Click here to

Session: Imaging: Multimodality Imaging Topics [Return to Session]

Spatial Transformation Network-Based MRI-CBCT Deformable Registration for CBCT-Guided Prostate Radiotherapy

S Momin*, Y Lei, T Wang, Y Fu, J Roper, P Patel, A Jani, J Bradley, T Liu, X Yang, Department of Radiation Oncology and Winship Cancer Institute, Emory Univ, Atlanta, GA

Presentations

MO-IePD-TRACK 1-3 (Monday, 7/26/2021) 5:30 PM - 6:00 PM [Eastern Time (GMT-4)]

Purpose: Radiotherapeutic dose escalation to dominant intraprostatic lesions (DIL) in prostate cancer could potentially improve tumor control. This study aims to propose an unsupervised method to accurately register multiparametric MRI to CBCT images for improved DIL delineation, treatment planning, and dose monitoring in CBCT-guided prostate radiotherapy.

Methods: Our proposed spatial transformation network (STN) consists of two consecutive networks: zoomed-out and zoomed-in feature networks. Both networks have adversarial learning integrated to further force the realism of derived deformation vector field (DVF). The network begins with MRI and CBCT as inputs to the zoomed-out network to predict an overall approximation of DVF, which is transferred to spatial transformer to generate zoomed-out version of deformed MRI. For subsequent zoomed-in network training, 3D patches are extracted from zoomed-out version of deformed MRI. Zoomed-in network learns smaller changes within the extracted patches to generate more local DVF, which is transferred to spatial transformer for generating deformed patch. We performed leave-one-out cross-validation experiment on 30 pelvic cancer patients to evaluate the proposed registration method. The target-registration-error (TRE) as the Euclidean distance of the identified landmarks was used to quantify the accuracy of the proposed MRI-CBCT registration.

Results: Overall results of this work showed feasibility of our proposed network architecture to accurately register multimodal images with good spatial alignment of different structures. An average TRE of 2.9mm (range 2.1-3.7mm) was achieved after registration for all patients, compared with 5.7mm (range 4.9-8.3mm) before registration. The proposed network can generate a final DVF in a single forward prediction, which is faster than the conventional iterative registration algorithms.

Conclusion: We developed a new framework to accurately register the prostate on MRI to CBCT images for external beam radiotherapy. The proposed method could be used to aid DIL delineation on CBCT, treatment planning, dose escalation to DIL, and dose monitoring.

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email

Share: