Method and impact of attenuation correction on quantitative accuracy for the PET/X breast imaging scanner

Objectives We are constructing a dual-modality PET/mammography scanner (the PET/X) for dedicated breast imaging. For quantitative imaging, e.g. to assess response to therapy, a method is needed to estimate attenuation and scatter.

Methods We have developed a method for estimating the necessary attenuation image from a single-view mammogram. The method is based on the constraints that biological breast materials have essentially identical attenuation coefficients at 511 keV. In other words, only the boundaries of the breast being imaged are needed, which can be estimated from the mammography image and the known separation between compression surfaces. Second-order corrections are available for the imaging geometry of the mammogram scanner, the compression paddle distortion, the variable thickness at the edge of the breast, and implants. We used simulations to evaluate the impact of attenuation correction as a function of location in the compressed breast and for different breast thicknesses of 3, 5, 7, and 9 cm.

Results With attenuation correction the correct PET image values were recovered independent of location in the breast and breast thickness. Without attenuation correction, there were large variations in the estimated PET value, decreasing by 50% or more from the edge to the center of the compressed breast. These errors and variations in errors increase with increasing breast thicknesses under compression.

Conclusions Accurate attenuation correction is feasible on a dual modality PET/mammography scanner. Breast shape and thus attenuation at 511 keV can be estimated from a mammography image and the known separation between detectors. For the geometry of the PET/X scanner, not performing attenuation correction induces errors of 50% or more that are dependent on breast thickness and location in the breast. Thus accurate attenuation correction is needed for quantitatively accurate PET breast images.

Research Support Supported by The UW/Coulter Translational Research Partnership Program, and the University of Washington C4C-CGF fund