MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

MRI-To-CBCT Registration for Alignment of Prostate Cancer Patients in An MRI-Only Simulation Workflow: Implementation of a GPU-Accelerated Elasticity-Based Biomechanical Model Mediated by Synthetic-CT Images

K Singhrao1*, B Stiehl1, M Lauria1, J Fu1, N Parikh2, A Kishan2, A Santhanam2, J Lewis3, (1) David Geffen School of Medicine at UCLA, Los Angeles, CA, (2) Dept. of Radiation Oncology, UCLA, Los Angeles, CA, (3) Cedars-Sinai Medical Center, Beverly Hills, CA

Presentations

(Tuesday, 7/14/2020) 3:30 PM - 5:30 PM [Eastern Time (GMT-4)]

Room: Track 2

Purpose: alignment between cone-beam CT (CBCT) and MRI-only simulation images is challenging due to CBCT’s poor soft tissue contrast. However, bladder and rectum structures can be identified on both MRI and CT with lower interobserver variability than the prostate. In this study, we propose a novel framework for CBCT-to-MRI soft tissue alignment using an elasticity-based biomechanical model driven by deformations of the bladder and rectum observed in CBCT images

Methods: patients with three fiducial markers implanted prior to MRI imaging were selected for this study. MRI prostate contours were defined using T2-weighted spin-echo MRI images. An MRI-based synthetic-CT image was generated using a deep-learning classification architecture. A CT-input GPU-accelerated finite element model-based pelvic biomechanical model was used to realistically deform the synthetic-CT. Four sets of contours were generated for this study: deformed prostate contours generated using the biomechanical-model (cDefBio), deformed prostate contours generated using a commercially available deformable-image-registration tool (cDefVel), physician-drawn CBCT contours (cPhysCBCT), and physician-drawn MRI contours (cDoc). Sørensen–Dice coefficients were calculated between the ground-truth cPhysCBCT contours and the other contours. Lastly, the three sets of CBCT contours were each registered to the MR image to simulate the expected differences in patient positioning. Target registration errors (TREs) were calculated based on the location of fiducial markers in the CBCT images.

Results: mean DICE coefficients were 0.80±0.06, 0.81±0.08 and 0.77±0.08 between cPhysCBCT and, cDefVel, cDoc and cDefBio, respectively. The mean absolute TRE alignment difference compared to fiducial marker-based registration was 5±2mm, 7±3mm and 3±3mm for the cDoc, cDefVel and cDefBio structures, respectively. cDefBio contours produced the smallest TREs and similar prostate contour agreement.

Conclusion: have developed a biomechanical model-based framework to accurately predict the prostate shape and location in CBCT images using structures derived from MRI-only simulation images.

Funding Support, Disclosures, and Conflict of Interest: This work was partially supported by Varian Medical Systems, Inc.

Keywords

MRI, Treatment Planning, Elastic Matching

Taxonomy

IM/TH- MRI in Radiation Therapy: MRI for treatment planning

Contact Email