PHSPP03 Presentation Time: 2:50 PM: Dosimetric Considerations of the Use of Synthetic MRI for Prostate HDR Brachytherapy

Abstract

<h3>Purpose</h3> In prostate HDR brachytherapy, MRI, while providing exquisite soft-tissue contrast and anatomic delineation, its acquisition slows the treatment workflow and the image registration between CT and MRI adds uncertainty to the treatment planning process. Deep learning-based MRI synthesis from CT is possible but has yet to be clinically validated. To address this, we assessed the dosimetric and workflow impact of synthetic MRI (sMRI) use in prostate HDR brachytherapy. <h3>Materials and Methods</h3> 58 paired pelvis CT and T2-weighted MRI datasets from patients treated with HDR brachytherapy to either a prescription dose of 13.5 Gy or 15 Gy were collected. A deep learning-based framework was developed to synthesize MRI data from CT and was validated using 17 additional patient datasets. The prostate was contoured on the sMRI, which was used to create a new planning target volume (PTV) with a 3 mm isotropic expansion and removal of the urethra, bladder, and rectum. New brachytherapy treatment plans were generated to cover the sMRI PTV with the catheter positions visible in the CT scans by adjusting the source dwell times from the original treatments. This dose was overlaid onto a second PTV, created from the prostate contour by the same observer on real MRI (rMRI). This observer is different from the one who contoured the prostate for the clinically delivered treatment, so the inter-observer variability of the prostate contour on rMRI can be compared with the inter-modality variability between the prostate contour on sMRI and rMRI using the dice similarity coefficient (DSC). The impact of sMRI use was assessed with dosimetric volume histogram (DVH) parameters PTV V100%, V150%, V200%, bladder D1cc, urethra D1cc, and rectum D1cc (percent of prescription dose averaged over 17 test cases ± standard deviation, institutional constraints based on ABS prostate high-dose rate task group). The sMRI PTV DVH metrics were compared with the rMRI DVH metrics, such that we assessed how delivering dose to the sMRI PTV would cover the ground truth rMRI PTV. The bladder, rectum, and urethra DVH metrics were compared with those of the original treatment. The workflow impact was assessed by comparison of the time for sMRI generation with the time from the end of the CT acquisition to the end of the MRI acquisition during the clinically delivered plan. <h3>Results</h3> The inter-modality variability of the prostate contour between sMRI and rMRI was found to be statistically similar to the inter-observer variability of the prostate contour on rMRI assessed with DSC, indicating that the prostate contour is similarly sensitive to observer as it is to sMRI (table a). Target coverage of sMRI PTV was not statistically different to that of rMRI PTV (table b). While there were statistical increases (p<0.05) in bladder, rectum, and urethra doses in the sMRI-planned cases (table c), in no case did this lead to an increase above the constraint for D1cc. Lastly, a full sMRI volume can be generated from the corresponding CT in approximately 10 seconds, compared to the 73.6 minutes on average between the completion of the CT and rMRI scans during the clinically delivered treatment workflow, not including image registration which will add further time to the procedure. <h3>Conclusions</h3> To our knowledge, this is the first report which details the dosimetric similarity and workflow advantages from the use of synthetic MRI for prostate HDR brachytherapy.