Sinogram bow-tie filtering in FBP PET reconstruction.

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Image Reconstruction

Background:

• Filtered back projection (FBP) in positron emission tomography (PET) commonly uses radial low-pass filtering to mitigate noise.
• Standard radial filtering can lead to a loss of image resolution.
• Alternative filtering strategies have shown varied success in balancing noise reduction and resolution preservation.

Purpose of the Study:

• To introduce a novel filtering scheme for FBP reconstruction in PET.
• To evaluate the performance of the proposed bow-tie filtering scheme against existing methods.
• To assess the impact on image noise, resolution, and contrast.

Main Methods:

• A binary mask, termed 'bow tie', was defined in the Fourier domain based on sinogram properties.
• The bow-tie mask was smoothed using Gaussian kernel convolution to prevent ringing and enhance low frequencies.
• The proposed filter combines the smoothed bow-tie filter with standard radial and axial filters for FBP reconstruction.

Main Results:

• The bow-tie filtering scheme achieved a 33% noise reduction with a 10% gain in radial resolution and a 1% increase in contrast.
• Compared to stackgram-domain filtering (33% noise reduction), the bow-tie filter showed less degradation in radial (14.5%) and tangential (16%) resolutions.
• Angular filtering resulted in greater contrast degradation (3%) and tangential resolution loss (46%) for similar noise reduction.

Conclusions:

• The proposed bow-tie filtering scheme offers an optimal balance of noise reduction, resolution enhancement, and contrast preservation in PET image reconstruction.
• This method demonstrates superior performance over standard radial, angular, and stackgram-domain filtering techniques.
• Experimental results with rodent images confirm significant image quality improvements using the bow-tie filtering scheme.