Manual measurement of aortoiliac angulation may be excessively variable (6
). Clinically significant variability has also been reported in manual measurements of angulation using CT in other structures such as in extremities (8
) and pulmonary veins (9
). The interobserver variability of manual measurements in this paper is slightly less than that reported by Diehm et. al. (~6° vs ~13°, 20% vs 32%) (10
). This could be due to differences in patient population, but also could be because our readers were given explicit written and pictorial instructions on how measurements were to be performed. Our automated approach is faster, yields measurements that are less variable and as accurate as manual methods, and the absolute amount of variability is very small. This may be partly because the median centerline derived from the user-defined points has been shown to be relatively insensitive to positional variability in the user-defined points (15
). If there is complete occlusion of the aorta or common iliac arteries, the interface allows the user to bridge the occluded segment manually, allowing the calculation of a branched median centerline. Our algorithm does not require hardware acceleration and is operating system independent, and so can potentially be implemented as an add-on to commercial CT scanner consoles or postprocessing workstations.
One limitation of the study is that we used a relatively small set of patients, but this set was enough to achieve statistical significance. Our algorithm still requires some user input, but the time required for this is small. We have not tested the algorithm in aortic aneurysms that have concomitant aortic dissections, where there is the potential of interference with the algorithm's ability to delineate the aortic lumen. However, this can also be addressed with manual input of centerline points. A common problem in evaluating algorithms of this type is that there is seldom a reliable reference standard, hindering the determination of accuracy. Thus, many studies report only variability (i.e., precision) and compare it to that derived using some other (e.g., manual) established method, as we have also done. However, we also evaluated accuracy using a novel approach – mathematical manipulation of clinically derived aneurysm shape in which true angulation could be known and varied. While perhaps not completely representative of what might be obtained if true angulations were known over a wide range of CTA data from patients, our approach, which generated realistic aortic morphology over just such a range of angulations, showed that the algorithm had a very good accuracy (<=1 degree) in this setting. This, when combined with the low variability between the automated and manual methods reported above, provides reasonable assurance of clinical applicability even in the absence of a solid reference standard for accuracy.
Our automated algorithm could enhance the clinical utility and reliability of CTA for preoperative assessment for EVAR. Our preliminary results support continued development and evaluation in a larger set of patients.