Analysis of factors affecting positron emission mammography (PEM) image formation
Image reconstruction for positron emission mammography (PEM) with the breast positioned between two parallel, planar detectors is usually performed by backprojection to image planes. Three important factors affecting PEM image reconstruction by backprojection are investigated: 1) image uniformity (flood) corrections, 2) image sampling (pixel size) and 3) count allocation methods. An analytic expression for uniformity correction is developed that incorporates factors for spatial-dependent detector sensitivity and geometric effects from acceptance angle limits on coincidence events. There is good agreement between experimental floods from a PEM system with a pixellated detector and numerical simulations. The analytic uniformity corrections are successfully applied to image reconstruction of compressed breast phantoms and reduce the necessity for flood scans at different image planes. Experimental and simulated compressed breast phantom studies show that lesion contrast is improved when the image pixel size is half of, rather than equal to, the detector pixel size, though this occurs at the expense of some additional image noise. In PEM reconstruction counts usually are allocated to the pixel in the image plane intersected by the line of response (LOR) between the centers of the detection pixels. An alternate count allocation method is investigated that distributes counts to image pixels in proportion to the area of the tube of response (TOR) connecting the detection pixels that they overlay in the image plane. This TOR method eliminates some image artifacts that occur with the LOR method and increases tumor signal-to-noise ratios at the expense of a slight decrease in tumor contrast. Analysis of image uniformity, image sampling and count allocation methods in PEM image reconstruction points to ways of improving image formation. Further work is required to optimize image reconstruction parameters for particular detection or quantitation tasks.
