SPECTRAL ESTIMATION FILTERS FOR NOISE REDUCTION IN X-RAY FLUOROSCOPY IMAGING

Abstract

In clinical x-ray fluoroscopy, moving images are acquired at very low x-ray dose so that only 10–500 x-ray quanta contribute to each pixel. The resulting Poisson statistic causes the images to be strongly affected by quantum noise, which, in the observed images, is spatially correlated and signal-dependent. In this contribution, we develop a spatial frequency domain method for intraframe quantum noise reduction, which takes the nonwhite noise power spectrum into account. Each image is subjected to a block DFT or DCT. The magnitude of each observed spectral coefficient is compared to the expected noise variance for it, which is derived from a suitable quantum noise model. Depending on this comparison, each coefficient is more or less attenuated, leaving the phase unchanged. Finally, the image is backtransformed and re-assembled. Using this method, noise power reductions of 60% are possible.