Published in:
01-08-2015 | Original Paper
CT evaluation prior to transapical aortic valve replacement: semi-automatic versus manual image segmentation
Authors:
Borek Foldyna, Camelia Jungert, Christian Luecke, Konstantin von Aspern, Sonja Boehmer-Lasthaus, Eva Maria Rueth, Matthias Grothoff, Stefan Nitzsche, Matthias Gutberlet, Friedrich Wilhelm Mohr, Lukas Lehmkuhl
Published in:
The International Journal of Cardiovascular Imaging
|
Issue 6/2015
Login to get access
Abstract
To compare the performance of semi-automatic versus manual segmentation for ECG-triggered cardiovascular computed tomography (CT) examinations prior to transcatheter aortic valve replacement (TAVR), with focus on the speed and precision of experienced versus inexperienced observers. The preoperative ECG-triggered CT data of 30 consecutive patients who were scheduled for TAVR were included. All datasets were separately evaluated by two radiologists with 1 and 5 years of experience (novice and expert, respectively) in cardiovascular CT using an evaluation software program with or without a semi-automatic TAVR workflow. The time expended for data loading and all segmentation steps required for the implantation planning were assessed. Inter-software as well as inter-observer reliability analysis was performed. The CT datasets were successfully evaluated, with mean duration between 520.4 ± 117.6 s and 693.2 ± 159.5 s. The three most time-consuming steps were the 3D volume rendering, the measurement of aorta diameter and the sizing of the aortic annulus. Using semi-automatic segmentation, a novice could evaluate CT data approximately 12.3 % faster than with manual segmentation, and an expert could evaluate CT data approximately 10.3 % faster [mean differences of 85.4 ± 83.8 s (p < 0.001) and 59.8 ± 101 s (p < 0.001), respectively]. The inter-software reliability for a novice was slightly lower than for an expert; however, the reliability for a novice and expert was excellent (ICC 0.92, 95 % CI 0.75–0.97/ICC 0.96, 95 % CI 0.91–0.98). Automatic aortic annulus detection failed in two patients (6.7 %). The study revealed excellent inter-software and inter-observer reliability, with a mean ICC of 0.95. TAVR evaluation can be accomplished significantly faster with semi-automatic rather than with manual segmentation, with comparable exactness, showing a benefit for experienced and inexperienced observers.