REVIEW ON DETECTING BRAIN TUMOUR USING CLUSTERING APPROACH
Archana D A1 Dr Chidananda Murthy M V2 Dr M Z Kurian3
1M.Tech Student, Dept of ECE, Sri Siddhartha Institute of Technology, India
2AssociateProfessor, Dept of ECE, Sri Siddhartha Institute of Technology, India
chidanandamurthymv@ssit.edu.in
3HOD, Dept of ECE, Sri Siddhartha Institute of Technology, India
ABSTRACT
Image segmentation place a prominent role in image processing and computer vision disciplines. It means the process of segregating a digital image into multiple non-overlapping regions. There are so many image segmentation techniques like edge base, clustering and region-based segmentation technique. Although of the variety of image segmentation techniques, still the selection of an appropriate technique for a special type of image is a most difficult task. Not all techniques are appropriate for all types of images. The main obstacle in segmentation algorithms is the difficulty of balancing the over-segmentation and under-segmentation. But still medical image segmentation is one of hot research topic in the research area. Though it’s quite a challenging problem because of images with poor contrasts, noise, and missing or diffuses boundaries. The magnetic resonance images (MRI) scan is comfortable for diagnosis. It will not upset the human body because it uses magnetic field and radio waves and it doesn't use any of the radiation which is harmful.
Keywords: Segmentation, clustering, MRI
INTRODUCTION
The brain is an organ which acts as master of each individual; any malfunction of brain may leads to serious hazardous to lead a life. One of the major diseases related to brain is brain tumor. A brain tumor means it is considered as unusual progression of the cells inside the brain. Brain tumors are of two types i.e., primary and secondary. Primary tumors are again classified as benign and malignant tumors. Benign tumors can be removed easily as they usually have a border or an edge. Malignant tumors are more serious as they grow rapidly in crowd and also occupy the nearby healthy tissue. Physician will usually concentrated in treating strokes not in dealing with tumor. As a result tumor detection plays very vital role in diagnosing of brain tumor. It is possible to augment the lifespan of the brain tumor affected patient if it is detected as early as possible. Hence the development of a proficient medical image segmentation technique is evolved with the benefits of least possible user interaction, fast computation, accurate, and robust segmentation results will help physicians to diagnose precisely. The most extensively used techniques in image segmentation are clustering techniques. Clustering means it is an unsupervised learning technique which requests the user to determine the number of clusters in advance to categorize the pixels. Thus, the cluster can be defined as it is a collection of similar pixels and it is dissimilar to the pixels which are belonging to other clusters. Clustering techniques carried out by performing clustering in 2 methods, either by partitioning method or by grouping similar pixels.
In any living organisms it is very much necessary to control and co-ordinate overall body parts. To do this each individual has a brain. Brain acts as master to all other organ systems, if it undergoes to any of the disorder then it will leads to critical condition. Among those brain tumor is one of the dangerous disorder.
A brain tumor, in generally called as an intracranial tumor. In other words it is a condition of growing abnormal cells and then starts to multiply uncontrollably, apparently unchecked by the mechanisms that regulate normal cells. Till now it’s clear that there are more than 150 different types of brain tumors are been characterized, but the two main types of brain tumors are named as primary and metastatic. [1]
Primary brain tumors are characterized based on its spreading i.e., it will originate from the tissues present in the brain or the brain's immediate surroundings. Primary tumors are again classified as glial which means that they are composed of glial or non-glial i.e., they are grown on or in the brain, including nerves, blood vessels and glands and then benign or malignant.
Metastatic brain tumors are characterized into another type of brain tumor since it arises anywhere in the body (may be in the breast or lungs) and then they move on to the brain, via bloodstream. Metastatic tumors are recognized as cancer which is malignant in nature.
The World Health Organization (WHO) has evolved a grading system that indicates a tumor's malignancy or benignity formulated on its histological features seen under a microscope.
Symptoms [2]
Image courtesy by MDPI
Diagnosis
Brain tumors are easy to find by using well defined imaging techniques. Those imaging techniques are computed tomography scan in simple words CT scan and magnetic resonance imaging scanning (MRI). Except this other MRI sequences will aid the surgeon to plan assist in surgeon to segment the tumor which is basically depends on the location of the normal nerve pathways of the brain. Intraoperative MRI is also implemented in the course of surgery to guide the tissue biopsies and tumor exclusion. It is common to use Magnetic resonance spectroscopy (MRS) because it inspects the tumor's outline and thus defines the nature of the tumor seen in MRI. Along with that Positron emission tomography (PET scan) will be used to detect recurrent brain tumors.
RELATED WORKS
K-Medoids algorithm is also known as partitioning around medoid algorithm. Initially it was proposed by Kaufman and Rousseeuw in 1987. Medoid is a point in any cluster whereas the dissimilarities with other points present in that cluster must be as less as possible. To achieve this clustering it is necessary to initiate the algorithm with all necessary requisites [3].
Algorithm:
As K-Medoids algorithm is simple to understand also it is easy to implement so it is used in medical image segmentation. But K-Medoids algorithm is not much applicable to cluster non-spherical or irregular objects. Also the results will keep on varying on applying same dataset repeatedly. Since the medoids are selected randomly at first.
If the Fuzzy C means is combined with the spatial constraints then the resultant algorithm is suitable for image segmentation process. It is effective because the SFCM algorithm utilizes fuzziness to cluster each pixel and also it exploits the information related to spatial context. But this spatial context can raise its sensitivity into some sort of noise up to some extent. Spatial is one type of function which integrates the membership function of each pixel with some considerations. This method produces the regions into homogeneous region also it reduces noise. This technique is recognized because it works well with noisy image while segmenting the image. SFCM algorithm can be applied to both singular feature data image and multiple feature data embedded with spatial information [4][5]. One of the brain tumor images is used as input to Spatial Fuzzy C means algorithm and the result shown below.
Image by Egyptian informatics journal
In above figure (a) and (d) shows the original MR image with some noise, (b) and (c) shows SFCM clustered image.
Fuzzy K-C means algorithm is one type of hybrid clustering algorithm where it involves some of the properties from Fuzzy C means and K-Means algorithm [6]. This algorithm involves various steps they are as follows.
One of the brain tumor images is used as input to Fuzzy K-C means algorithm and the result shown below.
Image courtesy by IISTE
First image is input image and latter image shows the Fuzzy K-C means clustered image.
There are various clustering techniques to cluster the image properly. Also every clustering technique has its own advantages and limitations. Based on the information to be extracted, the appropriate clustering technique can be selected.
REFERENCES