The geometric and dosimetric effect of algorithm choice on propagated contours from CT to CBCTs

Abstract

Purpose or Objective Efficient adaptive radiotherapy requires rapid plan assessment based on the anatomy of the day to determine whether plan adaptation is required. To determine the relevant DVH parameters, contouring the organs at risk (OARs) on the daily image is required. Manually contouring in such a short timescale puts a significant burden on oncologists. Using software to propagate the planning contours onto the daily images can theoretically remove this burden. Since there are many commercially available options, the geometric and dosimetric differences of the propagated contours should be assessed, before evaluating their role in plan adaptation. Material and Methods Planning CT (pCT) contours and 5 shading-corrected CBCTs [1] (for more accurate dose calculation) acquired weekly throughout treatment for 10 randomly selected head and neck cancer patients were used. The patients were replanned with VMAT in RayStation v6.99 (RS). The spinal cord (SC), brainstem (BS), parotids and larynx were propagated using 2 strategies: from pCT to each CBCT (pCT2CBCT) and in cascade, i.e. from the pCT to CBCT1 to CBCT2. The pCT2CBCT strategy was used in ADMIRE, Mirada and RS, and the cascade in ADMIRE and RS, for a total of 5 methods. Mean distance to agreement (mDTA) between propagated contours was computed between pairs of contours for all propagating methods. BS and SC were expanded to form PRVs. For the dosimetric impact, the plan was calculated on each CBCT after setup correction. The relevant DVH parameters were extracted (D1cc for BS and SC PRVs, mean dose parotids and larynx) for each contour and each method, and compared using the Kolmogorov-Smirnoff test to determine statistically significant differences. The doses were compared against the clinical constraints for BS and SC PRVs (D1cc< 54 and 46 Gy respectively). Results All methods produced similar contours evidenced by the range of the mDTA (up to 1.7 mm) (table 1). Figure 1 shows the dose differences for the OARs. There were no statistically significant differences except for SC PRV, with no particular method outperforming the others. The presence of significant differences on the SC is indicative of the high dose gradients present in the area, confirming that small variations on the propagated contours cause larger dose changes. The median dose for both left and right parotids showed a small increase for most methods. The larynx dose remained similar with time. There were 6 instances of the spinal cord failing against clinical constraint, and 5 for the brainstem - all were in cases of high dose gradients near to the OAR. No particular method failed consistently. Conclusion Most propagated contours were geometrically and dosimetrically similar, regardless of method used. For the SC, however, the variations were large enough to causes significantly different D1cc. Clinician review of propagated contours in areas of high dose gradients is critical.