Session Item

To assess the residual non-rigid setup errors due to anatomical changes of the clinical target volumes (CTVs) after online verification for clinical head and neck cancer (HNC) proton therapy patients.

Eleven HNC patients treated with simultaneous integrated boost – 70Gy_RBE and 54.25Gy_RBE to the primary CTV (CTV70) and elective CTV (CTV54.25 ), respectively – were retrospectively analysed to determine the residual non-rigid setup errors. Using an in-house validated deep convolution neural network (DCNN) architecture daily cone beam CTs (CBCTs) were converted to synthetic CTs (sCTs). The sCTs have comparable quality to the CTs, validated against weekly verification CTs acquired during the course of treatment. The CTV was propagated from the original CT where planning was made to the daily sCTs using a hybrid deformable image registration (RaySearch). These auto-propagated CTVs (dirCTVs) were also reviewed by the HNC radiation oncologist (pCTVs) (see Figure 1). We focused here only on the region above the cricoid cartilage where the CBCT quality is good. This ensured high quality derived sCTs for contour review. 

The original CTV was first uniformly expanded in discrete steps of 1 mm. The CTVs of the daily sCTs were co-registered  and intersected with the expanded planning CTV to determine the volume overlap. For CTV70 we determined for the patient group the interpolated shell distance at which 95% of daily CTVs have at least 98% volume overlap with the planning CTV + d_r. The elective CTV54.25, was examined at 95% volume overlap with the planning CTV + d_r. Both pCTVs and dirCTVs were evaluated, and the dice similarity coefficient (DSC) were compared.

For the primary target, d_r was 1.7 mm and 1.86 mm for pCTV and dirCTV, respectively (see Figure 2). For the elective target, d_r was 1.29 mm and 1.51 mm for pCTV and dirCTV, respectively. Furthermore, minor non relevant differences were noted for the mean DSC for all patients between the pCTV and dirCTV for both primary and elective targets (ΔDSC_mean < 3%).

Smaller setup uncertainty settings of <2 mm are feasible when using online CBCT guided proton therapy with 6D correction. A larger study cohort would be required to see if robust planning using a 2 mm setup uncertainty setting or smaller is possible. The feasibility of using the auto-propagated CTVs instead of physician corrected CTVs to analyse the errors for a larger HNC patient population is further demonstrated.