Optimization of Non-Local Means filtration for Cone Beam Computed Tomography
|Chalmers tekniska högskola / Institutionen för matematiska vetenskaper
|Chalmers University of Technology / Department of Mathematical Sciences
|Cone Beam Computed Tomography (CBCT) is a medical imaging technique used to visualize the internals of an object from a set of X-ray images. The X-ray images are taken with a cone shaped X-ray beam at many angles around the object being visualized. The images are then back projected through a volume forming a 3D representation of the object. Non-local means (NLM) denoising was applied to the X-ray images before they were used in the back projection algorithm. NLM was implemented for the graphics card and runs in real time while the X-ray images are taken. Compared to current denoising methods the result was less noisy and had higher sharpness. It was found that using NLM on the X-ray images led to streak artifacts and smearing of fine details in the reconstructed volume. Many versions of NLM were tested to reduce these effects, including variance reduction, total variation and using adjacent images in the scan. The different versions however gave very similar results. The state-of-the-art denoising algorithm BM3D was also tested for reference and the results were very similar to NLM. NLM denoising in the reconstructed volume as a post-processing step was also tested which showed improved results over denoising the projection images. Denoising in 3D led to less artifacts and better preservation of details. The runtime for the 3D-denoising is only a few seconds on the graphics card, making it useful in practice. 3D-NLM denoising offers the possibility of lowering the radiation dose to the patient since the noise can be reduced while keeping edges and details intact. The noise can be reduced such that a low dose scan is comparable to a high dose scan denoised with current CBCT denoising methods. A comparison of spatial resolution and preservation of details indicates that the dose could be lowered by as much as half without significantly lowering the image quality.
|Optimization of Non-Local Means filtration for Cone Beam Computed Tomography
|Examensarbete för masterexamen
|Engineering mathematics and computational science (MPENM), MSc