297. Radiomic CT features of pancreatic neuroendocrine tumors are robust against inter-observer delineation uncertainty

Published on Dec 1, 2018in Physica Medica2.485
· DOI :10.1016/J.EJMP.2018.04.306
Carla Sini9
Estimated H-index: 9
M. Mori5
Estimated H-index: 5
+ 6 AuthorsC. Fiorino7
Estimated H-index: 7
Purpose High accuracy in delineating pancreatic neuroendocrine tumors (NET) is a pre-requisite for the assessment of CT-based biomarkers, including radiomic features (RF). The aim of this study was to quantify inter-observer variability in delineating NET on CT images and its impact on RF. Methods Three radiologists contoured volumes of 32 NET patients. The contours were delineated using arterial or venous phase. The contours delineated on contrast enhanced images were overlayed onto CT images acquired before contrast injection, and possibly corrected to take into account small anatomical discrepancies. Inter-observer contouring variability was assessed by DICE index and volume agreement; paired analysis between observers was performed. Seventy-one RF (12 of first and 59 of higher order) were extracted for all contours, for each observer by using the CGITA software (v 1.4). Their robustness was investigated against contour variability by Spearman R values and intra-class correlation coefficients (ICC, taking the worst value of the tested pairs). RF were classified as optimal, good and modest according to their ICC: ⩾0.90, ⩾0.80 and Results The mean/median values of NET volumes were 14.7/1.3 cc. Despite the prevalence of small tumors, a satisfactory agreement was found (mean DICE = 0.78; SD 0.09) with no significant differences between observers. Sixty-three RF were classified as optimal and only 1 showed ICC   0.9 and 0.82 for the remaining ones. Examples of correlation for four different features (including asphericity) are shown in the Figure. Conclusions Results show a relatively small inter-observer variability in delineating pancreatic NET with little impact on RF. This finding suggests the possibility to robustly extract RF features in large pancreatic NET populations to assess CT-based radiomic biomarkers.
📖 Papers frequently viewed together
13 Authors (M. Mori, ..., Francesco De Cobelli)
11 Citations
12 Authors (Sara Broggi, ..., C. Fiorino)
Cited By0