User talk:Meshal1981

Digital imagery in health care
A digital image typically is represented in a computer as a two-dimensional array of numbers, called as bit map. Each element of the array represents intensity of small square area of the picture, called a ‘pixel’. In case of 3-D image, the image is represented by three-dimensional array of numbers. Each element in the three-dimensional array is called as ‘voxel’ that ultimately represents the volumetric image. Thus it is obvious that the 3-D image needs more physical storage than its 2-D counterpart. 3-D images reveal structures that are difficult or sometimes even impossible to see in 2-D images. It makes images easier to read and more attractive visually.

There are quite a few challenges involved with 3-D imagery from informatics perspective. Usually during medical examination, if the doctor needs more than one slice orientations, the patient has to undergo scanning process more than once. This is the case with 2D imagery. 3-D imagery comes to rescue in this particular problem. It obviates any need of rescanning the patient because; only one scanning provides equal resolution images from all the orientations by MPR. However, it introduces new challenge of increased noise. In order to overcome this challenge, a sophisticated SNR improvement algorithms are needed. This calls for more initial investments in the specialized hardware and software. A radiologist is needed to handle the sophisticated instruments in order to produce decent quality scans in optimized time. One such software Facor claims that they produce 3D MRI scans of the size 270 x 288 x 144 in 34 s on Intel® workstation (htt). They also claim that their software is easy enough to be handled by technician instead of a radiologist, which saves money for the HCO.

Just like for 2D image processing, for 3D image processing, we need image segmentation, and there are parallel methods to do that. Image segmentation is a process of partitioning an image, which typically involves locating boundaries in images, which can be extended into identifying objects. Image segmentation outputs a set of pixel sets or regions that share common properties. The properties are calculated and mapped using the computer algorithm such that the two adjacent regions are distinct when compared on the set of properties used for distinguishing the regions. These distinct regions help the radiologists and the physicians detect tumors, measure tissue volumes or study any other anatomical aspects to figure out the abnormalities and administer the treatment plan accordingly. There are many image binarization and segmentation methods that have been proposed and which are suitable for the processing of volume images (Wirjadi), such as thresholding, region growing, clustering and shape–based methods, etc.

Thresholding is the simplest method of all. This method depends on a clip-level to turn a gray-scale image into a binary image. Clip-level is nothing but a threshold value around which the image gray scale is moderated. The heart of this method is selecting appropriate threshold values. Maximum entropy method, Otsu’s method, Isodata Method, Bayesian Thresholding, Niblack Thresholding (Wirjadi) are examples of it.

One more image segmentation technique is region growing. Basic principle behind region growing methods is that all voxels belonging to one object share common properties, also called as predicate in algorithmic lingo. Appropriate neighborhood selection and seed selection is at the heart of this technique. There are two approaches to this technique. In top to bottom approach the image is taken and recursively split until the smallest possible voxel. This method is called as split and merge method. In bottom up approach (voxel-based) an object is formed from a group of voxels. In seeded region growing, seed selection is crucial. This is usually taken care of by an expert.

- Works Cited

Edward H. Shortliffe, James I Cimino (Eds.), Biomedical Informatics: Computer Applications in Health Care and Biomedicine (3rd edition), Springer, 2006 Facor™ for MRI images. (2009). Retrieved from www.facor.info:http://www.facor.info/mri2d.htm

Wirjadi, O. Survey of 3d image segmentation