Regularized image reconstruction with an anatomically adaptive prior for positron emission tomography

Area of Science:

• Medical Imaging
• Image Reconstruction
• Positron Emission Tomography (PET)

Background:

• Anatomical information aids PET image reconstruction but requires precise lesion outlines.
• Clinical anatomical images often fail to delineate lesion boundaries accurately.

Purpose of the Study:

• To develop a novel PET image reconstruction method using anatomical priors when lesion boundaries are unclear.
• To improve lesion contrast and quantification in PET imaging.

Main Methods:

• An anatomically adaptive anisotropic median-diffusion filtering (AAMDF) prior was developed.
• A two-step joint estimation reconstruction scheme iteratively updated images and priors.
• The AAMDF prior was compared against standard maximum a posteriori (MAP) methods.

Main Results:

• The AAMDF prior enhanced lesion-to-background contrast compared to other methods.
• The AAMDF prior reduced lesion estimation errors.
• Improved performance was observed with synthetic and physical phantom data.

Conclusions:

• Anatomically derived smoothing priors can improve PET lesion contrast and quantification.
• This method does not require explicit knowledge of lesion boundaries.
• Applicable to clinical PET/CT where CT-based lesion delineation is often insufficient.