Publication: Brain tumor image segmentation using graph-cut technique
Loading...
Date
2022-08-01
Authors
Ting, Tze Cheng
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Brain tumor is a collection of brain cells that have grown out of control which can cause serious issues to our health. It can be either cancerous or noncancerous, but both can cause brain damage when they keep on growing. The common diagnosis for the brain tumor is carried out by computed tomography (CT) and magnetic resonance imaging (MRI). Medical image segmentation is crucial in computer-aided diagnostic systems. It separates an image into sections based on a predefined description, such as segmenting humanorgans/tissues for boundary detection, tumor detection/segmentation, and mass detection in medical applications. The complicated brain structure is a big challenge and the large brain data cause the time consuming to radiologist. The algorithm of image segmentation must produce a clear brain tumor image. This paper presents a proposed graph-cut technique for the digital image segmentation. The graph-cut technique has generally higher accuracy than other method such as histogram thresholding method. The graph-cut technique illustrates the image as a graph, marked the targeted object as foreground and remained object as background. Preprocessing is carried out in this paper for enhance the quality of the image. There are set of two results will be analyzed in this paper which are proposed graph cut method and imageSegmenter, a Matlab toolbox. The results are evaluated by using two parameters, Sørensen-Dice similarity coefficient and Jaccard similarity coefficient. The similarity coefficient is measured in between the output image and ground truth image. The Sørensen-Dice similarity coefficient from proposed graph cut method are 0.9350, 0.8878, 0.9144, 0.9130 for Image1 to Image4 respectively whereby the Jaccard similarity coefficient are 0.8799, 0.7982, 0.8422, 0.8400 for Image1 to Image4 respectively. The Sørensen-Dice similarity coefficient from imageSegmenter APP are 0.9389, 0.8889, 0.9177 and 0.9202 for Image1 to Image 4 respectively whereby the Jaccard similarity coefficient are 0.8848, 0.8000, 0.8480 and 0.8522 for Image1 to Image4 respectively. From the results, both methods can successfully segment tumor from image.