Interobserver Variability And Dosimetric Impact In Structure Delineation Of Organs At Risk On Cone Beam Ct

The purpose of this study was to evaluate the feasibility of contouring organs at risk (OAR) relevant for potential adaptation decisions on cone beam CT (CBCT) and to determine the variability in inter-observer delineation and its dosimetric impact. The contours were compared to automatically deformed contours to assess the feasibility of utilizing deformed contours clinically. The accuracy of deformable registration has been highly investigated but how that may influence dose is rarely addressed yet important to consider prior to clinical utilization. Images and contours from three patients were evaluated. The planning CTs with original contours and CBCT images (fraction 10) from two separate image guidance systems (XVI and OBI) were imported into RayStation v4.5.2. Five observers delineated 15 OARs on each image set, while blinded to all other delineated volumes. A 4-point Likert scale survey was administered for subjective confidence in contouring. The original contours from planning CT were deformed to the CBCT and compared to observer generated contours. Inter-observer variability was quantified using the DICE; a DICE ≥ 0.8 indicated acceptable agreement. The estimated delivered dose was calculated on the CBCT and the dosimetric impact of contour delineation was evaluated by comparing representative dose metrics of observer contours against clinical contours on planning CT and deformed contours on CBCT, respectively. The mean DICE index, respectively, for (1) observer contours on XVI CBCT, (2) observer contours on OBI CBCT, (3) deformed contours on XVI CBCT, and (4) deformed contours on OBI CBCT were as follows: brainstem 0.7,0.7,0.7,0.8; chiasm 0.2,0.3,0.2,0.3; cord 0.7,0.7,0.7,0.8; larynx 0.4,0.7,0.6,0.7; mandible 0.9,0.9,0.9,0.9; left optic nerve 0.5,0.5,0.5,0.5; right optic nerve 0.5,0.5,0.5,0.6; left parotid gland 0.6,0.8,0.7,0.9; right parotid gland 0.7,0.8,0.7,0.8; left plexus 0.3,0.2,0.4,0.2; right plexus 0.3,0.2,0.3,0.2; postcricoid 0.4,0.5,0.4,0.6. The representative dose metrics of observer contours are similar to clinical contours on planning CT and deformed contours on CBCT: 0.1cc max dose to brainstem was 216 cGy (SD=6 cGy) vs. 208 cGy on planning CT and 200 cGy (SD=13 cGy) vs. 188 cGy on CBCT; 0.1cc max dose to chiasm was 120 cGy (SD=17 cGy) vs. 113 cGy on planning CT and 118 cGy (SD=17 cGy) vs. 117 cGy on CBCT. The DICE of automatically deformed contours is comparable to the DICE of contours delineated by observers. The brainstem, spinal cord and parotid glands achieved a DICE of 0.8 with automated contours. The chiasm and plexus contours achieved poor agreement. The dosimetric impact of contour delineation is small suggesting that it is feasible to use automatically deformed contours in the dose accumulation process. The interaction of site and dose distribution with these metrics warrant further investigation.