Image reconstructions from super-sampled data sets with resolution modeling in PET imaging.

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Image Reconstruction

Background:

• Spatial resolution is a key limitation in Positron Emission Tomography (PET) imaging.
• Existing PET scanners with fixed crystal sizes can be improved through mechanical movements like scanner wobbling and object shifting.
• Multiple acquisitions from varied positions offer complementary data and enhanced spatial sampling.

Purpose of the Study:

• To develop an efficient and effective reconstruction framework for generating super-resolution PET images from super-sampled datasets.
• To explore methods for improving spatial resolution in PET imaging without altering hardware.

Main Methods:

• Introduced a super-sampling data acquisition model incorporating tomographic, downsampling, and shifting matrices.
• Extended MLEM and Landweber algorithms for super-sampled data reconstruction.
• Developed a backprojection-filtration-like (BPF-like) method and explored variant reconstruction approaches.

Main Results:

• All tested algorithms significantly improved contrast recovery coefficients and reduced background artifacts using super-sampled data.
• MLEM outperformed Landweber, which in turn outperformed the BPF-like method in image quality.
• Reconstructions using finer system matrices and higher counts at increased iterations yielded superior image quality.

Conclusions:

• A novel super-sampling reconstruction framework was developed for PET imaging, capable of reconstructing super-resolution images from motion-aware super-sampled datasets.
• Super-sampling PET acquisition offers an economical and effective strategy to enhance image quality, particularly for region-of-interest imaging in preclinical and clinical settings.