Brain tumor segmentation in magnetic resonance imaging using OKM approach

Abstract

Brain tumor is considering life-threatening disease. Brain tumors can be of many types. Categorization is done based on behavior of tissue. Glioblastoma is one type of brain tumor, which is most aggressive in nature. The average survival rate of person having this type of tumor is 9.9 months, may vary depending on the treatment and response of the patient. Whatever the kind of brain tumor is, treatment requires proper diagnosis, treatment and follow-up. Magnetic resonance imaging, which is noninvasive in nature, is consider powerful image modality for diagnosis and follow-up purposes. During surgery, surgeons may take image as guiding tool. During planning of treatment, one of the major challenging tasks is localization of tumor tissues in the MRI image. Tumor localization is performed with the help of tumor segmentation. Brain tumor segmentation is one of the most challenging tasks, which can be found in literature. Scope of machine learning and deep learning are also investigated by some of researcher in brain tumor segmentation. Results reported were also encouraging but no one is till now able to claim its robustness. Data limitation is also one of the challenges in deep learning segmentation. Paper presents a method OKM i.e. Otsu K-means Method for tumor segmentation, which uses the combination of concepts i.e. Otsu thresholding and K-means clustering. Image modalities, which are considered for generating tumor masks, are T2-W, FLAIR. Final output of the proposed method is segmented mask, which contains enhancing tumor, necrosis, non-enhancing tumor and edema. BRATS’ data set was considered during development of proposed method. Results obtained through proposed method were compared with ground truth, which were included with BRATS data. Dice coefficient shows encouraging results. Proposed algorithm is robust and efficient, as it does not involve training and large size databases.