To improve the robustness and anatomical alignment accuracy of CT-CBCT deformable image registration for cervical cancer adaptive radiotherapy.
Approach:
Deep Learning Framework: A deep learning-based registration framework (NGF-UTSRMorph) was developed, enhancing a transformer-based model with a normalized gradient field constraint.
Training Methodology: The network integrates convolutional and transformer modules and uses a composite loss function combining mutual information, deformation regularization, and gradient-based similarity.
Evaluation Metrics: The method was evaluated on internal and external CBCT-CT datasets using Dice, HD95, and Jacobian determinant.
Key Findings:
On the internal dataset, the method achieved Dice scores comparable to the baseline across all structures, with bowel at 82.84%, and showed improved deformation regularity with low non-positive Jacobian determinant values (%|J|≤0: 0.13), which is comparable to UTSRMorph (0.12) and lower than VoxelMorph (0.19) and TransMorph (0.21).
On the external dataset, the method demonstrated improved generalization, achieving higher Dice for the bowel (83.40% vs. 82.99% and 82.90%) and reduced HD95 (11.70 mm vs. 11.91 mm and 11.85 mm).
Improvements were particularly evident in boundary alignment in high-contrast regions, while maintaining comparable deformation smoothness.
Interpretation:
The proposed method (NGF-UTSRMorph) improves the robustness of CT-CBCT registration and enhances boundary alignment without compromising deformation smoothness.
Limitations:
The study does not address potential limitations related to the generalizability of the model across different clinical settings, which may affect its applicability in diverse patient populations.
The reliance on specific datasets may limit the applicability of findings to broader populations, potentially impacting the robustness of the model in real-world scenarios.
Conclusion:
The NGF-UTSRMorph method enhances CT-CBCT registration robustness and boundary alignment, particularly in cross-scanner tests and boundary-sensitive metrics.
