Oversampled filters for quantitative volumetric PET reconstruction

Area of Science:

• Medical Imaging
• Image Reconstruction
• Signal Processing

Background:

• Three-dimensional filtered backprojection (3D FBP) commonly uses filters defined in the Fourier domain.
• Direct sampling of these filters in the Fourier domain introduces a DC shift artifact due to reconstruction filter aliasing.

Purpose of the Study:

• To develop an improved filter construction technique for 3D FBP to mitigate the DC shift artifact.
• To present an efficient method for filter construction without requiring storage of the entire oversampled filter array.

Main Methods:

• Implemented a filter construction technique utilizing Fourier domain oversampling.
• Developed an efficient method for filter construction avoiding the need for large array storage.
• Applied the technique to multiple-pass algorithms for cylindrical PET scanner data reconstruction.

Main Results:

• The Fourier domain oversampling technique effectively reduces the DC shift artifact in reconstructed images.
• The efficient filter construction method allows for practical implementation without excessive memory requirements.
• The developed algorithms enable reconstruction without fitting reprojected views to scanner data, enhancing quantitative accuracy.

Conclusions:

• Fourier domain oversampling is a viable strategy to reduce artifacts in 3D FBP.
• Efficient filter construction methods are crucial for implementing advanced image reconstruction algorithms.
• The improved 3D FBP technique enhances quantitative accuracy in PET imaging, particularly for cylindrical scanners.