A denoising method forwhole-body low-dose x-ray imageswith adaptable parameter control

A flexible patch based approach for combined denoising and contrast enhancement of digital X-ray images

“…Two of these approaches are simplifications of the proposed one as they consist of fully enhancing the input (D I t ) or the denoised (DÎ t ) band-pass images, and are denoted by NE (noise enhancement) and DE (denoised enhancement), respectively. Note that the method DE is similar to the one that we have proposed in Irrera et al (2013) with the difference that manual tuning of the filter parameters is avoided as presented in Section 2.3. The third method is a EOS proprietary algorithm, denoted by EOSE, that is used by default for exams in clinical routine.…”

“…Some aspects introduced in this paper are partially related with two of our former works. The first one (Irrera et al, 2013) has allowed showing that the denoising filter parameters need to be tuned for different anatomical regions in order to efficiently restore a full body X-ray image. However, while in this previous work the parameters were set by manually adjusting the shape of a curve, this is not the case for the XNLM filter here introduced, that now exploits automatic estimates of the noise levels.…”

Contrast enhancement of Micro Dose X-ray images

“…In a former work [9], we pointed out that the choice of one value of α on a full body X-ray image is sub-optimal: there is a strong risk of loosing anatomical information in low absorption regions, e.g. bone textures.…”

“…So, since the noise propagates through coarser levels than the first one, the aforementioned noise containment technique is not optimal anymore. Then, a possible method consists in denoising the image before enhancing the local contrast [8], [9]. Nevertheless, this kind of approach could lead to an artificial look of structures at medium frequencies, e.g.…”

“…For this aim, we combine the decomposition of the original image and of its denoised version, including the denoising directly into the enhancement loop. The denoised image is obtained using a Non-Local means (NL-means) filter [10], [11] with an adaptable parameter control related to X-ray absorption that we described in [9]. We want to stress that the proposed method is not just a MS denoising technique.…”