Comparison of low computational complexity filters suitable for real-time fluoroscopy image denoising

Fluoroscopy devices provide real-time, radiographic movies of patient and it is widely utilized as support for surgery and in diagnostic. Low X-ray dose results in intense quantum noise, which is modeled as Poisson-distributed stochastic signal. Recently, a specific filter technique was introduced to suppress quantum noise in fluoroscopy. Filter operation relies on the estimation of the relationship between noise variance and mean pixel intensity relative to the fluoroscopy device setup. By holding this information, noise suppression can be exclusively performed by averaging the only adjacent data in space and time that have high probability to belong to the noise statistics. The performances of this filter were compared to those of another filter based on the maximum a posteriori probability criterion designed for Poisson's noise suppression. The performances of the two filters, in terms of SNR and PSNR, resulted very similar, but they are a bit lower than more sophisticated filters such as BM3Dc. Nevertheless, they offer a simplicity of the algorithms that allows their realization in real-time to support interventional fluoroscopy application