Generating patient specific pseudo-CT of the head from MR using atlas-based regression

Area of Science:

• Medical Imaging
• Radiotherapy Physics
• Computational Anatomy

Background:

• Accurate radiotherapy planning and PET image attenuation correction rely on computed tomography (CT) derived electron density information.
• In specific clinical scenarios like stereotactic neurosurgery and PET/MR imaging, only magnetic resonance (MR) images are available, posing a challenge for radiation transport simulation.
• The absence of CT data necessitates alternative methods for generating accurate electron density maps.

Purpose of the Study:

• To develop and validate a method for generating patient-specific pseudo-CT images of the head from anatomical MR images.
• To enable radiotherapy planning and PET image correction in cases where CT data is not acquired.
• To provide a viable alternative to specialized MR imaging techniques for CT synthesis.

Main Methods:

• Introduced atlas-based regression, a technique analogous to atlas-based segmentation, to synthesize pseudo-CT images.
• Utilized a database of MR and CT image pairs (atlases) and registered them to a target MR image.
• Employed the Morphon deformable registration algorithm, enhanced with a certainty mask, and a novel iterative fusion method (joint mean registration and voxelwise median) to generate the final pseudo-CT.

Main Results:

• The atlas-based regression method successfully generated realistic, patient-specific pseudo-CT images from MR data.
• The proposed novel fusion method, particularly the voxelwise median, yielded results comparable or superior to existing techniques.
• The generated pseudo-CTs demonstrated accuracy suitable for clinical applications, comparable to methods requiring specialized MR sequences.

Conclusions:

• Atlas-based regression offers a robust and effective solution for synthesizing pseudo-CT images from MR data.
• This method addresses the limitations of CT unavailability in specific neuroimaging and combined imaging applications.
• The developed technique shows significant promise for clinical integration in radiotherapy and PET imaging workflows.