Noise reduction in PET attenuation correction by maximum likelihood histogram sharpening of attenuation images.

UNLABELLED

A new method for PET transmission data processing was developed and found to reduce transmission noise in 18F-FDG cardiac emission images. This method is based on a model that describes the measured attenuation image histogram as some unknown true underlying histogram, blurred by noise.

METHODS

Emission data from an elliptical phantom (cardiac insert) and three humans were reconstructed using transmission data of varying duration with varying levels of smoothing. Biases and noise levels (cardiac sector analysis) were evaluated for the phantom (simulated replicates) and human emission images.

RESULTS

The estimated attenuation histograms typically displayed three distinct peaks corresponding to air, lung and soft tissue without a priori assumptions of the underlying mu values. This method effectively sharpened the histogram peaks and performed well for the phantom and human data. For intermediate transmission noise levels, biases in the phantom sector values were <4%. The human sector results were more variable but consistent with the phantom results. Noise reduction (approximately 30%) was demonstrated across all smooth levels for the phantom data.

CONCLUSION

This histogram sharpening method introduces only small bias in the cardiac sector values while achieving an increase in effective transmission scan time of 50-100%. Alternatively, histogram sharpening permits less transmission data smoothing without increased noise.