Logo image
Back
A feasibility study on the use of synthetic CT in pelvic radiotherapy: automated pipeline for multilevel AI workflows in CT-CBCT deformable registration
Conference proceeding

A feasibility study on the use of synthetic CT in pelvic radiotherapy: automated pipeline for multilevel AI workflows in CT-CBCT deformable registration

Mahya Ahmadzadeh, Santhosh Vadivel, Keyur D. Shah, Gregory C. Sharp, Nagarajan Kandasamy and John Walsh
v 13930, 139300Q
02 Apr 2026
DOI: https://doi.org/10.1117/12.3088056
Featured in Collection :   Drexel's Newest Publications

Abstract

Prostate cancer is among the most common cancers in men, and radiotherapy is a standard treatment option. Accurate dose delivery requires accounting for daily anatomical changes in the prostate and nearby organs at risk, particularly the bladder and rectum. To enable this, image-guided radiotherapy (IGRT) protocols typically acquire a cone-beam CT (CBCT) at each treatment fraction and register it to the pre-treatment planning CT. Deformable image registration (DIR) between the planning CT and daily cone-beam CT (CBCT) is commonly used for this purpose. However, DIR often fails on CBCT due to its low soft-tissue contrast and imaging artifacts. To address these limitations, recent work has explored the use of deep learning models such as CycleGANs to generate synthetic CT (sCT) from CBCT, improving image quality for registration. Building on this idea, we developed an improved CycleGAN architecture to reduce CBCT distortions and hallucinations. Beyond intensity information, registration accuracy can also be enhanced with point-to-distance map (PD) constraints derived from organ contours. These constraints may come from manual expert segmentations or from fully automated approaches such as TotalSegmentator, which generates multi-organ segmentations directly from CT volumes. Our project is structured in two phases: (i) refining sCT generation with our improved CycleGAN, and (ii) developing an automated pipeline to benchmark sCT performance in downstream tasks including organ segmentation and deformable registration. The overall aim is to provide clinicians with a risk-aware framework that allows them to choose between manual, semi-automated, or fully AI-driven workflows for pelvic image-guided radiotherapy.

Metrics

1 Record Views

Details

Logo image