Medical image segmentation:
Medical image segmentation is often challenging since the images can involve high levels of noise and
unclear MBSedges. We consider the method of background subtraction (MBS) in order to minimize such difficulties. When an appropriate background is subtracted from the given image and segmentation is applied to the residue of image (upper right of right figure), most segmentation methods can detect desired edges more effectively. The bottom left of right figure is a segmentation without the MBS and bottom right is segmented image with the MBS. When figures have oscillatory backgrounds, conventional segmentation methods can also detect the background as an object. Such error can be prevented by using the MBS as a pre-process of various segmentation methods. In the figure below, conventional segmentation model was applied to MRI image of heart without and with the MBS. The original image reveals noise and unclear edges over the whole domain. Witout the MBS(middle), the model shows difficulties in the detection of edges, particularly around the lower left corner of the image. On the other hand, when the MBS is incorporated (right), the model can detect the desired edges successfully.
For this problem, students will first learn some conventional image segmentation methods and their numerical techniques. Then they will learn the basic concept and procedure of MBS for image segmentation applied to medical imagery so that they can later modify and test the methods to broader examples.
Left: Original medical image, middle: segmentation without MBS, right: with MBS segmentation