Comparison of the noise power spectrum properties of medical x-ray imaging systems

Abstract

Noise in medical images is recognized as an important factor that determines the image quality. Image noise is characterized by noise power spectrum (NPS). Four methods of NPS determination were compared: Wagner, Beutel, Dobbins and Samei's methods on Lanex Regular/ TMG screen-film system and Hologic Lorad Selenia Full Field Digital Mammography system, with the aim of selecting the best method to use. These methods differ in terms of various parametric choices and algorithm implementations. The one-dimensional moving-slit method has been used in the past to characterize the NPS of analogue screen film system (Wagner's method). Beutel's method offers the advantage of providing a value of the NPS at zero frequency along with NPS calculated via autocorrelation function (ACF). The moving slit and ACF methods have been replaced by a 2-D Fourier analysis method with the advent of fast Fourier transform and faster computers. This method is based on two techniques. The first is based on the extraction of a one-dimensional slice through the two-dimensional NPS parallel to and immediately adjacent to the axes (Dobbins's method). The second is based on the extraction of a one-dimensional slice through the two-dimensional NPS, just on the axes (Samei's method). NPS computation using different methods have been attempted using codes written in MATLAB. Overall, the four methods generate a practical value of noise power spectrum between 10–3 – 10–6 mm2 at spatial frequency range 0 – 10 mm–1. It was found that the Dobbins’s method was the best method for NPS determination.