A generalization of the maximum likelihood expectation maximization (MLEM) method: Masked-MLEM.

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Image Reconstruction

Background:

• Weighted Robust Least Squares (WRLS) algorithm is effective for 2D image reconstruction with uncertain system matrices and projection data.
• Extending WRLS to 3D reconstruction is challenging due to the non-smooth and non-strongly-convex nature of the optimization problem.
• Existing methods struggle with robust image reconstruction in the presence of system uncertainties and noise.

Purpose of the Study:

• To develop a generalized iterative method for robust 3D image reconstruction in scintigraphy, addressing system uncertainties.
• To introduce the Masked-Maximum Likelihood Expectation Maximization (Masked-MLEM) algorithm, an adaptation of the MLEM algorithm.
• To overcome the limitations of existing algorithms in handling system uncertainties and noise during image reconstruction.

Main Methods:

• The Masked-MLEM algorithm utilizes selected subsets (masks) of the system matrix and projection data for image updates, accommodating system uncertainties.
• Validation involved experimental data from collimated and uncollimated imaging instruments, including SPECT and scintigraphy.
• Comprehensive Monte Carlo simulations for 3D collimatorless tomography were performed to assess performance under varying system uncertainties.

Main Results:

• Masked-MLEM and standard MLEM reconstructions showed similar results when system uncertainties were negligible.
• Masked-MLEM significantly outperformed standard MLEM when the system matrix was an approximation.
• The algorithm demonstrated reliable image reconstruction across different levels of system uncertainties.

Conclusions:

• The Masked-MLEM algorithm provides more robust image reconstruction than standard MLEM, particularly when system matrix knowledge is imperfect.
• It effectively reduces the occurrence of false activity in regions without radioactive sources.
• The method offers a reliable approach for image reconstruction in scintigraphy with system uncertainties.