Anatomy assisted PET image reconstruction incorporating multi-resolution joint entropy

Area of Science:

• Medical Imaging
• Image Reconstruction
• Signal Processing

Background:

• Positron Emission Tomography (PET) image reconstruction can be enhanced by incorporating anatomical information from MRI or CT using similarity measures like joint entropy (JE).
• Existing JE-based methods classify voxels solely on intensity, neglecting crucial structural spatial information.
• This limitation hinders optimal performance, especially under varying noise levels in PET imaging.

Purpose of the Study:

• To develop and evaluate an anatomy-assisted maximum a posteriori (MAP) algorithm for PET image reconstruction that integrates spatial information.
• To introduce a novel wavelet-based joint entropy (WJE) measure to enhance the anatomical prior in PET reconstruction.
• To assess the performance of the WJE-MAP algorithm against intensity-only JE-MAP and maximum likelihood methods across different noise levels and in real patient data.

Main Methods:

• Developed a WJE-MAP algorithm integrating spatial information via wavelet multi-resolution analysis into the JE measure.
• Calculated derivatives of subband JE measures with respect to PET voxel intensities, analogous to inverse wavelet transform implementation.
• Validated the WJE-MAP algorithm using simulated PET data (BrainWeb phantom) at varying noise levels and real patient PET/MRI datasets (DPA-173 and Florbetapir).

Main Results:

• WJE-MAP performed similarly to JE-MAP at low noise levels in gray matter (GM) and white matter (WM) regions.
• At medium noise levels, WJE-MAP began to outperform JE-MAP, particularly in the less uniform GM region.
• In high noise simulations and real patient studies, WJE-MAP demonstrated clear improvement over JE-MAP, reducing noise while maintaining comparable regional mean values to maximum likelihood reconstruction.

Conclusions:

• The WJE-MAP algorithm effectively incorporates spatial information into PET image reconstruction, improving performance in higher noise conditions.
• The algorithm shows robust performance across simulations and clinical patient studies, demonstrating its potential for quantitative PET imaging.
• WJE-MAP offers a promising advancement for clinical quantitative PET by enhancing image quality and reducing noise through anatomically guided reconstruction.