Preliminary investigations on the use of MRI for decision making in adaptive proton radiotherapy for head and neck cancer

Abstract

Introduction For proton therapy of head and neck (H&N) virtual CT-scans (VCTs) generated from CBCTs can aid decision making for treatment adaptation. Purpose The aim was to investigate the accuracy of VCTs generated from daily MRIs. This work is relevant for proton centers without CBCT, but equipped with an out-of-room MRI scanner. Materials and methods For 5 H&N tomotherapy patients, 4 weekly CT-scans and MRIs were used. To allow comparison, CBCTs were generated from CTs using an in-house algorithm and for each pair of weekly CT and MRI pose changes were minimized by non-rigid registration. VCTs were generated by non-rigidly deforming the planning CT into the anatomy of the MRIs (VCT_MRIs) and CBCTs (VCT_CBCTs). The geometric accuracy of VCTs was evaluated by the Dice’s coefficient (DSC) between structures on VCTs and on weekly CTs. Water equivalent thickness (WET) maps at the surface of structures were calculated from stopping power maps for projections at 90° and 270 ° angles. Differences in WET maps between weekly CTs and VCTs were analyzed using a gamma index of 1mm/1mm. Results The average DSCs for VCT_MRI and for VCT_CBCT were respectively, 0.96 ± 0.01 and 0.98 ± 0.01 (external), 0.78 ± 0.09 and 0.86 ± 0.04 (parotids), 0.88 ± 0.04 and 0.91 ± 0.02 (brainstem), and 0.55 ± 0.21 and 0.62 ± 0.22 (GTV). The percentage of pixels that passed the gamma index criterion was significantly higher for VCT_CBCT (75.30 ± 13.07% versus 45.84 ± 21.27%). Conclusion Geometrically accurate VCTs could be generated from daily MRIs and CBCTs. Full dosimetric calculations will provide further insight on the use of MRIs for decision making in proton radiotherapy. Disclosure This work was supported by the Fonds de la Recherche Scientifique - FNRS under Grant(s) n° 7463215F.