Alzheimer Disease Diagnosis using the K-means , GLCM and K _ NN Wed

Investigation of medical images have major consequence in the field of treatment.in this work ,MR images have been used to distinguish the normal brain from brain with Alzheimer disease .Texture is an native property of all surfaces it contains important facts about the structural organization of the surfaces and their connections neighboring area. In direction to classify texture must be segmented into a number of section that has the similar properties, for this purpose we used kmeans algorithm with GLCM for feature extraction ,finally we used k-nearest neighbor algorithm to distinguish between normal and abnormal brain. Keyword: Normal brain, Alzheimer disease, kmeans algorithm, GLCM, k-nearest neighbor algorithm. ةصلاخلا لا ثحب ةيبطلا روصلا يف ريبكلا رثلاا هل يف .جلاعلا لقح يف غامدلا زييمتل يسيطانغملا نينرلا روص مادختسا مت ،ثحبلا اذه لا يعيبط غامدلا نم باصملا ب . رمياهزلا ضرم (جيسنلا texture ) وه لا لأا ةيصاخ ه م ثيح يمكيهلا ميظنتلا لوح ةماه قئاقح ىمع يوتحي ا لال حطس هتلاصو قطنملاب ةرواجملا جيسنمل يف . فينصت لقح لا جيسن نأ بجي ي نوك همسقم لا نم ددع ىلإ قطانم لا يت كراشتت صئاصخلا ، ةيمزراوخ انمدختسا ضرغلا اذهل kmeans عم GLCM جا رختسلا صئاصخلا ةيمزراوخ انمدختسا ا ريخأو ، k-nearest neighbor يعيبط ريغو يعيبط غامدلا نيب زييمتمل . كلا : ةيحاتفملا تامم باسحلا طسوتم ةيمزراوخ , برقلأا راجلا ةيمزراوخ , رمياهزلا ضرم , يعيبطلا غامدلا Introduction Images are reflected the vital middle in transmission information. A key phase of Machine learning were studying images and extracting the information from them, which can be used for other tasks. The initial phases in track of studying images is the segmentation, which the way to finds the different objects in them, segmentation is nothing but pixel classification. Image segmentation requires the separation or division of the image into parts of similar attributes. For MRI grouping using k-means algorithm is useful. Isolated the object and contextual clear regardless the MRI has cloudiness borderline, is the key of segmentation (Ahsan et al., 2012). Texture analysis is very useful, applied almost any digital image. If the spatial level of the MRI brain can be identified by an independent means, then the application of texture analysis can be limited to a set of predefined areas of attention .GLCM based texture analysis has become a common feature extraction method for the detection and classification masses and micro calcifications in digital image(Ahsan et al., 2012). The classification of textural features are related to a radiologist’s clinical analysis and contains dividing the streamlined feature space according to tissue class or diagnostic category (Liu et al., 2012; Kassner et al., 2010). Journal of Babylon University/Pure and Applied Sciences/ No.(2)/ Vol.(26): 2018 58 (Imad Zyout et al., 2011) classified MC clusters by PSO-KNN and GLCM features inserted feature selection approach. Addition of the PSO-KNN methodology to other feature spaces and confirming the results of this work using a greater dataset of mammograms are planned as a current and future work of this paper. Results of exploratory the relation between the size of the mammographic regions that used to calculate GLCM features and the discriminative power of GLCM features indicated the positive impact of the texture close to MC clusters. (Ana Simões, 2013) the thesis propose methods to help diagnose Alzheimer disease at an initial stage of advance. The results of the three methodologies that used in this thesis show that: first, texture descriptors are able to reach high classification degrees, equivalent to structural-based features; second, using limited spots above the whole brain, no expectations must be made about the expectedly affected brain sections, and thus no earlier segmentations are necessary; third, confine discriminative brain sections using exceptionally sampled patches in the brain, by affine-registering the images only. Computer assisted diagnosis is design using a k-NN classifier in the (Papakostas et al. ,2015), present a Computer Assisted Diagnosis system for Alzheimer’s disease. The system using the data of the MRI features and using a Lattice Computing method, Lattice Computing setting by treatment this mission by two different views; first, it does dimensionality lessening on the tall dimensional feature vectors. Second, by creating adaptive class boundaries it classifies the issues in the lattice region. Calculation testing using a standard MRI dataset around Alzheimer’s disease patient’s appearance that the obtainable classifier does sound reasonably to state-of-the-art classification prototypes. (Han et al., 2001) proposed way for clustering. Representing data by less clusters necessarily misses certain many data object by little clusters and therefore, it representations data by its clusters. An Improved K-means Clustering Algorithm proposed by (Chunfei Zhang et al., 2013). The idea of the K-means clustering algorithm analysis the advantages and disadvantages of the traditional K-means clustering algorithm, the traditional K-means algorithm is a widely used clustering algorithm, with a wide range of applications. Elaborates the method of improving the K-means clustering algorithm based on improve the initial focal point and determine the K value. 2Methodology In this research, the K-means algorithm used as pre-processing image by segmentation the MR image of brain into several cluster based on their inherent space after each other. For texture classification and feature extraction, GLCM has a good base, and at last the k_ nearest neighbor used for classify the MRI in to normal and abnormal brain, as it explain in the figure (1). Journal of Babylon University/Pure and Applied Sciences/ No.(2)/ Vol.(26): 2018 59 Figure (1) the methodology of the proposed system 2.1 K-means Clustering Algorithm Image segmentation is a significant apparatus in image processing which is classing an image into pixels that is homogeneous with respect to some standard. K-Means is an unsupervised clustering algorithm that used to processing the input data points of the MRI by organize it (Kassne et al., 2010). The method work of the algorithm stands by take the data features form a vector space and work on find natural group for it. The points are clustered about centroids, the center of a cluster for the k-means algorithm is the mean point of all points in the cluster, which are found by minimizing the objective, Where there are k =4, where each MRI segment to four areas (Liu et.al., 2012). The algorithm takes as a input 2D MRI. Steps of the algorithm are as follows: Input: 2D test MRI, 2D train MRI Output: clustering2D test MRI, clustring2D train MRI. Begin For each object in 2D test M I and 2D train MRI do Test MRI Train MRI


Introduction
Images are reflected the vital middle in transmission information.A key phase of Machine learning were studying images and extracting the information from them, which can be used for other tasks.The initial phases in track of studying images is the segmentation, which the way to finds the different objects in them, segmentation is nothing but pixel classification.Image segmentation requires the separation or division of the image into parts of similar attributes.For MRI grouping using k-means algorithm is useful.Isolated the object and contextual clear regardless the MRI has cloudiness borderline, is the key of segmentation (Ahsan et al., 2012).Texture analysis is very useful, applied almost any digital image.If the spatial level of the MRI brain can be identified by an independent means, then the application of texture analysis can be limited to a set of predefined areas of attention .GLCM based texture analysis has become a common feature extraction method for the detection and classification masses and micro calcifications in digital image (Ahsan et al., 2012).The classification of textural features are related to a radiologist's clinical analysis and contains dividing the streamlined feature space according to tissue class or diagnostic category (Liu et al., 2012;Kassner et al., 2010).(Imad Zyout et al., 2011) classified MC clusters by PSO-KNN and GLCM features inserted feature selection approach.Addition of the PSO-KNN methodology to other feature spaces and confirming the results of this work using a greater dataset of mammograms are planned as a current and future work of this paper.Results of exploratory the relation between the size of the mammographic regions that used to calculate GLCM features and the discriminative power of GLCM features indicated the positive impact of the texture close to MC clusters.(Ana Simões, 2013) the thesis propose methods to help diagnose Alzheimer disease at an initial stage of advance.The results of the three methodologies that used in this thesis show that: first, texture descriptors are able to reach high classification degrees, equivalent to structural-based features; second, using limited spots above the whole brain, no expectations must be made about the expectedly affected brain sections, and thus no earlier segmentations are necessary; third, confine discriminative brain sections using exceptionally sampled patches in the brain, by affine-registering the images only.
Computer assisted diagnosis is design using a k-NN classifier in the (Papakostas et al. ,2015), present a Computer Assisted Diagnosis system for Alzheimer's disease.The system using the data of the MRI features and using a Lattice Computing method, Lattice Computing setting by treatment this mission by two different views; first, it does dimensionality lessening on the tall dimensional feature vectors.Second, by creating adaptive class boundaries it classifies the issues in the lattice region.Calculation testing using a standard MRI dataset around Alzheimer's disease patient's appearance that the obtainable classifier does sound reasonably to state-of-the-art classification prototypes.(Han et al., 2001) proposed way for clustering.Representing data by less clusters necessarily misses certain many data object by little clusters and therefore, it representations data by its clusters.
An Improved K-means Clustering Algorithm proposed by (Chunfei Zhang et al., 2013).The idea of the K-means clustering algorithm analysis the advantages and disadvantages of the traditional K-means clustering algorithm, the traditional K-means algorithm is a widely used clustering algorithm, with a wide range of applications.Elaborates the method of improving the K-means clustering algorithm based on improve the initial focal point and determine the K value.

2-Methodology
In this research, the K-means algorithm used as pre-processing image by segmentation the MR image of brain into several cluster based on their inherent space after each other.For texture classification and feature extraction, GLCM has a good base, and at last the k_ nearest neighbor used for classify the MRI in to normal and abnormal brain, as it explain in the figure (1).

K-means Clustering Algorithm
Image segmentation is a significant apparatus in image processing which is classing an image into pixels that is homogeneous with respect to some standard.K-Means is an unsupervised clustering algorithm that used to processing the input data points of the MRI by organize it (Kassne et al., 2010).
The method work of the algorithm stands by take the data features form a vector space and work on find natural group for it.The points are clustered about centroids, the center of a cluster for the k-means algorithm is the mean point of all points in the cluster, which are found by minimizing the objective, Where there are k =4, where each MRI segment to four areas (Liu et.al., 2012) 3.Each object (point) assign to the group according to the adjacent center.4. Stop when no more change in the center, otherwise, go back to step 2. End End Primary cluster centers effect on the classification consequences significantly.Oppositely, clustering algorithm use the data of color for pixels to segment images, but do not consider the spatial data.Therefore, it is delicate to noise and result in concludedsegmentation (Kassner et al., 2010), As shown in figure (2).

2.2Gray level co-occurrence matrix (GLCM):
To aid understand the facts and the more information about the complete image there are several textural factors compute by the GLCM.Texture occurs to be a key characteristic to the mechanical or semi-mechanical analysis of digital images.GLCM can be define as one of statistical methods that reflect the second order histograms to evaluation the dual probability of a gray level pixel for a one pixel spacing and path.Used statistical features the minimum correlated and those maximum recurrently were computed, by the GLCMs (Kassner et al., 2010;Everitt et al., 2011).
In this research eight features were calculate in each direction .thefirst step of the image processor was the gray levels of the MRI s were change to (0-255) gray level values and then compute the texture features for four theta directions (0-45-90-135) and one pixel distance (1).When the neighbor pixel was not similar, the pixels pairs were unaccepted by the GLCM.
Using the normalized GLCM matrix (Cnorm(i,j)), the texture features are computed as follows (Kassner et al., 2010).: Step 1-For I = 0 to 255(gray level) For j =0 to 255 (gray level) Begin 1-Sum all pairs of pixels in which the first pixel has a value i, and the second has a value of j; and distance between them is (d=1).

This Sum is saved in the ith row and jth column of the matrix G[i,j] End
Step 2-Each elements of G[i,j]can be normalized by dividing it by the summation of total element value in the matrix G; to get Normalized GLCM N[i,j], Step

K-Nearest Neighbor (KNN):
The KNN is kind of a non-parametric algorithm that used to classification and regression.The result of the algorithm rest on whether is used to classification or regression.Generally, the input consists of the k closest training patterns in the feature space (Kassner et al., 2010).One of a data mining algorithm with a comprehensive choice of field image processing application is the KNN algorithm several components of this approach must be taken in account: 1-A set of training and test patterns.2-Calculate the distance between the training set and the test pattern.
3-The value of k (k is a positive integer, typically small).
The object is just goes to the class of that single nearest neighbor if k = 1.In this search, the Euclidean distance used to classify the testing set examples from the two classes ,were we compute Euclidean distance for each sample in the MRI test for all MRI train .theEuclidean distance expressed mathematically as, (Kassner et al., 2010;Everitt et al., 2011;Altman et al., 1992).Euclidean distance = √ (∑ n i=1 (x iyi) 2 )……… (2.9) In this work, according to the test the KNN algorithm the best result was, when the value of k=3, were we take the first three of Euclidean distance after the ascending Order for the distance.

The data set used
The MR Image of Brain that used in this study taken from Open Access Series of Imaging Studies (OASIS) database (Daniel et al. 2007)..inthis research we used 40 sample MRI mixed of normal brain and brain with Alzheimer disease, and other 30 MRI for the test stage.A sample of MRI which is used in this research shown in Fig (5).each of MRI is convert to bmp file with size 100*100 pixel , the first step in the proposed system was convert each MRI to gray level (0-255) (Kekre et al., 2009) .

4-Conclusion
This paper has presented a supervised classification algorithm (KNN) for MRI for brain with Alzheimer disease.An effective segmentation method was proposed using Kmean algorithm for clustering.The GLCM was used for feature extract finally we used knearest neighbor algorithm to distinguish between normal and abnormal brain .Experimental results of MR images show that the estimation of clusters and classified performs well in Gray level feature spaces.The experimental results have shown the advantage of the system in the field of medical diagnosis, the accuracy of the system was 86.6%, the results of the search acceptable according to the accuracy and computational speed of KNN.

Figure ( 4 )
Figure (4): the K-NN to find the class of test MRI and compute accuracy of the algorithm . The algorithm takes as a input 2D MRI.Steps of the algorithm are as follows: The objects Partition into k=4 not empty subgroups 2. Calculate centers of the each group.