Session Item

Changes in prostate shape significantly affect the accuracy of treatment delivery and that effect is intensified in hypofractionated regimes or in prostate proton radiotherapy, in which sensitivity to anatomical changes is more dominant than in conventional photon radiotherapy. The aim of this study was to investigate principal modes of  prostate deformation at subsequent time points, mimicking intra-fractional deformations.

Prostate deformations were analyzed in the 2D cine MR series which were acquired for a total of 8 minutes in 9 patients. The average number of frames per patient was 300. The landmarks were placed on the first frame of the cine MR sequence by a Harris Corner Detector algorithm, and then tracked automatically, frame-by-frame by using a Kanade-Lucas-Tomasi algorithm (Figure 1(a)). In this way, on every frame, we created shape instances annotated by tracked landmarks. For statistical analysis of shapes, Generalized Procrustes Analysis (GPA) was used. The GPA provides a metric distance among the shapes, estimates mean shape and shape variability within samples. 

Principal Components with the largest variation of landmarks in individual patients showed that deformation of the prostate and seminal vesicles is differently distributed and highly correlated with changes in the rectum. The landmarks representing the posterior part of the prostate showed larger variation over time. The full bladder protocol constrains changes in the bladder, hence having minimal impact on prostate deformation. The sudden movement of the prostate caused by gas in the rectum elongates the prostate in the superior-inferior direction. Smaller or almost negligible changes were noted in anterior parts of the prostate, apex, or base. In Figure 1, the methodology is illustrated for one example patient. 

We propose a methodology for describing prostate intra-fractional deformations using landmark tracking and GPA. This enables us to show intra-individual changes in prostate shape. From statistical shape analysis of landmark data configurations we obtained a measure of distance between shapes, generated in successive time points of cine MR, mean posture of landmarks, and shape variability within the sample. The patterns of intra-fraction prostate changes and the main mode of deformation could eventually improve definition of treatment margins and robustness settings in hypofractionated photon and proton treatment planning.