Automatic detection of selective arterial devices for advanced visualization during abdominal aortic aneurysm endovascular repair

摘要

Here we address the automatic segmentation of endovascular devices used in the endovascular repair (EVAR) of abdominal aortic aneurysms (AAA) that deform vascular tissues. Using this approach, the vascular structure is automatically reshaped solving the issue of misregistration observed on 2D/3D image fusion for EVAR guidance. The endovascular devices we considered are the graduated pigtail catheter (PC) used for contrast injection and the stent-graft delivery device (DD). The segmentation of the DD was enhanced using an asymmetric Frangi filter. The segmented geometries were then analysed using their specific features to remove artefacts. The radiopaque markers of the PC were enhanced using a fusion of Hessian and newly introduced gradient norm shift filters. Extensive experiments were performed using a database of images taken during 28 AAA-EVAR interventions. This dataset was divided into two parts: the first half was used to optimize parameters and the second to compile performances using optimal values obtained. The radiopaque markers of the PC were detected with a sensitivity of 88.3% and a positive predictive value (PPV) of 96%. The PC can therefore be positioned with a majority of its markers localized while the artefacts were all located inside the vessel lumen. The major parts of the DD, the dilatator tip and the pusher surfaces, were detected accurately with a sensitivity of 85.9% and a PPV of 88.7%. The less visible part of the DD, the stent enclosed within the sheath, was segmented with a sensitivity of 63.4% because the radiopacity of this region is low and uneven. The centreline of the DD in this stent region was alternatively traced within a 0.74 mm mean error. The automatic segmentation of endovascular devices during EVAR is feasible and accurate; it could be useful to perform elastic registration of the vascular lumen during endovascular repair. (C) 2015 IPEM. Published by Elsevier Ltd. All rights reserved.