Accuracy of registrations between cone-beam computed tomography and conventional computed tomography images and dose mapping methods in RaySearch software for the bladder during brachytherapy of cervical cancer patients

Purpose
The aim of the study was to assess selected methods of image registration available in the RaySearch software and their impact on the accuracy of mapping of doses deposited in the bladder during brachytherapy (BRT) of cervical cancer in images used during external beam radiotherapy (EBRT).

Material and methods
The study was based on data from ten patients. Cone-beam computed tomography (CBCT) images (BRT) were aligned with CT images (EBRT) using four registration methods: Reg_1 (rigid), Reg_2a, Reg_2b (hybrid), and Reg_3 (biomechanical). Image mapping accuracy was evaluated based on bladder’s anatomy. Sørensen-Dice coefficient (DSC) values were analyzed for all the registrations. Discrepancies between triangular mesh points set on the basis of bladder contours were analyzed. Dose distributions from BRT were transformed according to registration results and mapped on CT images. Original BRT doses deposited in 2 cm3 volume of the bladder were compared to those transformed and associated with bladder’s volume determined on CT images.

Results
Mean DSC values amounted to 0.36 (Reg_1), 0.87 and 0.88 (Reg_2a and Reg_2b), and 0.97 (Reg_3). Significant differences were found between DSC for the following comparisons: Reg_3/Reg_1 (p = 0.001), Reg_2a/Reg_1 (p = 0.011), and Reg_2b/Reg_1 (p = 0.014). The lowest discrepancies between triangular mesh points were for Reg_3 (p < 0.001, Reg_3 vs. Reg_1, and p = 0.039, Reg_3 vs. Reg_2b). Finally, the lowest discrepancies between the original and transformed doses were found for Reg_3. Nevertheless, only 5 out of 10 observations for Reg_3 yielded error of less than 5%.

Conclusions
Biomechanical registration (Reg_3) enabled the most accurate alignment between CBCT and CT images. Satisfactory registration results of anatomical structures do not guarantee a correct mapping of primary BRT doses on the bladder delineated on CT images during EBRT. The results of dose transformation based on biomechanical registration had an error of less than 5% for only 50% of the observations.