm segmentation and geometric analysis for WEB sizing utilizing a cloud-based computational platform

Background Accurate aneurysm measurements are important for selecting the WEB device. The objective was to validate a cloud-based platform, SurgicalPreview (SP) against manual measurements for aneurysm analysis. Methods Two sets of measurements each for SP and manual methods were obtained for 40 aneurysms. Reliability and agreement were assessed with intra-class correlation coefficient (ICC) and Bland-Altman plots respectively. Kappa coefficient was used to assess agreement for predicting WEB size. Results There was good reliability for repeat SP measurements: aneurysm diameter (ICC-1, 95%CI 0.98-1), height (ICC-1, 95%CI 0.99-1) and neck diameter (ICC-0.96, 95%CI 0.93-0.98). There was good reliability for the two manual diameter (ICC-0.97, 95%CI 0.9-0.97) and height (ICC-0.93, 95%CI 0.87-0.96) measurements and moderate for neck diameter (ICC-0.76, 95%CI 0.54-0.87). There was greater agreement for SP versus manual repeat measurements on Bland-Altman plots. Reliability between the SP and manual methods was good for aneurysm diameter (ICC-0.98, 95%CI 0.95-1) and height (ICC-0.96, 95%CI-0.93-0.98) and moderate for neck. (ICC-0.6, 95%CI -0.22-0.87). The Bland-Altman plots confirmed better agreement between the two methods for the aneurysm diameter and height than the neck. There was strong agreement between the methods for predicting the WEB diameter (Kappa-0.84, 95%CI 0.71-0.97) and moderate for predicting WEB height (Kappa-0.66, 95%CI 0.43-0.89). There was moderate agreement for predicted versus deployed WEB diameter: SP (Kappa-0.56, 95%CI 0.38-0.74), Manual (Kappa-0.53, 95%CI 0.34-0.71). Conclusion The SurgicalPreview (R) had greater agreement for repeat measurements. There was good reliability between the two methods for predicting WEB diameter and height and moderate agreement between predicted versus deployed WEB diameter.