Dynamic PET reconstruction using temporal patch-based low rank penalty for ROI-based brain kinetic analysis.

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Image Reconstruction

Background:

• Dynamic positron emission tomography (PET) enables tracking radiopharmaceutical biodistribution over time.
• Limited photon counts per frame in dynamic PET compromise temporal resolution, leading to noisy images and degraded time activity curves (TACs) with conventional methods like OSEM.
• Existing advanced algorithms utilize spatio-temporal regularizations to improve image quality.

Purpose of the Study:

• To develop a novel temporal regularization technique for dynamic PET image reconstruction.
• To leverage the self-similarity of image patches for improved temporal regularization.
• To address the non-convexity and memory limitations of existing optimization frameworks.

Main Methods:

• Exploiting self-similarity of patches in dynamic PET images through a novel temporal regularization.
• Employing a low-rank constraint, insensitive to global intensity variations, to capture patch correlations.
• Developing a new optimization framework using the concave-convex procedure (CCCP) to handle non-Lipschitz and non-convex terms.

Main Results:

• Demonstrated the effectiveness of exploiting patch correlation via a low-rank constraint for temporal regularization.
• Proposed a novel optimization framework that overcomes limitations of direct PIDAL application.
• The CCCP-based approach addresses the challenges posed by non-convexity and memory requirements.

Conclusions:

• The novel temporal regularization method enhances the quality of dynamic PET images and extracted TACs.
• The proposed optimization framework offers a viable solution for complex dynamic PET reconstruction problems.
• This approach holds promise for improving quantitative analysis in dynamic PET studies.