• Nu S-Au Găsit Rezultate

View of Pre-Processing on Alzheimer MRI images

N/A
N/A
Protected

Academic year: 2022

Share "View of Pre-Processing on Alzheimer MRI images"

Copied!
9
0
0

Text complet

(1)

Pre-Processing on Alzheimer MRI images

Bharathi. A, A.S.Arunachalam Research Scholar Associate Professor

Department of Computer Science , School of Computing Sciences,

Vels Institute of Science,Technology and Advanced Studies( VISTAS), Chennai, India [email protected], [email protected]

Abstract

Health issues are common developing factor in the recent era. Brain is an important part of the body which activates the human functionalities. There are many health issues related to brain, Alzheimerdisease is one among them. If brain is dead it cannot be reversed. So, each distortion of the brain ought to be grant significance. Early diagnosis helps to reduce and avoid death rate due to Alzheimer. There are many medical diagnosis methods applied to provide medication on this issue. Computing algorithm plays a vital role in giving assistance to physicians in the diagnosis of Alzheimer disease. Objective of the work is detection ofAlzheimer disease from MRI data set. In this paper, the Alzheimer image is implemented with pre-processing techniques to reduce noise in the image which leads to the next level. In the pre-processing, filtering techniques are used to blur, and smoothen the image without changing the pixel values. Filtering techniques like Arithmetic mean, median, Gaussian, bilateral are used to remove the din in the image and Fast FourierTransform (FFT) is used to find the frequency distribution in image. Before pre-processing the image, the distribution of pixels is determined using histogram distribution for both normal and three types of Alzheimer images are Mild demented, moderate demented, very mild demented. The pixels values before pre-processing and after applying pre-processing are compared. The comparison shows the outliers removed in the types of Alzheimer image. The accuracy has proved the efficiency of the techniques used for this work.

Keywords: Digital image pre-processing, Filtering techniques, Histogram equalization, Fast Fourier transform, Alzheimer image.

I. INTRODUCTION

Human beings are the most beautiful creatures in this world. For long life human beings ought to watch out for their wellbeing. Brain is an important part of the body that sends signals to the nerves and to the other parts of the body [1]. Even human actions and emotions are controlled by this brain. Alzheimer disease is the deformity which occur due to the age factor, hyper stress and trauma. This may lead to loss of sensation, paralysis, stops the daily task, and stops the thinking skill, even it will cause death if it is not diagnosed well in advance. Early diagnosis and the medication will help to reduce the problems due to this deformity [2].

There are many tests available to find the Alzheimer disorders. Some of them are memory test, neurological test for functionality, CT and MRI. The MRI shows the clear image of the objects in the image with its deviations [3]. There are three types of Alzheimer diseases. They are mild dementia, moderate dementia and very mild dementia. Very mild is the starting stage of Alzheimer disease that occurs due to age factors. People with this kind will forget the familiar words and names. Next is the mild Alzheimer disease which leads the memory lapses, struggle in daily activities.Moderate Alzheimer disease will lead to the sever stage. This stage needs assistance to do the daily activates. So,decision should be takenat the earlystage throughMRI to avoid complications. There are many ongoing researches to solve this issue. This research attempts to find the Alzheimer disorder in the MRI brain image.

Computing algorithms are performed well in different field. It plays a vibrant role in medication [4]. It simplifies the diagnosis by the models and methods developed using its techniques. Digital image processing used for the medical image analysis. So from the images from different modalities are processed using the digital image techniques. Researches which provided support [5] to do this work are listed follows.

William R et al [2010] proposed the method instigated with the arithmetic mean to know the functionalities and structural measurements of the brain image. Ibrahim et al [2013] designed a neural network model to identify the tumour in the brain image. They have pre-processed the image using arithmetic mean and used PCA algorithm to extract the feature values. George et al [2012] enhanced their brain image using filtering techniques. They removed the skull and uniform the distribution of intensity by histogram equalization. Finally, it is applied with the enhanced median filter. senel et al [2002] tested the median filter they introduced which is not disturbed by the neighbourhood pixel changes. phophalia et al[2015] used bilateral filtering techniques to

(2)

4434

remove noisingin the brain MRI. Mustafa et al [2011] used bilateral filtering techniques to remove the noise.

This technique is enhanced filter multiresolution bilateral fileting produced the better result.

Herholz et al [2002] gave a comparison study to discover the Alzheimer disease between PET and SPECT scan.

Gaussian filtering techniques is used to smoothen the image. Chaddad et al [2016] used the laplician of Gaussian filter which was used to remove the noise and smoothen the Alzheimer image. They achieved the accuracy above 70%. They have extracted the identified the features, brain, texture feature. Saeed et al [2016]

used the Fourier transformation to identify the tumor and leakage of cerebral fluid. This Fourier transformation is implemented with segmentation algorithm to identify the issues. They have achieved better results.Zahng et al [2015] used the wavelet with supporting the Fourier concept to identify the Alzheimer disease and the mild cognitive disorder. SVM, PCA are used to classify the mild dementia.

The above are the base papers that gave the key idea to use the filtering techniques for further enhancement. In this work, the research is aimed to detect the Alzheimer disease in the brain through MRI. To do that, all the images taken for the processing should be pre-processed. Because the MRI image has noise due to thermal effect and other effects.In the pre-processing, filtering techniques is used to blur, smoothen the image without disturbing the pixels of the images. It also saves the borders of the image. Arithmetic mean, median, Gaussian, bilateral are used to smoothen the image by preserving the edges of the images. Fast Fourier transformations is used to find the frequency of the image. Before pre-processing the histogram for image which give the pixel distribution is determined. This helps to know how the pixel values are distributed in the grey image. Finally, the pixels values of pre-processed image compared with its original image to find the accuracy of the image.

This entire paper is organised in five sections. First section talks about the introduction about the Alzheimer disease and role of the digital computing algorithms, explains the problem found in the existing research.

Second, Methodology describes how the work is carried out and explained with its schemas. The work flow diagram shows the step by step process of the work. Third section presents the dataset, fourthdisplays results and discussions which explain the entire work process with its results. Finally conclusion which summarizes the wok and its future enhancement.

II. METHODOLOGY

Methodology explains the design process of the work. There are five filters namely arithmetic mean, median, bilateral, Gaussian and fast Fourier transformations (FFT). The following figure 1 shows the block diagram of the implementation methodology.

The Alzheimer normal image and its three types namely very mild Alzheimer, mild Alzheimer, and moderate Alzheimer images are given as an input. Each filtering techniques applied on the data set are explained below.

a) Arithmetic mean: it is used to eliminate the identical noise in the image [].

𝑚𝑠𝑥𝑦 = 1

𝑚𝑛 (𝑠,𝑡)∈𝑠𝑥𝑦𝑔(𝑠, 𝑡) ……….. (1)

Where sxy is the coordinates of the window in rectangle shape. m*n are the window size. This mean is calculated based on the pixel values represented by s. average computed in g(x,y).

b) Median Filtering: it is used to remove the noise which are very strong. Itprovides the less blurs than smoothening the image [].

𝑦 𝑚, 𝑛 = 𝑚𝑒𝑑𝑖𝑎𝑛{𝑥 𝑖, 𝑗 , (𝑖, 𝑗) ∈ 𝑤}……….. (2)

Where m*n are the window size, i & j are the coordinates. W is the neighbourhood pixel of image.

c) Gaussian Filtering: It protects the edges of the image and smoothen the image by specifying the weight value. It is defined as sigma.

𝑮(𝒙, 𝒚) = 𝒆^ (−(𝒙^𝟐 + 𝒚^𝟐)/ (𝟐 ∗ 𝒔𝒊𝒈𝒎𝒂^𝟐)) ……….(3) Where, x and y are the pixel coordinates of the image.

d) Bilateralfiltering: It preserves edges by blurring the flatten region in the image. Weight is set as the pixels of centre value. It is compared with neighbourhood pixel [].

𝒈( 𝒙) = (𝒇 ∗ 𝐺 𝑠)( 𝒙) = 𝒇 𝒀 𝑹𝟎 𝐺 𝑠 𝑥 − 𝑦 𝑑𝑦 ……….(4) Where, ||xy||gives the spatial distance of G(xy) , weight is set in f(y)

e) Fast Fourier Transformation: It transfer the signal form in to frequency domain. It always carry with finding factorization []. It reduce the complexity in frequency transformation.

𝒙𝒏−𝒌= 𝒙𝒌……….………..(5)

Where,n is the size, and k =0, which defines the n output.

(3)

Input: MRI Alzheimer image

Output: Alzheimer pre-processed image Begin

Import cv2, numpy// file loaded for the implementation Define 1 to n images // total no. of images

Read image G// read images in jpeg format Convert dimensions 512*512 //image conversion

Computewindows m,n or i,j and pixel coordinates x,y , G ground truth image Calculate

𝑚𝑠𝑥𝑦 = 1

𝑚𝑛 (𝑠,𝑡)∈𝑠𝑥𝑦𝑔(𝑠, 𝑡) // arithmetic mean 𝑦 𝑚, 𝑛 = 𝑚𝑒𝑑𝑖𝑎𝑛{𝑥 𝑖, 𝑗 , (𝑖, 𝑗) ∈ 𝑤} // median

(𝑥, 𝑦) = 𝑒^ (−(𝑥^2 + 𝑦^2)/ (2 ∗ 𝑠𝑖𝑔𝑚𝑎^2)) // Gaussian 𝒈( 𝒙) = (𝒇 ∗ 𝐺 𝑠)( 𝒙) = 𝒇 𝒀 𝑹𝟎 𝐺 𝑠 𝑥 − 𝑦 𝑑𝑦//bilateral 𝒙𝒏−𝒌= 𝒙𝒌 // fast Fourier transformation based on kernel Store the output in folder (…n)// save near the input image

These are the above filtering techniques used to implement on the Alzheimerimage.

This can be explained in the following pseudocode.

Filtering Techniques (FT) Comparison of (FT)

Figure 1. Block diagram of pre-processing techniques forAlzheimer MRI image Table1. Pseudocode of pre-processing techniques

Arithmetic mean (AM)

FFT Median

Bilateral

Pixel values AM

Gaussian Filtering Pixels of median

Pixels of Bilateral

Pixels of FFT

Best FT

Input of Alzheimer MRI

image image

1.

Gaussian Pixels of Gaussian

(4)

4436

2.1. Dataset

The dataset taken for the implementation is Alzheimer MRI image. It is collected from different sources like hospitals, scan centres, and online repositories. The MRI dataset collected will be in the form of DICOM: digital imaging communication in medicine. To process this image it must be converted into .Jpeg image with the dimensions of 512*512. The MRI image has three types of plane such as Axial, sagittal, coronal. The axial view of the Alzheimer MRI is taken for the analysis. There are three types of Alzheimer disease image 1. Very mild Alzheimer dementia image (VM), 2. Mild dementia Alzheimer image (MD) 3. Moderate dementia Alzheimer image (MO). These three types and the normal Alzheimer images are processed in this work are given in figure 2.

A B C D

Figure 2. Input images. A) Normal image B) Very mild dementia C)Mild dementia D) Moderate dementia III. RESULT AND DISCUSSION

The Alzheimer image is pre-processed using filtering techniques to remove the noise, blur and smoothen the image by preserving the details. The implantation work and the result obtained are explained in three ways. First finding the pixels distribution in the image. Secondly, pre-processing the image. Third finding the difference between the pixels of different types of Alzheimer image to determine the accuracy. As per the discussion in dataset section, the DICOM is converted in to .jpeg and preserved the uniformity in the dimensionality as 512*512.

3.1. Histogram

The first step of the work is finding the pixels distributions in the Alzheimer image. The intensity range is lower to higher. The intensity of thegrey image starts from 0 to 255. The black is given in 0and the pure white will be in 1.The figure 3 shows the intensity values. The normal brain image is shown in A, the very mild dementia is in B, mild dementia is shown in C and moderate Alzheimer is shown in D. The variation in the intensity is marked in red colour rectangle box. From the analysis the Alzheimer disease has the little variation between them but it has high variation between the normal images. The threshold value of the abnormality image is determined based on the trial and error method. The range-based threshold set as 90 to 140 for the Alzheimer image tested with 50 images. The pixels distribution in the image is calculated by finding the black and white pixels count.

The difference of these pixel value shows exactly how much the nerves are compressedin each types of disease.

The normal data pixel distribution and the Alzheimer disease image pixel values are compared.

A B

(5)

C D

Figure 3. Histogram of dataset A) Normal image B) Very mild dementia C) Mild dementia D) Moderate dementia

3.2. Pre-processing

Processing the noisy image will culminate in bad result when the segmentation or the feature extraction are applied. Enhancement technique helps to achieve the best result when we do the identification and extraction.

There are many enhancement techniques available in digital image processing. Filtering techniqueis one among them which is used to sharpen the image, smoothen the image and blurs the noise, removes the high-density noise, provides density equalization.

In this work arithmetic mean (AM) which was used to remove the uniform noise, median was used to remove the strong noise in the image. Gaussian filter was used to remove the noise by changing the values sigma and it sanctuaries the facts of the image. The sigma value must be 1.0 for the Gaussian filtering on medical images.

The kernel with sigma value of Gaussian for Alzheimer image is 0.96 .This was attain by trial and error method.

Bilateral filtering works based on finding the spatial distance between the neighbourhood pixels using the tonal weight. And it smoothens the image. Fast Fourier transformations (FFT) are used to represent the frequency of the grey image. Where the spectrum in the image shows the high frequency. Corners shows the low frequencies.

All the pre-processing (pp) result is shown in figure 4.

PP Original image VM MD MO

Gaussian

Bilateral

(6)

4438

FFT

PP Original image VM MD MO

AM

Median

Figure 4. Pre-processing filtering techniques for Alzheimer images

IV. SELECTING THE BEST FILTER BY COMPARISON ANALYSIS

The best filter is selected based on the comparison analysis. The pixels of the image from Alzheimer ground truth image and the pre-processed images are computed. Find the variance between the processed and unprocessed images. The difference shows the noise removed without changing the pixels in the image.The values derived are given in the table 2.

Table 2. Comparison of pixels in the Alzheimer’s image

NI MD AM M B G R1 R2

W B W B W B W B W B W B AM M B G

12340 24268 10626 25982 10477 26131 10521 26087 10762 25846 10634 25974 1714 149 105 136 8 11930 24678 10311 26297 10067 26541 10097 26511 10366 26242 10271 26337 1619 244 214 55 40 9045 27563 10012 26596 9738 26870 9770 26838 10039 26569 9945 26663 967 274 242 27 67

MO AM M B G R1 R2

W B W B W B W B W B AM M B G

27010 27010 9490 27118 9518 27090 9687 26921 26998 27024 14670 10517 17492 28 12

11498 25110 11432 25176 11490 25118 11524 25084 11490 25118 432 66 8 58 8

10012 2619 9515 27093 9535 27073 9718 26890 9982 26999 967 497 477 20 30

VM AM M B G R1 R2

W B W B W B W B W B AM M B G

10744 25864 10517 26091 10567 26041 10823 25785 10735 25873 1596 227 177 79 9

11848 24760 11781 24827 11851 24757 12031 24577 11889 24719 82 67 3 183 41

4870 31738 9515 27093 4398 32210 4548 32060 4847 31761 4175 4645 472 322 23

(7)

In the above table 1, NI refers the normal image AM is the arithmetic mean, M is the median, B is the bilateral and G is the Gaussian filtering. R1 is the ratio between the NI and the three types of Alzheimer image MD, MO, VM. The best fileting techniques is selected based 1. How much noise is reduced? 2. How much pixels are affected when reducing the noise (white pixel difference of Alzheimer type and the fileting techniques).

So, based on this condition the Gaussian filter performs well in all types of Alzheimer image and it is proved by calculating R1 and R2. The filter which has reduce more noise and pixel disturbance is very less are computed by comparing the one filters to other fileting techniques. It should preserve the edges of the image without affecting the pixels. Accuracy of the pre-processing technique is determined based on these two values R1 and R2.

R1 gives the value of noise reduced in an image.

R1= ATi–FTi ………..(6) Where,FTi is the arithmetic mean, median, Bilateral, Gaussian

ATiis the Alzheimer image types

The values found using this formula are plotted in the graph. They are shown in figure 5.

Figure 5. Chart representation for R1.

Chart represents how the pixels are varied from one to another. Three types of Alzheimer image R1 values are computed andgiven in the table 1 are compared in the chart. Three images from each category taken for the analysis. The result shows that the maximum of noise pixels removed from the original image after applying filtering techniques.

R2 describes how much it has distressed the original pixels. The more changes in the pixel lead to elimination.

The less pixel changes consider for selection.

R2=FTwpI-ATi ……… (7)

Where, FTwp i is the pre-processed white pixel of arithmetic mean, median, Bilateral, Gaussian ATiis the Alzheimer image types

The values of R2 derived using this formula are plotted in the chart. They are shown in figure6.

(8)

4440

Figure 6. Chart representation for R2.

Figure 6 represents how the pixels are disturbed in the original image after applying the three filters. Three types of Alzheimer image R2 values are computed by finding the difference between the original Alzheimer types and the pre-processed Alzheimer types of images. It is plotted based on the determined values in the table 1. One image from each category taken for the clear analysis. The result shows that the minimum affected pixels in the image are in Gaussian filter.So, from the table values and the chart analysis the Gaussian filtering method performed well for three types of Alzheimer images by removing more noise, smoothen and preserve the details of the pixels.

V. CONCLUSION

The MRI Alzheimer image is applied with the computing algorithms to detect the existence of the abnormality.

Three ways are followed to implement and prove the accuracy of the pre-processing techniques in this work. In the beginning the MRI images are converted in to processing data format and then the histogram intensity distribution is calculated among the Alzheimer images. This is applied with the filtering techniques. Pixels of white and black are compared and estimated under two ratiosthat help us to find the accuracy. The values are determined based on the noise removal ratio and the pixels distressed ratio when it is applied with filtering.

Filtering techniques Arithmetic mean, median, Gaussian, bilateral are used and Fast Fourier transformation used to find the frequency of the image. Gaussian filter satisfies the condition fixed for finding accurateness. It removed the noise and it haddistressed very less pixels in the image.

References

1. Damasio, Hanna. Human brain anatomy in computerized images. Oxford university press, 1995.

2. Raj, Ashish, and Yu-hsien Chen. "The wiring economy principle: connectivity determines anatomy in the human brain." PloS one 6.9 (2011): e14832.

3. Alexander‐Bloch, Aaron, et al. "Subtle in‐scanner motion biases automated measurement of brain anatomy from in vivo MRI." Human brain mapping 37.7 (2016): 2385-2397.

4. Sagheer, Sameera V. Mohd, and Sudhish N. George. "A review on medical image denoising algorithms." Biomedical signal processing and control 61 (2020): 102036.

5. Bhrathi A, A.S arunachalm .”A Survey on Early Detection and Prediction of Alzheimer’s Disease using Machine Learning Methods” journal of xidian university , ISSN No:1001-2400, VOLUME 14, ISSUE 6, 2020,pgno1262-1268.

6. Crum, William R. "Magnetic resonance brain image processing and arithmetic with FSL." Magnetic Resonance Neuroimaging. Humana Press, 2011. 109-126.

7. Ibrahim, Walaa Hussein, Ahmed AbdelRhman Ahmed Osman, and Yusra Ibrahim Mohamed. "MRI brain image classification using neural networks." 2013 international conference on computing, electrical and electronic engineering (ICCEEE). IEEE, 2013.

8. George, E. Ben, and M. Karnan. "MRI brain image enhancement using filtering techniques." International Journal of Computer Science & Engineering Technology (IJCSET), ISSN (2012): 2229-3345.

9. Senel, Hakan Güray, Richard Alan Peters, and Benoit Dawant. "Topological median filters." IEEE Transactions on Image processing 11.2 (2002): 89-104.

(9)

10. Phophalia, Ashish, and Suman K. Mitra. "Rough set based bilateral filter design for denoising brain MR images." Applied Soft Computing 33 (2015): 1-14.

11. Mustafa, Zeinab A., and Yasser M. Kadah. "Multi resolution bilateral filter for MR image denoising." 2011 1st Middle East Conference on Biomedical Engineering. IEEE, 2011.

12. Herholz, Karl, et al. "Direct comparison of spatially normalized PET and SPECT scans in Alzheimer’s disease." Journal of Nuclear Medicine 43.1 (2002): 21-26.

13. Chaddad, Ahmad, Christian Desrosiers, and Matthew Toews. "Local discriminative characterization of MRI for Alzheimer's disease." 2016 IEEE 13th International Symposium on Biomedical Imaging (ISBI).

IEEE, 2016.

14. Saeed, Soobia, Afnizanfaizal Abdullah, and N. Z. Jhanjhi. "Implementation of fourier transformation with brain cancer and CSF images." Indian Journal of Science and Technology 12 (2019): 37.

15. Zhang, Yudong, et al. "Detection of Alzheimer's disease and mild cognitive impairment based on structural volumetric MR images using 3D-DWT and WTA-KSVM trained by PSOTVAC." Biomedical Signal Processing and Control 21 (2015): 58-73.

Referințe

DOCUMENTE SIMILARE

Approximation problems of functions of several variables defined on cubes of the Euclidean space are usually studied by assuming that the functions are periodic in each of

, Convergence of the family of the deformed Euler-Halley iterations under the H¨ older condition of the second derivative, Journal of Computational and Applied Mathematics,

In the very recent paper [5], Strichartz investigated the behaviour of the arclengths of the graphs Γ(S N (f )) of the partial sums S N (f ) of the Fourier series of a piecewise

Keywords: trickster discourse, meaning, blasphemy, social change, transgression of social norms.. The Myth of the trickster and its

rnetric sþacets X^rbsþectiael,y Y are-NS-isomorþkic, tken the corresþond'ing quoti,ent sþaces læ ønd, lo øre homeomorþhic.. Rernarh

For the quasioperations the interval arithmetic is not inclusion mono- tonic.. \Miss' Berlin ected with inversability of complex ervals) is the scope of the

The evolution to globalization has been facilitated and amplified by a series of factors: capitals movements arising from the need of covering the external

Using a case study designed for forecasting the educational process in the Petroleum-Gas University, the paper presents the steps that must be followed to realise a Delphi