Interobserver variation in cervical cancer tumor delineation for image‐based radiotherapy planning among and within different specialties

Radiation therapy for cervical cancer involves a team of specialists, including diagnostic radiologists (DRs), radiation oncologists (ROs), and medical physicists (MPs), to optimize imaging-based radiation therapy planning. The purpose of the study was to investigate the interobserver variations in tumor delineation on MR images of cervical cancer within the same and among different specialties. Twenty MRI cervical cancer studies were independently reviewed by two DRs, two ROs, and two MPs. For every study, each specialist contoured the tumor regions of interest (ROIs) on T2-weighted Turbo Spin Echo sagittal images on all slices containing tumor, and the total tumor volume was computed for statistical analysis. Analysis of variance (ANOVA) was used to compare the differences in tumor volume delineation among the observers. A graph of all tumor-delineated volumes was generated, and differences between the maximum and minimum volumes over all the readers for each patient dataset were computed. Challenges during the evaluation process for tumor delineation were recorded for each specialist. Interobserver variations of delineated tumor volumes were significant (p < 0.01) among all observers based on a repeated measures ANOVA, which produced an F(5,95) = 3.55. The median difference between the maximum delineated volume and minimum delineated volume was 33.5 cm3 (which can be approximated by a sphere of 4.0 cm diameter) across all 20 patients. Challenges noted for tumor delineation included the following: (1) partial voluming by parametrial fat at the periphery of the uterus; (2) extension of the tumor into parametrial space; (3) similar signal intensity of structures proximal to the tumor such as ovaries, muscles, bladder wall, bowel loops, and pubic symphysis; (4) postradiation changes such as heterogeneity and necrosis; (5) susceptibility artifacts from bowels and vaginal tampons; (6) presence of other pathologies such as atypical myoma; (7) factors that affect pelvic anatomy, including the degree of bladder distension, bowel interposition, uterine malposition, retroversion, and descensus. Our limited study indicates significant interobserver variation in tumor delineation. Despite rapid progress in technology, which has improved the resolution and precision of image acquisition and the delivery of radiotherapy to the millimeter level, such "human" variations (at the centimeter level) may overshadow the gain from technical advancement and impact treatment planning. Strategies of standardization and training in tumor delineation need to be developed.