Mostafa Mahdipour - Department of Biomedical Engineering, Amirkabir University of Technology (Tehran Polytechnic), Tehran, Iran- Institute of Medical Science and Technology, Shahid Beheshti University, Tehran, Iran
Majid Zohrevand - Department of Medical Radiation Engineering, Shahid Beheshti University, Tehran, Iran
Mohammad Mohammadzadeh - Department of Medical Radiation Engineering, Shahid Beheshti University, Tehran, Iran

IntroductionGlioma is the most common tumor among all brain tumors. Low-Grade Glioma(LGG) is one of the four grades based on the World Health Organization(WHO) criteria. Using MRI images, the accurate segmentation of LGGs is one of the most crucial treatment procedures. Meanwhile, automatic segmentation is still hard to be achieved because of the diversity of LGGs in size, shape, texture, and location. This study compares two methods of LGG segmentation working based on the thresholding and gradient vector flow techniques.MethodsTwo segmentation methods were applied to the ten subjects of The Cancer Imaging Archive(TCIA) data set [۱]. The first one was based on using Otsu's Thresholding technique[۲], morphological operations, and extraction of tumor features [۳]. The second approach was parametric active contour(Snake) based on Gradient Vector Flow[۴]. A snake is an energy minimizing, deformable spline influenced by constraint and image forces that pull it towards object contours and internal forces that resist deformation.For evaluating, all results were obtained from two methods compared with the ground truth mask, which was downloaded from the data set using the Dice and Jaccard similarity ratio.ResultsBoth similarity ratios of the two methods were obtained close to each other. However, the snake method had better similarity scores than the method working on the thresholding technique. The mean similarity scores for both methods are given in Table ۱. Furthermore, the results of segmentation for one of the subjects is shown in Figure ۱.ConclusionWe examined both methods so far. Although the Snake method had a slightly better similarity score than the first approach and both methods had relatively good results, the results of both methods are still far from Grand Truth and should be closer to it. Our suggestion for the future works is to use these methods in machine learning and deep neural networks.