Implementation and validation of a 3-dimensional deformable registration algorithm for targeted prostate cancer radiotherapy

Abstract

The daily position and shape changes of the prostate pose an obstacle to reducing the safety margin and decreasing radiation to the surrounding normal tissues during prostate cancer external beam radiotherapy. The daily prostate deformation also complicates accurate dose tracking. The uncertainty in dosimetry is further accentuated when intraprostatic targets based on molecular imaging need to be treated to a desired dose. Against this background, we have implemented and validated a 3-dimensional deformable registration algorithm to aid targeted prostate radiotherapy. We implemented and enhanced a deformable 3-dimensional image registration algorithm. This algorithm performs registration between two computed tomography (CT) images based on iteratively minimizing the difference in their image intensities (CT numbers). We used three methods to quantify the accuracy of the algorithm. First, a deformed pelvic CT set was created using a mathematically transformed set of patient pelvic CT images that was obtained with an in-room CT scanner during a course of radiotherapy. The precise voxel-to-voxel correspondence was available from the known mathematical transformation. We measured the accuracy of the deformable registration algorithm in registering the original pelvic CT set to the deformed CT set. In the second test, a custom-made deformable pelvic phantom was used. The phantom contains bony structures, and deformable bladder, rectum and prostate. Two separate CT sets of the phantom were obtained with an endorectal balloon inserted into the rectum with or without inflation with 100 cc of air. To simulate intraprostatic targets, 23 CT-opaque seeds were embedded into the prostate. We measured the accuracy of the algorithm in registering the images of the phantom with and without deformation. Lastly, two sets of pelvic CT images of the same patient were obtained on two different days during the course of treatment. These CTs were 9 days apart. We tested the goodness of the registration. In each case, correlation coefficients that measure the similarity of CT image intensities of two CT data sets were computed for a 256 × 256 − 120 cubic mm region of interest around the prostate. Correlation coefficients were computed at each CT slice as well as globally. Based on point-to-point comparison between the mathematically transformed and the registered CT sets, we found the errors were as follows: in x (lateral) direction: −0.10 ± 0.52mm; y (anterior-posterior) direction: 0.10 ± 1.11mm; and z (superior-inferior) direction: −0.21 ± 1.10mm. The corresponding 3 dimensional error was 0.57 ± 1.56mm. The overall correlation coefficient improved from 0.971 to 0.997 with the deformable registration. Using the deformable pelvic phantom, the range of shifts of the intraprostatic seeds in space created by inflating the endorectal balloon with 100 cc of air was 4.69 mm to 9.54 mm. The corresponding mean (S.D.) shift due to deformation for the 23 seeds was 6.81(±1.33) mm. The range of error in registering the seeds was 0.23 mm to 3 mm. The corresponding mean (S.D.)error was 0.86 ± 0.53mm. The correlation coefficient improved from 0.979 to 0.993 with deformable registration. Using deformable registration on the two patient CT sets obtained on two separate CT sessions, the correlation coefficient improved from 0.945 to 0.972; discounting the top and bottom CT slices that did not have sufficient information for 3-D registration, the correlation coefficient improved from 0.945 to 0.989 for the remaining 38 CT slices. Our 3-dimensional deformable registration algorithm performs accurate registration between the reference pelvic images and its deformed counterparts. The absolute registration accuracy is in the order of 1mm. In particular, the algorithm tracked the intraprostatic seeds accurately. We argue that our algorithm may aid in optimizing dose calculation and delivery for prostate radiotherapy. It may also help to allow daily mapping of intraprostatic targets for prostate cancer radiotherapy