A tumor is an anomalous mass in the brain which can be cancerous. Such anomalous growth within this restricted space or inside the covering skull can cause problems. Detecting brain tumors from images of medical modalities like CT scan or MRI involves segmentation (Division into parts) for analysis and can be a challenging task. Accurate segmentation of brain images is very essential for proper diagnosis of tumor and non-tumor areas for clinical analysis. This paper details on segmentation algorithms for brain images, advantages, disadvantages and a comparison of the algorithms.

1. Integrated Intelligent Research (IIR) International Journal of Data Mining Techniques and Applications
Volume: 07, Issue: 01, June 2018, Page No.124-129
SSN: 2278-2419
124
An Investigation into Brain Tumor
Segmentation Techniques
S.Rathnadevi1
,T.N.Ravi2
1
Ph.D Scholar, PG & Research Dept. of Computer Science, Periyar E.V.R. College (Autonomous),
Trichirappalli – 620023
2
Research Co-ordinator & Asst. Professor, PG & Research Dept. of Computer Science, Periyar E.V.R. College
(Autonomous)
Trichirappalli - 620023
Email : rathnajjc@gmail.com, proftnravi@gmail.com
Abstract— A tumor is an anomalous mass in the brain which can be cancerous. Such anomalous growth within
this restricted space or inside the covering skull can cause problems. Detecting brain tumors from images of
medical modalities like CT scan or MRI involves segmentation (Division into parts) for analysis and can be a
challenging task. Accurate segmentation of brain images is very essential for proper diagnosis of tumor and non-
tumor areas for clinical analysis. This paper details on segmentation algorithms for brain images, advantages,
disadvantages and a comparison of the algorithms.
Keywords— MRI Scan; CT Scan; Tumor; CAD
I. INTRODUCTION
Brain tumors can erupt anywhere within the brain with the growth of anomalous cells. There are at least over 100
histologically distinct brain tumors, each having its own range of clinical presentations and treatment. Brain
tumors have an effect that it can totally change a human’s way of life. Brain disease is a standout amongst the
riskiest in light of the fact that almost all tumors that emerge in the brain are malignant. Diagnosis of a brain
tumor begins with physical examination and medical history. Muscle strength, body coordination and memory are
checked by a doctor. Medical imaging modalities like CT scan of the head or MRI of the head are additionally
used by clinicians to identify the location and size of the tumors.
Figure. 1 – Brain Tumor
These medical modalities are a boon to doctors as they provide multiple images of the brain in high resolution.
Tumors cause direct harm by attacking brain tissue where their manifestations depend on size and location..
Separating various tissues of the brain before analysis of tumors is an uphill task in image processing as the brain
has various tissues like Gray Matter White Matter and Cerebrospinal Fluid. Though Intensity is an important
element in separating diverse tissues of the brain in images, utilizing intensity alone to section the complex brain
structure may be inefficient. Radiologists perform visual and qualitative analysis of these medical images.
Intricate diagnosis with an elaborate quantitative analysis for assessment is facilitated by Computer aided
diagnosis (CAD). CAD systems are effective as they have lesser negative impacts on diagnosis results when

2. Integrated Intelligent Research (IIR) International Journal of Data Mining Techniques and Applications
Volume: 07, Issue: 01, June 2018, Page No.124-129
SSN: 2278-2419
125
compared to manually driven diagnosis. Segmentation is an underlying step of image analysis and partitions the
input image based on common characteristics like color, intensity or texture [2]. It helps quantification of required
regions in images for analysis by establishing relationships of data [9]. This paper discusses brain image
segmentation methods. Figure1 shows a human brain with tumor
II. BRAIN IMAGE SEGMENTATION AND CHALLENGES
Image segmentation is a key step in computerized medical analysis like radiotherapy planning, diagnosis and
treatment [12]. Segmentation methods can be classified as manual, semi-automated or automated segmentations
based on human interactions [16]. Brain image segmentation is also prone to challenges and problems. The input
brain images may have noise or heterogeneous levels in image intensities. The contrast level of the brain images
may also vary, thus making identification of tumors a challenge. It becomes complex when the distinction
between the healthy tissues and tumor regions is thin or cannot be identified. Further, the size of the tumors also
plays an important role in identification as tiny tumors may be overlooked. Human errors can occur in
considering brain tissues as tumors, creating problems in accuracy.
A. Segmentation Methods
The above discussed difficulties in brain image segmentation, has been the driving force for researchers to
develop various algorithms in brain image segmentations. The complexities involved in segmentations have also
helped in researcher trying combination of algorithms for the same [15]. Image segmentation can be done in
many ways and though there are hundreds of such techniques these papers details on a selected few techniques.
Manual Segmentation: In manual segmentation processes an expert physician or individual with intrinsic
knowledge on brain anatomy labels brain regions using brain imaging software. This slice-by-slice introspect
segmentation is the most accurate method, but time-consuming and likely fail with novices. Manual
segmentation used to prepare ground truths and labels and validate other segmentation techniques.
Intensity-based Methods: These segmentation methods categorize brain regions based on pixel intensity or
image intensity-based features extracted from brain images. Though these methods are popular their
applications are limited to classification of Gray Matter, White Matter and Cerebrospinal Fluid regions of the
brain. A detailed or sub classification is challenging and may fail. It is also dependent on efficient preprocessing
of brain images. Thresholding is an example of intensity based method. Thresholding is an effective and simple
method, where objects in an image are classified based on object intensities and intensity thresholds. Using
histograms of an image, single objects can be separated from their background using global thresholding method
(single threshold). In cases where there are more than two intensities, segmentation is carried out using local
thresholding. The threshold values are based on local statistical properties and prior knowledge[10]. Also,
partial volumes of regions are calculated for thresholding [3] or thresholding occurs using Gaussian distribution
of data values [6]. Figure 2 shows the output of a thresholded Brain Image
Figure. 2 - Brain Image Thresholding
Region based segmentation: Pixels in an image are examined in these methods. Pixels with similar
characteristics are grouped to form regions. The regions also satisfy common criteria. Regions thus formed can
be assembled to get the original image. The simplest region-based segmentation technique is the region
growing, which is used to extract connected regions of similar pixels from an image. Pre-processing methods
help segmentation with realistic results [8]. Figure 3 shows the output of a Region segmented brain image.

3. Integrated Intelligent Research (IIR) International Journal of Data Mining Techniques and Applications
Volume: 07, Issue: 01, June 2018, Page No.124-129
SSN: 2278-2419
126
Figure. 3 - Region Segmented Image.
Pixel classification based segmentation: In pixel classification segmentation images are represented in a
feature space using their attributes of gray level, color and texture in the image. These segmentation have
constraints in clustering pixels using supervised or unsupervised classifiers in the feature space. Figure 4 shows
the output of Pixel based segmentation.
Figure. 4 - Pixel based segmentation
Model based segmentation: Segmentation of 3D images are usually done using model based segmentation
techniques like parametric deformable models. A continuous connected model is built for specific anatomic
structures based on prior knowledge of object like shape, orientation and location. Geometric representation of
object structures is difficult due to variability in their shapes [4].Parametric deformable models (active contour
models or snakes) can segment required anatomic structures with a priori knowledge about the shape , size and
location of the structures. These segmentation techniques are less efficient in handling topological changes in an
image.
Machine Learning Methods: Machine learning (ML) techniques are the most used for intrinsic evaluation of
anomalous growth in the brain like tumors. Supervised ML methods are trained on training samples
where the ground truths and labels are provided for each sample. In unsupervised ML methods, no such labels
are provided for samples. Though supervised and unsupervised methods are used in segmentation, unsupervised
methods have been of greater interest to researchers. Supervised learning methods need a larger sample set in
training. In contrast, unsupervised techniques are less sensitive to data size and can function even in the case of
small data size. However, the accuracy of classification might not be as good as supervised methods, but it can
compete in certain cases. A few of these methods namely KNN, K-Means Clustering, Fuzzy C Means and
Morphological based segmentation are detailed below

4. Integrated Intelligent Research (IIR) International Journal of Data Mining Techniques and Applications
Volume: 07, Issue: 01, June 2018, Page No.124-129
SSN: 2278-2419
127
KNN: K-Nearest Neighbor (k-NN) is a simple technique that can provide classifications for image
segmentation[5]. It is based on an Euclidean distance and voting function within a neighbourhood [11]. KNN
consumes more processing time. Segmentation is done in four steps namely determined k value, Euclidian
distances calculations, sorting the minimum distance and determination of classs based on majority. Figure 5
shows KNN segmentation Output.
Figure. 5 - KNN segmentation.
K means Clustering: K-means (Unsupervised ML algorithm) is a non-complicated technique used for solving
issues in clustering [7].A defined value of ‘k’ is the input corresponding to the number of clusters and centers.
It is continuous clustering and an iterative process. Mean value of pixels in one iteration becomes the centroid
of next iteration. Thus, the final iteration will have the best result achieving k-level segmentation of the
image[17]. Figure 6 shows the output of K means Clustering.
Figure. 6 - K means Clustering
Fuzzy C Means: fuzzy c means divides a given set of data into a clusters or a group, with homogeneity inside
the group and heterogeneity between groups [13].In FCM the initial points are determined. Fuzzification of a
technique is allowing partial membership in one or more clusters to each data point. Each pixel or member of a
fuzzy set can belong to multiple clusters or sets. Membership functions define the fuzziness in an image.[14].
Figure 7 shows the output of FCM.
Figure. 7 - Fuzzy C means Output
Morphological Based Segmentation: These segmentation techniques remove demerits of segmentation and
operate on bi-level images [18]. Morphological operations used in images can be opening/ closing , boundary
extractions, Region fillings, extraction of connected components and thinning/thickening. Pixels are structured
using a structuring element with dynamic sizes to give a new processed image output. Figure 8 shows the output
of Morphological based segmentation. Thus, multiple techniques are employed to evaluate for segmenting brain
images. Table 1 lists a comparative performance of different techniques in brain image segmentation.

5. Integrated Intelligent Research (IIR) International Journal of Data Mining Techniques and Applications
Volume: 07, Issue: 01, June 2018, Page No.124-129
SSN: 2278-2419
128
It can be observed that most classifiers score above 90% in accuracy of segmentations. Morphological based
operations involve many pre-processing methods which enhance the quality of images and thus pave way for a
higher segmentation accuracy.
Figure. 8 - output of Morphological based segmentation
Table 1 – Comparative performances of Segmentation Techniques
SNO Technique Accuracy
1 K-NN 91-94%
2 K-Means 91-94.5%
3 FCM 92-95.5%
4 Morphological Operations based segmentation 97-99%
III. CONCLUSION
Brain image segmentation’s importance can not be denied and is present in all types of analysis starting from 3D
MRI data visualization to structural MR data analysis, brain segmentation represents an essential step in the
process. Various techniques, such as threshold-based, machine learning-based or hybrid methods, have been
developed and optimized to perform brain parcellation. However, not all techniques produce a high accuracy rate,
which is due to a variety of issues. For instance, supervised machine learning methods require enough labeled
data (ground truth) to train a model than can segment the brain with a high accuracy rate. As mentioned above,
providing the ground truth by manually segmenting the data is a timeconsuming approach. Consequently,
unsupervised machine learning using unsupervised methods, or so-called clustering algorithms, are of greater
interest. The extant literature reveals that several successful segmentation cases in which fuzzy c-means (FCM)
was utilized have been reported. Furthermore, researchers have indicated that by combining various methods to
develop hybrid algorithms, they have achieved more robust segmentation methods that yield higher accuracy
rates.
References
[1] Luc Vincent, Edward R. Dougherty, "Morphological Segmentation for Textures and Particles", Published
as Chapter 2 of Digital Image Processing Methods, E. Dougherty, Editor, Marcel-Dekker, New York, 1994,
Pages 43—102
[2] Bhandarkar S, Koh J, Suk M. Multiscale image segmentation using a hierarchical self-organizing map.
Neurocomputing 1997;14:241–72.
[3] ChangSun Y, SunHan K, JunSong C, MooNoh S, WonPark J. Threshold estimation for region segmentation
on MR image of brain having the partial artifact.ICSP2000 2000:1000–9.
[4] McInerney T, Terzopoulos D. Deformable models. Handbook of medical imaging processing and analysis.
San Diego,CA: Academic Press; 2000. p. 127–45
[5] S. Warfield, Duda, “K-Nearest Neighbour Classification”, 2001
[6] Stadlbauer A, Moser E, Gruber S, Buslei R, Nimsky C, Fahlbusch R, et al. Improved delineation of brain
tumors: an automated method for segmentation based on pathologic changes of 1H-MRSI metabolites in
gliomas. Neuroimage 2004;23:454–61
[7] Zhang DQ, Chen SC (2004), “A novel kernelized fuzzy c-means algorithm with application in medical
image segmentation”, ArtifIntell Med 32(1):37–50

6. Integrated Intelligent Research (IIR) International Journal of Data Mining Techniques and Applications
Volume: 07, Issue: 01, June 2018, Page No.124-129
SSN: 2278-2419
129
[8] Bleau A, Leon L. Watershed-based segmentation and region merging. Comput Vis Image Under
2000;77(3):317–70.][Gies V, Bernard T. Statistical solution to watershed over-segmentation. IntConf Image
Process; 2004. p. 1863–6. Image Technol 2004:207–11
[9] Wong K. Medical image segmentation: methods and applications in functional imaging. Handb Biomed
Image Anal Segmentation Models Part B 2005;2: 111–82.
[10]Shanthi KJ, Sasi Kumar M. Skull stripping and automatic segmentation of brain MRI using seed growth
and threshold techniques. Conf Intelligent Adv Sys 2007:422–6
[11]A. El-Sayed, El-Dahshan, M. S. Abdel-Badeeh, and H. Y. Tamer, “A Hybrid Technique for Automatic MRI
Brain Images Classification”, Digital Signal Processing, Vol. 20, Issue 2, March 2010, pp. 433-441.
[12]N. Sharma and L. Aggarwal, "Automated medical image segmentation techniques," Journal Medical
Physics, vol. 35, pp. 3-14, 2010
[13]Mohammad Shokouhifar, Gholamhasan Sajedy Abkenar, "An Artificial Bee Colony Optimization for MRI
Fuzzy Segmentation of Brain Tissue", 2011 International Conference on Management and Artificial
Intelligence IPEDR vol.6 (2011) IACSIT Press, Bali, Indonesia.
[14]Brain Tumor Segmentation and Its Area, Calculation in Brain MR Images using K- Mean,Clustering and
Fuzzy C-Mean Algorithm, IEEE-International Conference On Advances In Engineering, Science And
Management (ICAESM -2012) March 30, 31, 2012
[15]Saman Sarraf , Cristina Saverino, Ali Mohammad Golestani, "A Robust and Adaptive Decision-Making
Algorithm for Detecting Brain Networks Using Functional MRI within the Spatial and Frequency Domain,"
in The IEEE International Conference on Biomedical and Health Informatics (BHI) , Las Vegas, 2016
[16]Konstantinos Kamnitsas, Enzo Ferrante, Sarah Parisot, Christian Ledig, Aditya Nori, Antonio Criminisi,
Daniel Rueckert, and Ben Glocker, “DeepMedic for Brain Tumor Segmentation”,
https://www.microsoft.com/en-us/research/wp-content/uploads/2017/02/ kamnitsas2016brats.pdf
[17]Tatiraju S, Mehta A. Image Segmentation using k means clustering, EM and normalizzed Cuts, University
of California, Irvine, technical report
[18]ParthaPratim Roy, Umapada Pal, Josep Lladós, "Morphology Based Handwritten Line Segmentation Using
Foreground and Background Information", Computer Vision Centre.