Statistical image reconstruction from correlated data with applications to PET.

Area of Science:

• Medical Imaging
• Statistical Modeling
• Computational Science

Background:

• Emission tomography reconstruction typically assumes independent data, ignoring real-world correlations from detectors and processing.
• These correlations, often modeled by large covariance matrices, pose computational challenges for traditional algorithms.
• Ignoring correlations leads to less precise image estimates in emission tomography.

Purpose of the Study:

• To develop computationally tractable methods for reconstructing correlated emission tomography data.
• To improve image precision by effectively incorporating covariance information into reconstruction algorithms.
• To simplify the use of non-diagonal covariance matrices in statistical reconstruction.

Main Methods:

• Proposed two methods to simplify the use of non-diagonal covariance matrices by dimensionality reduction and flexible correlation modeling.
• Applied methods to simulated positron emission tomography (PET) data and data processed with Fourier rebinning.
• Incorporated methods into a penalized weighted least-squares 2D reconstruction framework.

Main Results:

• Demonstrated the feasibility of incorporating complex correlation structures into reconstruction.
• Achieved more precise image estimates compared to conventional independent data methods.
• Showcased computational tractability of the proposed methods for correlated PET data.

Conclusions:

• The proposed methods offer a computationally efficient way to handle correlated data in emission tomography.
• These techniques provide more accurate image reconstruction by accounting for data dependencies.
• This work advances statistical reconstruction for improved PET imaging analysis.