Improving lesion detectability of a PEM system with post-reconstruction filtering
We present a method to quantify the image quality of a positron emission mammography (PEM) imaging system through the metric of lesion detectability. For a customized image quality phantom, we assess the impact of different post-reconstruction filters on the acquired PEM image. We acquired six image quality phantom images on a Naviscan PEM scanner using different scan durations which gave differing amounts of background noise. The image quality phantom has dimensions of 130 mm Ã? 130 mm Ã? 66 mm and consists of 15 hot rod inserts with diameters of approximately 10 mm, 5 mm, 4 mm, 3 mm, and 2 mm filled with activity ratios of 3.5, 6.8 and 12.7 times the background activity. One region of the phantom had no inserts so as to measure the uniformity of the background noise. Lesion detectability was determined for each background uniformity and each activity ratio by extrapolating a fit of the recovery coefficients to the point where the lesion would be lost in the noise of the background (defined as 3 times the background's standard deviation). The data were reconstructed by the system's standard clinical software using an MLEM algorithm with 5 iterations. We compare the lesion detectability of an unfiltered image to the image after applying one of five common post-reconstruction filters. Two of the filters were found to improve lesion detectability: a bilateral filter (9% improvement) and a Perona-Malik filter (8% improvement). One filter was found to have negligible effect: a Gaussian filter showed a 1% decrease in lesion detectability. The other two filters tested were found to worsen lesion detectability: a median filter (8% decrease) and a Stick filter (7% decrease).