Objectives: In radiotherapy for prostate cancer the pelvic organs move and deform between fractions introducing geometrical uncertainties that make difficult to determine the delivered dose to healthy organs. This paper proposes a population model based on principal component analysis (PCA) to describe bladder motion and deformation between fractions by determining directions of geometric variability.

Materials and Methods: This study included a cohort of 20 patients treated for prostate cancer with external beam radiotherapy. For each patient a planning CT scan and several (ranging from 5 to 8) on-treatment CT scans were available. 17 patients were used to obtain directions of bladder motion and deformation called modes using PCA, and the three remaining patients were used for validation. Dominant modes were used to characterize bladder shapes and motion/deformation regions in the latent space. The modes were evaluated using a reconstruction error in both PCA model and left-out patients.

Results: A dimensionality reduction was obtained by representing any bladder in term of 104 scores instead of 1596375 voxel variables. 28 modes were considered as population directions to model bladder geometric variations.

Conclusions: A population-based PCA model was implemented to quantify uncertainties produced by bladder motion and deformation between fractions in prostate cancer radiotherapy. This model can be used to quantify uncertainties in dose delivering and tailoring of personalized radiotherapy treatments.