Clinical Report: Enhancing Deformable Image Registration of CT-CBCT for Cervical Cancer
Overview
This study presents a deep learning-based framework aimed at improving CT-CBCT deformable image registration for cervical cancer adaptive radiotherapy. The proposed method enhances anatomical alignment accuracy and robustness.
Background
Cervical cancer treatment often involves radiotherapy, which can be complicated by anatomical variations during treatment. These variations can lead to dosimetric uncertainties, necessitating accurate image registration methods to ensure optimal treatment delivery. The development of robust deformable image registration techniques is crucial for effective adaptive radiotherapy workflows.
Data Highlights
Metric
Internal Dataset
External Dataset
Dice Score (Bowel)
82.84%
83.40%
HD95 (Bowel)
-
11.70 mm
Non-positive Jacobian (%|J|≤0)
0.13
-
Key Findings
The NGF-UTSRMorph method achieved a Dice score of 82.84% for bowel on the internal dataset.
On the external dataset, the method achieved a higher Dice score for bowel (83.40%) compared to baseline methods.
Boundary alignment improvements were particularly noted in high-contrast regions.
The method maintained comparable deformation smoothness while enhancing anatomical alignment accuracy.
Clinical Implications
The NGF-UTSRMorph framework offers an approach to improve CT-CBCT registration in cervical cancer radiotherapy.
Conclusion
The study demonstrates that the NGF-UTSRMorph method significantly enhances the robustness of CT-CBCT registration, particularly in boundary-sensitive metrics.
