The effect of interobserver variability on transrectal ultrasonography-based postimplant dosimetry

To investigate interobserver variability in contouring the prostate on postimplant transrectal ultrasonography (TRUS) images and its effect on dosimetric parameters that quantify implant quality.

Twenty preplanned peripherally loaded prostate implants were performed using 125I seeds and spacers linked together in linear arrays that maintain precise seed spacing and prevent seeds from rotating about their longitudinal axis. A set of two-dimensional transverse images spaced at 0.50-cm intervals was obtained with a high-resolution TRUS probe at the conclusion of the procedure with the patient still under anesthesia. A high percentage of the seeds (> 85%) were localized based on their visible echoes. The remaining seeds were identified based on the known locations of the "missing" seeds in the arrays. Two experienced ultrasonographers and a prostate brachytherapist independently contoured the prostate on the postimplant TRUS images. The prostate volumes defined by each observer were used to calculate the minimal dose received by 90% of the prostate volume (D90) and the percentage of the prostate volume receiving 100% of the prescribed minimal peripheral dose (V100). The observers also contoured the prostate on six preimplant TRUS studies to compare the variability in defining the prostate on pre- and postimplant TRUS images.

The mean postimplant prostate volumes ranged from 20.8 to 66.9 cm3 (median: 45.7 cm3). The standard deviations (SDs), which reflect the variation in the volumes of the three observers, ranged from 1.4% to 26.1% of the mean (median: 11%). Multiple pairwise comparisons showed that the prostate volumes delineated by observer 3 differed significantly from those of observers 1 and 2 (p < 0.003). The volumes of observers 1 and 2 were not significantly different (p > 0.5). The mean values of D90 ranged from 124.2 to 171.1 Gy (median: 154.7 Gy) having SDs that ranged from 0.6% to 24.4% of the mean D90 (median: 7.8%). The mean values of V100 ranged from 82.3% to 95.1% (median: 92.8%) having SDs that ranged from 0.4% to 11.2% of the mean V100 (median: 4.0%). The values of both D90 and V100 calculated from the volumes of observer 3 were significantly (p < 0.003) different from those of observers 1 and 2, which did not differ significantly (p > 0.5). There was less interobserver variability in contouring the preimplant TRUS volumes. The mean volumes ranged from 20.3 to 54.3 cm3 having SDs that ranged from 1.9% to 14.1% (median: 8.6%).

Significant interobserver differences in delineating the prostate volume on postimplant TRUS images were observed; however, these differences were less than generally reported for postimplant CT images. The interobserver differences in contouring the prostate in both TRUS and CT images produced significant differences in the dosimetric parameters, D90 and V100.