Volumetric feature points integration with bio-structure-informed guidance for deformable multi-modal CT image...

Area of Science:

• Medical Image Processing
• Radiotherapy
• Artificial Intelligence

Background:

• Medical image registration, particularly CT-CBCT, is crucial for image-guided radiation therapy (IGRT).
• Traditional registration methods are computationally intensive.
• Deep learning methods can struggle with local optima, especially in low-contrast scenarios.

Purpose of the Study:

• To develop an accurate and efficient medical image registration method.
• To improve CT-CBCT registration for IGRT applications.
• To overcome limitations of existing deep learning registration techniques.

Main Methods:

• Introduced a registration method integrating volumetric feature points with bio-structure-informed guidance.
• Generated surface point clouds from segmentation labels.
• Co-guided training using surface-registered point pairs and voxel feature point pairs.

Main Results:

• Validated on paired CT-CBCT datasets.
• Achieved a 6% improvement in precision compared to other deep learning methods.
• Reached state-of-the-art performance in CT-CBCT registration.

Conclusions:

• Integrating voxel and bio-structure feature points effectively guides medical image registration networks.
• The proposed method shows significant promise for advancing medical image registration and IGRT.
• This approach offers a new direction for future research in the field.