Design and development of efficient segmentation algorithms for the analysis of brain tumor using MR images

Abstract

In recent times, brain cancer is the significant reason for mortality of larger section of population across the globe. Different varieties of cancers, among which brain tumor is the one which requires immediate diagnosis and effective treatment to save the patient life. Segmentation of the brain tumor is a common process in the medical setup and provides data which helps for preliminary diagnosis and planning for the treatment. Performing segmentation of tumors manually radiologists or physicians is a tedious and requires lot of time because of huge amount of clinical data that are produced today in hospitals. The challenging and significant issue associated with the medical image is segmenting the given image. Automatic brain tumor segmentation, classification and identification have increased the interest in medical field due to high accuracy and less computational time. Among the imaging modalities, MRI plays a major role in visualization of anatomical structure of the brain. This helps in treatment planning, dose estimation and also image guided surgery. The main contributions to this proposed work is the development of segmentation and classification algorithms. These algorithms in turn overcome challenges of the existing brain tumor segmentation algorithms. In the proposed research work, three algorithms are developed and implemented successfully for the efficient segmentation and classification of brain tumor. newlineThe first contribution is the development of hybrid algorithm for the segmentation of brain tumor. In this method, the Variational Level sets and K-Means Clustering are combined to develop the segmentation algorithm. Recently deformable models are extensively used to detect boundaries of tumors in MRI images of the human brain. Due to extremely weak convergence of the contour towards concavities of the tumor and moving towards unwanted features of image, traditional active contours have limited applications. To overcome these limitations, the hybrid approach i.e. region based K-means clustering