Journal of Brain and Neurology

Research Article - Journal of Brain and Neurology (2018) Volume 2, Issue 1

Alzheimer's stage classification using Support Vector Machines (SVM) and texture parameters of the brain MRI.

Shaik Basheera1*, Satya Sai Ram M2*

1Department of ECE, Acharya Nagarjuna University College of Engineering, Acharya Nagarjuna University, Guntur, India

2Department of ECE, Chalapathi Institute of Engineering and Technology, Lam, Guntur, India

*Corresponding Author:
Shaik Basheera
Department of ECE, Acharya Nagarjuna University, Guntur, India
E-mail: [email protected]

M Satya Sai Ram
Department of ECE, Chalapathi Institute of Engineering and Technology, Guntur, India
E-mail: [email protected]

Accepted on May 14, 2018

Citation: Basheera S, Ram MSS. Alzheimer’s stage classification using Support Vector Machines (SVM) and texture parameters of the brain MRI. J Brain Neurol 2018;2(1):19-25.

Visit for more related articles at Journal of Brain and Neurology


Medical imaging play major role in diagnosis of diseases, Machine learning play major role in diagnosis of medical images using computer-aided diagnosis. As India’s urban population is goon increasing Neurological disorders also increases, Alzheimer’s is one of the major dementia of neurons and make the death tally as high next to cancer. Estimating the stage of the Alzheimer’s is a challenging task. We use T2 Weighted Brain MRI and Extract the Texture Features from those images. Train the Classifier and perform crossover validation using those features. Support Vector Machines (SVM) give the good classification accuracy than comparing to the Naïve Bayes classifier and KNN. Test the classifiers with unknown images. The result is compared with Clinical information SVM gives 100% accuracy.


Support vector machines, Gray level co-occurrence matrix, Alzheimer’s, Dementia, Naïve bayes classifier, KNN.


The modern days working environment has changed due to globalization. Working stress is increasing on persons which cause neurological disorders such as dementia. Alzheimer’s disease is due to dementia it is a progressive neurological disorder. Generally, it appears at the age of 60’s. But due to the change in food habits and working environment it is appearing in young person’s below 40 years of age.

Alzheimer’s is the 3rd in major cause of death next to heart diseases and cancer. As India is the second largest in population and developing rapidly in economic. Change in the working Environment and food habits causes Alzheimer’s.

Memory loss is the major symptom of the Alzheimer’s. It is due to loss the connection between the nerves. White matter of the brain get decreases because of losing the nerve cell connection. Hippocampus volume reduction is observed as primary symptom of Alzheimer’s using MRI.

Radiologist diagnosis the disease using Imaging technology. But due to visual impairments, less frequent and uncharacterized imaging features, radiologist face problem to correctly diagnosis the disease.

Alzheimer’s changes the brain morphology. Figure 1 show the different stages of the Alzheimer’s disease affected MRI Images.


Figure 1. (a) Normal Stage, (b) Mild Stage, (c) Advance Stage.

The texture of the brain is different from one stage to another stage. We come to know that due to the Alzheimer’s brain get shrinks, enlarges its ventricles, change in white and gray matter volume, Change in Hippocampus size.

Existing Techniques

Biomarkers are used to evaluate the biological changes carried due to Alzheimer’s disease [1-5]. Which are used to measure β-amyloid, total tau and phospho-tau-181 in cerebrospinal fluid (CSF). This technique is the most acceptable method to diagnose AD with high specificity and sensitivity. But Biomarkers are not useful for early diagnosis of the disease. Moreover, it needs to use Intra cerebral ventricular injection.

To collect the CSF the clinical staff, need to take utmost care without damage brain tissues and spinal card. This is one of the most critical mechanism to estimate the Alzheimer’s where they estimate the tangle and plague of the brain tissues for Alzheimer’s analysis.

As CSF Biomarkers have remarkable drawbacks. Body Fluid Based Biomarkers are needed to diagnosis the Alzheimer’s [6]. Although saliva or urine can be easily collected, blood analysis is the gold standard it is still unknown how the concentration of analyses in the blood directly correlates with pathological changes in the brain, especially in AD.

Brain Imaging is used for different brain disorders such as Tumors, subdural hematomas and stroke but not used for severity of the AD [7-9]. Volumetric analysis is used by analyzing manually or semi-automatic techniques using SPM-5 [10] using MATLAB Environment. There the neurologist need to calculate the total volume of the different regions such as white matter, gray matter, CSF and sum together come to conclusion about the stage of the disease.

Proposed system

Morphological changes are appearing in Brain if the person is having Alzheimer’s disease. It means texture of the brain get change as the severity of the disease get increases. In primary stage hippocampus size get shrunk, plague and tangle get increases in the brain region.

In our proposed system, Alzheimer’s is classified using Support Vector Machine, by using texture parameters of the brain MRI. Gray level Co-occurrence matrix of the brain image is used to collect the texture information. The Block diagram of the proposed system is shown in the Figure 2.


Figure 2: Proposed system Block diagram.

To train the classifier 54 MRI Slice Images are used. The images are collected from Harvard Medical School. The images are pre-labeled as normal, mild, and advanced.


All the images are pre-processed by enhancing and stripping the redundant image information. Histogram Equalization is used to enhance the image. In order to strip the unwanted features such as skull, fat from the Head MRI Image we use a 3 × 3 Filter as shown in the Figure 3. It is designed by taking the Taylors series of exponential series into consideration.


Figure 3: Convolution filter.

We know and designed a 3 × 3 matrix. This produce the resultant image with edge are highlighted where the abrupt change of intensity carried.

f(x,y) is the input image pixel intensity. C(a,b) is the Convolution filter cell with the particular intensity. Convolution is performed between input image and filter using the equation (1).

g(x,y)=f(x,y) × c(a,b) (1)

g(x,y) is the resultant image

Snake active contour is used to segment and form a mask M(x,y) it is a binary image. By performing Logical AND operation brain image is extracted and remove the unwanted features from the Head MRI image. Figure 4 shows the original image and resultant Skull Stripped image.


Figure 4: Original image and skull stripped image.

Feature extraction

Gray scale co-occurrence matrix (GLCM) is used for feature extraction. The matrix that gives the probability successive pixels of same intensity in the image. It is a 256 × 256 Square Matrix. Figure 5 shows the idea of generating GLCM


Figure 5: Generating GLCM Matrix from the original image.

As the Figure 5a represent the original gray scale image and verifies the pixel intensities in horizontal direction and write the number representation in the GLCM Matrix.

The procedure continues for all the pixels and note down how many times the same pixels are come side by side based on this we calculate the probability used to find the texture parameters of the image.

Features and database generation

The GLCM used to generate statistical texture parameters such as Mean in x direction, Mean in y direction, Correlation, homogeneity, Energy, Entropy, standard deviation in x direction, standard deviation in y direction, Angular secondary moment, variance, Cluster shade, inertia, skew and skew coefficient etc. All these features are used to generate the required a data base. This data base is used to train and validate the Classifiers.


We use Classifier to bifurcate the stage of Alzheimer’s disease based on the features that extracted by using GLCM, the features are stored in data base. We use Support vector classifier having its path in statistics give good results even having small data set. SVM is used as Binary classifier used to separate two classes. But our data set is having multi classification for this one versus all (OVS) or One to One mapping method is used.

Let the training set is having x data set points as (a1,b1)……… ……..(an,bn).

Where equation is the p dimensional feature vector equation represent the class that be labeled to train the classifier. A hyper plane with great variance represented with a real vector equation which is perpendicular to the hyper plane and a offset parameter p is used to make the margin equation

If the offset become zero, then the hyper plane pass through the origin.

As the SVM is Risk minimizing algorithm it is interested in maximum margin between two classifiers. Two parallel planes are taken if the data set is lineally separable and no data is placed in-between the hyper plane.

As the SVM is binary classifier then we can apply this algorithm by considering one versus all or one to one mapping.

In one versus all algorithm one class data is taken as +ve and remaining all classes are labeled as –ve and train the classifier and repeat this procedure for all the class. The algorithm is carried using the given formula.

equation the given data set is belonging the bi=1 class.

≤ -1, the given data set is belonging the bi=-1 class.

equationuse as a classifier kernel.

Hθ(a) Let we have three classes as normal, medium, advance impairments as i={1,2,3} then find the probability of getting the given data set belong to that class as Hθi(a)=p(i/a;w) by using this the probability and using maximum value of Hθi(a) we can come to knowledge that the data set belong to the particular class.


We are taken 54 MRI Slices data base to train our classifier. Which are already labeled as Normal, Mild and Advanced. These images are collected from Harvard medical school. They are total 54 images 21 images are Normal, 11 Images are Mild, and 22 images are advanced images.


Pre-processed images are used to generate GLCM. Using GLCM Statistical Texture Features are Extracted. Totally 21 parameters are extracted for each image and tabulated in Microsoft excel and convert into .CSV file as shown in Figure 6.


Figure 6: Data base in .CSV file.

Our data set is having multiclass, so we compare the performance of the SVM with different Multiclass classifiers such as Navy baye’s classifier, KNN.

To compare the performance of the classifiers, we use 5 folded Cross validation and generate confusion matrix of KNN, Navy Baye’s Classifer and SVM as shown in Figure 7. Using the Respective Confusion Matrix different parameters are calculated such as Classification accuracy, F1 measure, Precision, Recall and Area under Curve (AUC).


Figure 7: Confusion Matrix of individual classifier (a) SVM (b) Naive based (c) KNN Classifier.

From the Table 1 all the classifiers are trained by folded 5 cross validation approach and we conclude from the parameters that SVM gives good AUC and Accuracy F measurement precision and Recall.

Classifier AUC CA F1 Precision Recall
SVM 1 0.981 0.951 0.971 1
KNN 0.983 0.981 0.957 0.971 1
Naive Bays Classifier 1 0.907 0.815 0.688 1

Table 1: Statistical Parameters of Training Classifiers.

And the classifier is also test using different unknown images collected from Internet source and verified using our algorithm. Unlabelled images are shown in Figure 8.


Figure 8: Unlabelled Test Images.

On testing SVM gives perfect prediction of the disease the metrics are given in Table 2. Table 3 Gives the Clinical Information of Unlabelled images based on Internet sources.

  Naive Bayes Classifier KNN Classifier SVM Classifier
Image Advance Mild Normal Status Advance Mild Normal Status Advance Mild Normal Status
Image a 0 1 0 M 0 1 0 M 0.06 0.74 0 M
Image b 0 0 1 N 0 0 1 N 0 0 1 N
Image c 0 0 1 N 0 0 1 N 0 0 1 N
Image d 0.03 0.97 0 M 0.2 0.8 0 M 0.2 0.8 0 A
Image e 0.99 0.01 0 A 0.4 0.6 0 M 0.52 0.46 0.01 A

Table 2: Prediction of the disease. **Status: N-Normal, M-Mild, A-Advanced.

Image Stage
Image a Mild
Image b Normal
Image c Normal
Image d Advanced
Image e Advanced

Table 3: Clinical Information of Un labelled images.


We conclude that the Alzheimer’s stages are get classified perfectly by the texture attributes collected from the brain using SVM classifier. The predicted results are compared with the clinical information. The Proposed method gives 100% classification accuracy. This can further be expanded to the Schizophrenia as the texture information of the brain is also play the main role in the classification.