Optimal Virtual-target Definition for Detecting Feeding Arteries of Renal Cell Carcinoma Using Automated Feeder-detection Software

Purpose: To determine the optimal virtual-target definition for detecting renal cell carcinoma feeders using transarterial computed tomography angiography with automated feeder-detection software.
Material and Methods: This retrospective study included 17 patients with 17 renal cell carcinomas who underwent transarterial ethiodized-oil marking before cryoablation. Tumor feeders were automatically detected on transarterial renal computed tomography angiography images using the automated feeder-detection software with three virtual-target definitions: small (ellipsoidal area maximized within the tumor contour), medium (ellipsoidal area covering the entire tumor with a minimal peripheral margin), and large (ellipsoidal area including the tumor and a 5-mm peripheral margin). The detected feeders were classified as true or false positives according to the findings of selective renal arteriography, by consensus of two interventional radiologists. Feeder-detection sensitivity and the mean number of false-positive feeders per tumor were calculated for each virtual-target definition.
Results: For 17 tumors, 25 feeding arteries were identified on the arteriography. The feeder-detection sensitivity of the software was 80.0% (20/25), 88.0% (22/25), and 48.0% (12/25) for small, medium, and large virtual targets, respectively. The mean ± standard deviation number of false-positive feeders per tumor was 0.82 ± 1.3, 1.41 ± 1.1, and 2.82 ± 1.6 when using small, medium, and large virtual-target definitions, respectively.
Conclusions: The detection rate of renal cell carcinoma feeders with the automated feeder-detection software varies according to the virtual-target definition. Using a medium virtual target, covering the entire tumor with a minimal peripheral margin, may provide the highest sensitivity and an acceptable number of false-positive feeders.