Apply Field-of-View Extension and Deformable Image Registration to Improve HDR-EBRT Cumulative Dose Estimation In Cervical Cancer Patients.
Abstract
Purpose
Accurate dose accumulation in cervical cancer radiotherapy combining external-beam radiotherapy (EBRT) and High-Dose-Rate brachytherapy (HDR) requires reliable Deformable-Image-Registration (DIR) between two drastically different planning CTs. Besides large anatomical changes, the truncated Field-Of-View (FOV) in HDR-CT compromises the anatomical context of registration and increases uncertainty in Organ-At-Risk (OAR) dose estimation. This study evaluates a workflow combining FOV-extension and DIR to improve anatomical alignment and cumulative dose assessment in cervical cancer patients.
Methods
A cohort of cervical cancer patients (N=32) treated with EBRT followed by HDR brachytherapy was retrospectively analyzed. HDR-CTs with truncated-FOV were first extended using locally trained CropGAN model to restore missing anatomy. Rigid, intensity-based, and contour-based DIR methods were applied to register extended HDR-CT to EBRT-CT. Registration accuracy was evaluated using DICE similarity coefficient and Mean Distance to Agreement for pelvic OARs, including bladder and rectum. Deformation vector fields were then used to map HDR dose distributions onto the EBRT-CT geometry, calculating dose accumulation. Dose-volume-histogram (DVH) metrics were plotted to assess the impact on cumulative OAR dose estimation.
Results
Using an intensity-based DIR, extending the HDR-CT improved the mean DICE from 0.52 to 0.59, whereas rigid registration achieved a lower DICE of 0.43, which demonstrates the benefit of FOV-extension input. When applying the contour- and intensity-hybrid DIR algorithm, a high DICE value (0.81) was achieved for both extended and unextended HDR-CT inputs. In addition, qualitative assessment demonstrated that FOV-extended inputs reduced deformation artifacts, including bone elongation and excessive organ stretching. DVH analysis for one patient shows 2.65 Gy and 2.52 Gy greater cumulative D2cc for the bladder and the rectum when more accurate DIR is applied.
Conclusion
This study demonstrates that addressing HDR-CT FOV limitations prior to DIR effectively improves registration accuracy in cervical cancer radiotherapy. The improved registration accuracy can lead to substantially different cumulative dose estimation.