MediVis project at Ostrava – CT prefiltering, segmentation and visualization

In this post I would like to continue the introduction section with the prefiltering, segmentation and visualization parts. As it was mentioned before, during the scanning process the CT machine recreates the inner image of the body in sequential axial slices.

Slices

The CT images are stored in a special format, so-called Digital Imaging and Communications in Medicine (DICOM) format. This is a well-known and widely adopted standard by hospitals and medical device industries for handling, storing, and transmitting all the necessary information in medical imaging. The most of the medical imaging machines use this as a default output standard or they have a capability to create it.

In the viewpoint of the medical treatments the precise and realistic image processing and visualization are important since they can obviate the repeated scanning and directly help doctors with the evaluation and decision making process. For example, regarding the liver carcinoma, one of the most important results of the segmentation is the exact evaluation of the organ volume. Based on this information, an exact liver resection is made. Without the information surgery can be lethal. The process consists of four main steps:

• pre-filtering
• segmentation
• post-processing
• 3D visualization.

In the pre-filtering step, the acquired images are denoised/prefiltered (all the slices). During the scanning process, random noise appears in the images.  The noise is tightly connected with the physical principle of the scanning. I have to mention here that it is just a half true, because during the scanning process many different kinds of noise can distort the final output (pictures) of the CT. Nonetheless based on the state of the art method, the application of this model can suppress the majority of the noise very effectively, so that way it is “just” a post-processing step regarding the CT scanning. This means you do not have to touch the CT machine to receive better results. Anyway the noise can be described with mathematical functions which are the base for the applied denoising/noise supressor filters. The “denoised” images support the segmentation algorithms to achieve better results.(I used the “” to give a feeling about the noise could not be completely removed in most cases. We can suppress to some extent and help with describing functions to achieve a better performance.)

It allows also the reduction of the radiation dose during the scanning. We focus on the state of the art methods. The segmentation process uses the results of the previous stage, so in all denoised images the algorithm tries to segment the previously selected organ (for instance, the liver). We focused on one of the state of the art methods, which uses the k-means algorithm. The pictures below shows the segmentation results in the case of one image slice regarding the non pre-filtered and pre-filtered.

The final step is the creation of the 3D object and visualization of it in a suitable environment based on the segmentation results, and provide the expected measurement such as volume measurement. During common work we concentrate on one part of the human body which is the liver.