International Journal of Computer Trends and Technology (IJCTT) Volume 43 Number 2 January 2017

Overview on segmentation and classification

for the Alzheimers disease detection from
brain MRI
Kajal Kiran Gulhare#1, S. P. Shukla*2, L. K. Sharma#3
1
Dr. C. V. Raman University, Kota, Bilasupur, INDIA
2
Dr. Bhilai Institute of Technology, Durg, INDIA
3
National Institute of Occupational Health, Ahmedabad, INDIA

Abstract The several studies projected that important task in MRI data analysis for the AD
approximately 115 million people will be affected detection. Image segmentation is intended to
from Alzheimer disease (AD) worldwide by the year partition images into well-defined regions, where
2050. Early detection of AD is important so that each region is a set of pixels that share the same
preventative measures can be taken place. The range of intensities, the same texture or the same
human brain Magnetic resonance imaging (MRI) neighbourhood. The purpose of segmenting images
data have been used to detection of AD. Due to the is to remove unwanted information in order to locate
variation and complexity of brain tissue the MRI meaningful objects from the processed images. The
data analysis for detection AD is considered as classification is used to produce meaningful patterns
difficult process. The objective of this study is to from raw data, classify them into different groups
explore the recent published segmentation and based on their characteristics and predict new
classification techniques and discuss the usability in patterns based on previous knowledge. Due to the
AD detection of the human brain MRI data. advancement of technology in the imaging the
various techniques for image segmentation and
Keywords Alzheimer disease, MRI Data, classification have been reported. The objective of
Medical Image, Segmentation, Classification. this paper is to study the recent published
segmentation and classification techniques and
I. INTRODUCTION discuss the usability in AD detection of the human
The worlds elder population comprises nearly brain MRI data.
900 million people [1]. As per the census 2011, The organization of the paper is as follows.
India is home to more than103 million people aged Section 2 presents the methodology for required to
60 years and older [2]. The chronic diseases become analyse and predict the AD from the MRI, Section
more prevalent with the increase of age and a trend 3 reviewing on the existing on the recent
exacerbated by changes towards lifestyles and segmentation and classification techniques applied
behaviours that predispose towards them [1]. for the AD detection, conclusion and future work
Alzheimers disease (AD) is globally recognized as will be reported in the section 4.
the most common form of dementia affecting seniors
age 65 and over [3] [4]. The several studies II. METHODOLOGY
projected that approximately 115 million people will The process of applying to analyse and classify
be affected worldwide by the year 2050 [3]. AD the Alzheimers disease is depicted in Fig. 1. The
causes nerve cell death and tissue loss throughout first step is collecting or access the MRI data. There
the brain, resulting to brain tissue shrinking and are several organizations which provide access the
larger ventricles. When AD is suspected, the MRI data of Alzheimer Disease for the research
diagnosis is first confirmed with behavioural purpose. The researchers are using the Alzheimers
assessments and cognitive tests and often followed disease Neuroimaging Initiative Database (ADNI) [6]
by a brain scan which can show whether certain [7] and Open Access Series of Imaging Studies
changes have taken place in the brain [3]. There are (OASIS) Data set [8]. The second step is data
a number of different types of brain scan. The most preparation and data pre-processing. In this step the
widely used are computerised tomography (CT) and non relevant information is required to remove and
magnetic resonance imaging (MRI) [5]. the data is reoriented for easier interpretation. In
Medical imaging is the technique and process of Step 3, image segmentation is performed on the pre-
creating visual representations of the interior of a processed 3D MRI Neuro-imaging brain data using
body for clinical analysis and medical intervention. different techniques in order to extract the
Medical image processing and machine learning ventricles area. The image segmentation techniques
tools can help neurologists in assessing whether a are classified in various categories on the basis of
subject is developing the Alzheimer disease. The their process in the image processing. The
image segmentation and classification is an thresholding Edge detection, Region growing,

ISSN: 2231-2803 http://www.ijcttjournal.org Page 130

 International Journal of Computer Trends and Technology (IJCTT) Volume 43 Number 2 January 2017

watershed, texture image segmentation techniques sMRI data from 49 cognitive normal and 49
were utilized for the MRI data segmentation. The patients subjects. They reported that both gray
attributes are selected and In Step 4, attribute matter (GM) and white matter (WM) carry useful
extraction, such as surface area, centre of gravity, information for classification of cognitively normal
average intensity and standard deviation in order to control and Alzheimer disease sMRI images,
analyse the shape of the ventricle. The producing high levels of accuracy, sensitivity, and
classification/prediction method in order to assess specificity.
whether the patient is developing the Alzheimers Sweety and Jiji [11] used particle swarm
disease (AD) is performed by utilizing the machine optimization and decision tree for the AD detection.
learning algorithms. In this regards, suitable machine Features such as eigen brain, eigen vectors, mean,
learning algorithm is selected and training of standard deviation, variance, skewness, kurtosis,
algorithm is performed for model construction. The area, perimeter, eccentricity were calculated from
trained model is evaluated with the test data set. This MRI Images and quantitative measures are derived.
process is performed until the acceptable result is not Zhang et al. [12] proposed eigenbrain and
obtained. machine learning method for the detection of AD
using 3D MRI scan data. This method achieved high
accuracy 92.360.94 and they reported eigenbrain
method is effective in AD subject prediction and
discriminant brain region detection in MRI scanning.
Lorenzi et al. [13] proposed statistical modelling
of non-local intensity correlations for the multimodal
image analysis in AD. The results reported that non-
local approach better than classical PM-based
multimodal local correlation models in terms of
modelling accuracy and predictive power. The
ensemble of the reported results proves the ability of
the proposed PLSR in capturing biologically
relevant features, and in generalising to unseen
structural imaging data of T1-MR scans.
Beheshti and Demirel [14] proposed probability
distribution function (PDF) based classification for
the AD detection. This study deals with statistical
patterns extracted from structural MRI (sMRI) data
on four systematic levels; voxel-based morphometric
(VBM) technique with 3-Tesla 3D T1-weighted
MRI for global and local differences of gray matter,
feature extraction based on the voxel clusters
detected, utilized the PDF of the VOI to represent
statistical patterns of the respective high-
dimensional sMRI sample and SVM classifier was
applied.
Zang et al. [15] proposed a landmark based
Fig. 1 The process of system implementation feature extraction method and used support vector
machine (SVM) classifier for fast AD diagnosis.
III. SEGMENTATION AND CLASSIFICATION This technique does not require nonlinear
TECHNIQUES FOR ALZHEIMERS DISEASE registration and tissue segmentation. 85% accuracy
for the AD classification is reported in this study.
Yang et al. [9] used independent component
Sarwinda and Bustaman [16] introduced as a 2D
analysis based classification of Alzheimers MRI
and 3D feature extraction descriptor. The principal
data. The normalized brain images are decomposed
component analysis (PCA) and factor analysis are
into MRI basis functions and the corresponding
used as feature selection and SVM classifier was
coefficients using the FastICA algorithm and then,
used for the classification.
the separated coefficients are fed into an SVM-based
Beheshti et al [17] applied feature ranking and
classifier for diagnosis of individuals with or without
classification error for the detection of AD from
AD.
Structural MRI. The voxel based morphometry
Casanova et al. [10] illustrated using sMRI data
technique was used to compare the gray matter of
from the Alzheimer Disease Neuroimaging Initiative
AD patients and health control. The raw features are
(ADNI) clinical database and applied large scale
ranked using statistical dependency (SD),
regularization approach based on penalized logistic
information gain (IG), mutual information (MI),
regression to automatically classify sMRI according
Pearsons correlation coefficient (PCC), t-test score
to cognitive status. They performed their study on

ISSN: 2231-2803 http://www.ijcttjournal.org Page 131

 International Journal of Computer Trends and Technology (IJCTT) Volume 43 Number 2 January 2017

(TS), Fisher's criterion (FC), and the Gini index (GI). [6] O. B. Ahmed et al., Alzheimer's disease diagnosis on
structural MR images using circular harmonic functions
The SVM method was utilized for the classification. descriptors on hippocampus and posterior cingulate cortex,
Computerized Medical Imaging and Graphics, vol. 44, pp.
IV. CONCLUSIONS 1325, September 2015.
With the advance of computational intelligence [7] L. Lillemark et al. Brain regions relative proximity as
marker for Alzheimers disease based on structural MRI,
and machine learning techniques, computer-aided BMC Medical Imaging, vol. 14, pp. 21, 2014.
detection attracts more attention for AD detection. It [8] D. S. Marcus et al. Open Access Series of Imaging
has become one of the major research subjects in Studies (OASIS): cross-sectional MRI data in young,
medical imaging. In this study, we reviewed current middle aged, nondemented, and demented older adults, J
Cogn Neurosci., vol. 19(9), pp. 1498-1507, 2007.
studies of the different segmentation, feature [9] Yang et al., Independent component analysis based
extraction and classification algorithms. classification of Alzheimers MRI data, J Alzheimers Dis.,
Future research could focus on investigating other 24(4), pp. 775783, 2011. doi:10.3233/JAD-2011-101371.
regions that might be more affected, coupled with [10] R. Casanova et al., High dimensional classification of
structural MRI Alzheimers disease data based on large
appropriate features set in order to characterize the scale regularization, Frontiers in Neuroinformatics, vol.
new regions. It is suggested to focus on reducing the 5(22), 2011. doi: 10.3389/fninf.2011.00022
cost and improving the precision of the algorithms [11] M. E. Sweety and G. W. Jiji, Detection of Alzheimer
by using more intelligent algorithms such as disease in brain images using PSO and Decision Tree
Approach, International Conference on Advanced
adaptive seeds initialization and image registration Communication Control and Computing Technologies
in order to get initial contour for the active contour (ICACCCT), 2014. DOI:
segmentation method. It is also suggested to focus 10.1109/ICACCCT.2014.7019310
on the enhancement of the classification algorithms [12] Zhang et al., Detection of subjects and brain regions
related to Alzheimers disease using 3D MRI scans based
and adding more data to the system. on eigenbrain and machine learning, Frontiers in
Computational Neuroscience, 9 Article 66, 2015. doi:
REFERENCES 10.3389/fncom.2015.00066
[1] World Alzheimer Report 2015, "The Global Impact of [13] M. Lorenzi et al., Multimodal Image Analysis in
Dementia An analysis of prevalence, incidence, cost and Alzheimers disease via Statistical Modelling of Non-local
trends", Alzheimers disease International (ADI), London. Intensity Correlations, Nature Scientific Reports, 6:22161,
August 2015. 2015. Doi 10.1038/srep22161.
[2] Census of India-2011, office of the Registrar General & [14] I. Behesthi and H. Demirel, Probability distribution
Census Commissioner, India, Govt. of India, function-based classification of structural MRI for the
http://www.censusindia.gov.in/2011census/Age_level_Dat detection of Alzheimers disease, Computers in Biology
a/India/Age_data.xls [Last access 18/12/2016] and Medicine, vol. 64, pp. 208-216, 2015.
[3] N. T. Aggarwal, R. C. Shah, and D. A. Bennett, [15] J. Zang et al., Detecting Anatomical Landmarks for Fast
Alzheimers disease: Unique markers for diagnosis & new Alzheimers Disease Diagnosis, IEEE Transactions on
treatment modalities, Indian J Med Res 142, pp 369-382, Medical Imaging, vol. 35(15), pp. 2524-2533, 2016.
October 2015. [16] D. Sarwinda and A. Bustamam, Detection of Alzheimer's
[4] S. Matoug, Predicting Alzheimers disease by Segmenting disease using advanced local binary pattern from
and Classifying 3D-Brain MRI Images Using Clustering hippocampus and whole brain of MR images,
Technique and SVM Classifiers, Laurentian University International Joint Conference on Neural Networks
Sudbury, Ontario, Canada, 2015. (IJCNN), 2016, DOI: 10.1109/IJCNN.2016.7727865
[5] Fact Sheet What is Alzheimers disease? Alzheimers [17] I. Beheshti et al., Structural MRI-based detection of
Society, England, 2014. Alzheimer's disease using feature ranking and classification
https://www.alzheimers.org.uk/site/scripts/documents_info. error, Computer Methods and Program in Biomedicine,
php?documentID=100 [Last access 18/12/2016]. 137, pp. 177-193, 2016.

ISSN: 2231-2803 http://www.ijcttjournal.org Page 132