Inter-individual organ-driven CT registration for dose mapping in prostate cancer radiotherapy

Toxicity prediction is crucial in prostate cancer radiotherapy. There is thus a need to model voxel-wise relationships between dose distribution and toxicity events. However, building such a model is challenging due to difficulties in the mapping of the organs and of the dose distribution. We designed an organ-driven registration method merging the deformation fields computed for registrations first of the computed tomography scans and second of the organs delineations. The merging was done by weighting the different deformation fields according to the distance between voxels and organs. We evaluated this strategy on 29 individuals and compared the results with those obtained with intensity-based affine and non-rigid registrations. The dice scores obtained were above 0.9 for each organ ans the accuracy was improved by 41.1% for the prostate, 31.5% for the bladder and 54.2% for the rectum. The method allows to accurately map inter-individual dose distributions to a single coordinate system to perform further voxel-based statistical analysis.