Geometric-feature-based spectral graph matching in pharyngeal surface registration

Area of Science:

• Medical Imaging
• Computational Geometry
• Radiotherapy Planning

Background:

• Accurate tumor targeting in radiotherapy requires precise alignment of medical imaging data.
• Fusing endoscopic video and CT scans aids in defining tumor volumes but necessitates robust surface registration.

Purpose of the Study:

• To develop a deformable surface registration method for aligning 3D endoscopic reconstructions with CT segmentations.
• To enable accurate tumor target volume delineation for radiotherapy planning.

Main Methods:

• Utilizing local geometric features to establish initial correspondences between surfaces.
• Constructing an association graph and employing spectral graph matching for registration.
• Defining a novel similarity measure for inter-surface affinity computation.

Main Results:

• The proposed method effectively handles large non-rigid anatomical deformations.
• The registration technique demonstrates robustness against missing data and topological changes.
• Successful registration results were achieved on both synthetic and real patient data.

Conclusions:

• The developed method provides a reliable approach for deformable pharyngeal surface registration.
• This technique enhances the fusion of endoscopic and CT data for improved radiotherapy planning.
• The approach shows promise for clinical applications requiring accurate image-guided interventions.