Abstract
The purpose of this study was to assess variability in contouring the gross tumor volume (GTV) and clinical target volume (CTV) of 3 clinical cervix cancer cases by a cohort of international experts in the field in preparation for the development of an online teaching atlas.
Twelve international experts participated. Three clinical scenarios: node positivity (PLN), retroverted uterus (RV), and parametrial invasion (PI) were used. Sagittal and axial magnetic resonance images of the clinical cases were downloaded to participants’ treatment planning systems for contouring. The GTV/cervix/uterus/parametria/vagina and nodal CTV were contoured. Contour consensus was assessed for sensitivity/specificity using an expectation maximization algorithm called Simultaneous Truth and Performance Level Estimation and experts’ overall agreement was summarized by kappa statistics.
Agreement for GTV in the 3 clinical cases was high (Simultaneous Truth and Performance Level Estimation sensitivity, 0.54-0.92; specificity, 0.97-0.98; and kappa measure for PLN, RV, and PI was 0.86, 0.76, and 0.42; P < .0001). Moderate to substantial agreement was seen for nodal CTV (kappa statistics for PLN, RV, and PI was 0.65, 0.58, and 0.62; P < .0001), uterus (kappa for PLN, RV, and PI was 0.45, 0.74, and 0.77; P < .0001), and parametria (kappa for PLN, RV, and PI was 0.49, 0.62, and 0.50; P < .0001). Contouring heterogeneity was greatest for the cervix (kappa measure for PLN, RV, and PI was 0.15, 0.4, and 0.24; P < .0001) and vagina (kappa for PLN, RV, and PI was 0.47, 0.36 and 0.46; P < .0001), reflecting difficulties in determining the interface between GTV and these tissues.
Kappa statistics of the different CTV components generally demonstrated moderate to substantial agreement among international experts in the field of gynecological radiation therapy. Further planning target volume margins accounting for organ motion and setup errors are a necessary addition to the CTV.
