Comparison between MAP and postprocessed ML for image reconstruction in emission tomography when anatomical knowledge...

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Image Reconstruction

Background:

• Previous studies compared penalized-likelihood (MAP) and postsmoothed maximum-likelihood (ML) reconstruction for uniform resolution in emission tomography.
• Penalized-likelihood reconstruction was not found to be superior to postsmoothed ML in those prior applications.
• Current research focuses on noise suppression tuned with anatomical information, assuming limited but exact anatomical data is available.

Purpose of the Study:

• To compare two methods of incorporating anatomical information for noise suppression in emission tomography.
• To determine the more efficient reconstruction method when anatomical information is available.
• To evaluate image quality and lesion detectability using quantitative metrics.

Main Methods:

• Method 1: Anatomical information incorporated into the prior of a maximum a posteriori (MAP) algorithm during reconstruction.
• Method 2: Anatomical information imposed as a postprocessing step on an unconstrained maximum-likelihood (ML) reconstruction.
• Simulations using noisy PET data with inserted lesions were performed, and images were analyzed using bias-noise curves and observer performance metrics (non-prewhitening and channelized Hotelling observers).

Main Results:

• The postprocessing method was found to be inferior to MAP reconstruction.
• The inferiority of the postprocessing method persists unless noise correlations between neighboring pixels are accounted for using a prewhitening filter.
• The prewhitening filter's shift-variant and object-dependent nature suggests MAP reconstruction is a more efficient approach.

Conclusions:

• Maximum a posteriori (MAP) reconstruction with integrated anatomical information is more efficient than postprocessing anatomical information into ML reconstructions for emission tomography.
• The findings are particularly relevant for applications requiring noise suppression tuned with anatomical data.
• MAP reconstruction offers a more robust and efficient solution for lesion detection and image quality enhancement in such scenarios.