Motion correction for separate mandibular and cranial movements in cone beam CT reconstructions

Area of Science:

• Medical Imaging
• Cone Beam Computed Tomography (CBCT)
• Image Reconstruction

Background:

• Patient motion during oral and maxillofacial cone beam CT scans frequently degrades reconstruction quality.
• Independent mandibular movements create complex, unpredictable motion patterns challenging existing correction methods.

Purpose of the Study:

• To develop a motion estimation and correction technique for separate cranial and mandibular motions using only 2D projection images.
• To address limitations of previous methods in handling complex, articulated patient movements.

Main Methods:

• An iterative, three-step algorithm was developed, modeling articulated motions as independent rigid motions.
• A deep neural network was used for segmenting the cranium and mandible in projection images.
• The algorithm iteratively refines motion poses by minimizing projection errors while keeping the 3D reconstruction fixed.

Main Results:

• The marker-free approach achieved up to 85% higher reconstruction quality compared to the projection error.
• The method outperformed existing techniques that model only a single rigid motion.
• Validation was performed using synthetic data and real-world data from a skull-hexapod setup.

Conclusions:

• The algorithm enhances visual quality in motion-affected CBCT scans, potentially avoiding re-scans and reducing patient radiation dose.
• The technique is flexible for various regions of interest and applicable to local tomography.