Contouring Variability Of Human And Deformable Generated Contours On Planning And Cone Beam Ct Datasets In Radiation Therapy For Prostate Cancer

To quantify the interobserver contouring variation on planning CT (PCT), CBCT (CBCT_Human), and contours on CBCT generated using deformable image registration (DIR) (CBCT_Def). Ten patients were studied in this IRB-approved retrospective analysis. For each patient, 1 PCT and 5 CBCT image sets (spaced evenly throughout the treatment course from start to finish) were used. Five radiation oncologists contoured the target (prostate) and organ-at-risk (OAR: bladder and rectum) on each image set. A consensus contour was generated for contour analysis through combining all observer contours using the STAPLE method in CERR. Each contour was compared to consensus according to the following metrics: Dice coefficient, contour distance, center of mass deviation, and contour volume. The contour distance metric measured the average magnitude of deviation between observer and consensus contour set for the overall contour and five prostate contour regions: superior, inferior, anterior, posterior, and lateral. The day 1 CBCT image was registered to the remaining 4 CBCT fractions using DIR to generate additional contours for analysis. These deformable-generated contours were compared to the same consensus contour as mentioned above. The average [st dev] contour metrics for prostate contours on PCT and CBCT were as follows: Dice– 0.89 [0.05] (PCT), 0.87 [0.06] (CBCT_Human), 0.82 [0.06] (CBCT_Def); COM Deviation – 2.0 mm [0.5] (PCT), 2.8 mm [0.5] (CBCT_Human), 3.5 mm [0.7] (CBCT_Def); contour volume – 57.7 cm3 [8.7 cm3] (PCT), 58.6 [9.2] (CBCT_Human), 60.7 [9.1] (CBCT_Def). The average contour distance [st dev] for the individual contour regions was as follows (all values in mm):ePoster Abstracts 1088; Table 1Contour RegionPCTCBCT_HumanCBCT_DefSuperior [st. dev]1.4 [1.2]1.8 [1.3]2.4 [0.9]Inferior2.2 [1.8]2.4 [1.9]3.0 [1.6]Anterior1.3 [0.8]1.4 [0.8]2.1 [0.7]Posterior0.9 [0.4]1.0 [0.5]1.9 [0.5]Lateral1.1 [0.5]1.2 [0.6]1.6 [0.4]Contour - Overall1.4 [0.7]1.7 [0.7]2.30 [0.5]The average [st dev] Dice coefficient for bladder contours was: 0.96 [0.02] (PCT), 0.95 [0.03] (CBCT_Human), 0.83 [0.11] (CBCT_Def). The average [st dev] Dice coefficient for rectum contours was: 0.87 [0.06] (PCT), 0.86 [0.06] (CBCT_Human), 0.76 [0.07] (CBCT_Def). Open table in a new tab The average [st dev] Dice coefficient for bladder contours was: 0.96 [0.02] (PCT), 0.95 [0.03] (CBCT_Human), 0.83 [0.11] (CBCT_Def). The average [st dev] Dice coefficient for rectum contours was: 0.87 [0.06] (PCT), 0.86 [0.06] (CBCT_Human), 0.76 [0.07] (CBCT_Def). Prostate contouring variation on PCT and CBCT images is largest in superior and inferior regions. Human contour variation is similar between PCT and CBCT for the in-plane prostate contour regions (anterior, posterior, and lateral). The CBCT_Def contours exhibit an increase in contouring variation, but the magnitude of increased contouring variation for these contours was similar in magnitude to interobserver variation seen on PCT and CBCT. Human OAR contouring variation was similar between PCT and CBCT; CBCT_Def OAR contours were prone to contouring variation due to daily deformations (bladder/ rectal filling).