Patient-specific assessment of dosimetric deformable registration uncertainties and respiratory motion effects for moving tumours in proton therapy

Purpose To develop a patient-specific assessment of the dosimetric impact of deformable image registration (DIR) by quantifying DIR uncertainties using the distance discordance metric (DDM), and to demonstrate whether it is sufficient for assessing respiration-induced dose degradation in proton therapy (PT) treatments. Methods 4D-CT images were acquired to evaluate tumour motion. The study included eight lung and two liver cancer patients. The DDM was computed for each voxel and patient using the ANACONDA algorithm. DDM was validated on two different thoracic 4D-CT patient datasets with expert-placed landmarks. We conducted 100 simulations by varying the deformation vector fields applying a normal distribution N(0, DDM(x,y,z)) to each voxel. To evaluate the dosimetric impact of DIR uncertainties, we evaluated the mean of the 4D accumulated perturbed doses ± its standard deviation ( 4D mean ± S D 100 ). To assess the dosimetric impact of respiratory motion, 4D mean 100 was compared with the 3D planned dose. Results DDM values for clinical target volume (CTV) were below the deformation grid voxel (2.5 mm). Dosimetric impact of DIR uncertainties showed a maximum variation of ±3.8% in the CTV. Dose degradation caused by anatomical changes was most evident in lung, resulting in a D99%[Gy] reduction of −5.9% for the CTV near the mediastinum and −11.8% in liver. Variations of up 10.2% were observed in the average dose to the great vessels. Conclusion DDM was validated for the assessment of DIR uncertainties and their impact on 4D accumulated doses in PT. Dose degradation in the CTV was larger than discrepancies caused by DIR uncertainties.