Brain tumor MRI identification and classification using DWT, PCA and kernel support vector machine Текст научной статьи по специальности «Медицинские технологии»

Современные инновации, системы и технологии // Modern Innovations, Systems and Technologies

2024; 4(1) eISSN: 2782-2818 https://www.oajmist.com

УДК: 004.5

DOI: https://doi.org/10.47813/2782-2818-2024-4-1-0133-0152

EDN: YGYKSV

Brain tumor MRI identification and classification using DWT, PCA and kernel support vector machine

Omar Faruq1'2, Islam Md Jahidul3, Md. Sakib Ahmed4, Md Sajib Hossain4, Narayan

Chandra Nath5

department of Electrical and Electronic Engineering, Bangladesh University of Engineering

and Technology, Dhaka- 1000, Bangladesh 2School of Communication and Information Engineering, Chongqing University of Posts and Telecommunications, P.O. Box 400065, P.R China 3School of Software Engineering, Northeastern University, P.O. Box 110819, P.R. China 4Department of Electrical and Electronic Engineering, Green University Bangladesh, Dhaka-

1207, Bangladesh

5Department of Control, Microsystems, Microelectronics, University of Bremen, Bremen,

Germany

Abstract. Classification, segmentation, and the identification of the infection region in MRI images of brain tumors are labor-intensive and iterative processes. Numerous anatomical structures of the human body may be envisioned using an image processing theory. With basic imaging methods, it is challenging to see the aberrant human brain's structure. The neurological structure of the human brain may be distinguished and made clearer using the magnetic resonance imaging technique. The MRI approach uses a number of imaging techniques to evaluate and record the human brain's interior features. In this study, we focused on strategies for noise removal, gray-level co-occurrence matrix (GLCM) extraction of features, and segmentation of brain tumor regions based on Discrete Wavelet Transform (DWT) to minimize complexity and enhance performance. In turn, this reduces any noise that could have been left over after segmentation due to morphological filtering. Brain MRI scans were utilized to test the accuracy of the classification and the location of the tumor using probabilistic neural network classifiers. The classifier's accuracy and position detection were tested using MRI brain imaging. The efficiency of the suggested approach is demonstrated by experimental findings, which showed that normal and diseased tissues could be distinguished from one another from brain MRI scans with about 100% accuracy.

Keywords: classification, DWT, Feature extraction, Image Segmentation, PCA, PNN, Pre-Processing.

Acknowledgements: The authors did not receive any funding for the research, writing, or publication of this work. We are encouraged by the research now available to investigate medical image technology. We believe we can develop a system that continuously increase better image processing results. In this study, we investigated various segmentation and classification techniques while processing image data via DWT, PCA, PNN, and KSVM. We have shown that our technique can distinguish the best results from others. We are not able to reveal our unique code. However, we have used Matlab, and C to create our code, schematics, and simulations. All authors contributed equally.

© O.Faruq, IJahidul, S.Ahmed, S.Hossain, N.Ch.Nath, 2024

0133

For citation: Faruq, O., Islam, M. J., Ahmed, M. S., Hossain, M. S., & Nath, N. C. (2024). Brain tumor MRI identification and classification using DWT, PCA and kernel support vector machine. Modern Innovations, Systems and Technologies, 4(1), 0133-0152. https://doi.org/10.47813/2782-2818-2024-4-1-0133-0152

МРТ-идентификация и классификация опухолей головного мозга с использованием DWT, PCA и машины

опорных векторов ядра

Омар Фарук1'2, Ислам Джахидул3, Сакиб Ахмед4, Саджиб Хоссейн4, Нараян

Чандра Натх5

1 Кафедра электротехники и электроники, Бангладешский Инженерно-Технологический

Университет, Дакка - 1000, Бангладеш 2Школа коммуникационной и информационной инженерии, Чунцинский Университет

Почты и Телекоммуникаций, Чунцин, КНР 3Школа разработки программного обеспечения, Северо-Восточный Университет, КНР 4Кафедра электротехники и электроники, Зеленый университет Бангладеш, Дакка-

1207, Бангладеш

5Кафедра управления, микросистем, микроэлектроники, Бременский университет,

Бремен, Германия

Аннотация. Классификация, сегментация и идентификация области инфекции на МРТ-изображениях опухолей головного мозга являются трудоемкими и итеративными процессами. Многочисленные анатомические структуры человеческого тела можно представить с помощью теории обработки изображений. С помощью базовых методов визуализации сложно увидеть аномальную структуру человеческого мозга. Неврологическую структуру человеческого мозга можно различить и уточнить с помощью метода магнитно-резонансной томографии. Подход МРТ использует ряд методов визуализации для оценки и регистрации внутренних особенностей человеческого мозга. В этом исследовании мы сосредоточились на стратегиях удаления шума, извлечении признаков из матрицы совместной встречаемости на уровне серого (GLCM) и сегментации областей опухоли головного мозга на основе дискретного вейвлет-преобразования (DWT) для минимизации сложности и повышения производительности. В свою очередь, это уменьшает любой шум, который мог остаться после сегментации из-за морфологической фильтрации. Сканирование МРТ головного мозга использовалось для проверки точности классификации и местоположения опухоли с использованием вероятностных классификаторов нейронных сетей. Точность классификатора и определение положения были проверены с помощью МРТ головного мозга. Эффективность предложенного подхода подтверждается экспериментальными результатами, которые показали, что нормальные и больные ткани можно отличить друг от друга на МРТ головного мозга с точностью около 100%.

Ключевые слова: классификация, DWT, извлечение признаков, сегментация изображений, РСА, предварительная обработка.

Благодарности: Авторы не получали финансирования для проведения исследования, написания или публикации данной работы. Нас воодушевляют исследования, посвященные технологиям медицинских изображений. Мы считаем, что можем разработать систему, которая будет постоянно улучшать результаты обработки изображений. В этой работе мы исследовали различные методы сегментации и классификации при обработке данных изображений через DWT, РСА, РМК и К^УМ. Мы показали, что наша методика позволяет отличить лучшие

результаты от других. Мы не можем раскрыть наш уникальный код. Однако для создания кода, схем и моделирования мы использовали Matlab и C. Все авторы внесли равный вклад в эту работу.

Для цитирования: Фарук, О., Ислам, Д., Ахмед, С., Хоссейн, С., & Натх, Н. Ч. (2024). МРТ-идентификация и классификация опухолей головного мозга с использованием DWT, PCA и машины опорных векторов ядра. Современные инновации, системы и технологии - Modern Innovations, Systems and Technologies, 4(1), 0133-0152. https://doi.org/10.47813/2782-2818-2024-4-1-0133-0152

INTRODUCTION

An Image in image processing symbolizes the under-standing that object data was also processed to produce an image. Clinical personnel are now better equipped to pro-vide patients with healthcare thanks to the recent use of technology and electronic healthcare systems in the medical industry [1]. In order to identify the issue of segmenting aberrant and normal tissues from MR images, this study employs the gray-level co-occurrence matrix (GLCM) for feature extraction and probabilistic neural network (PNN) classification. Brain tumors are caused by the uncontrolled growth of aberrant, malignant tissue. Both benign and malignant brain tumors are possible. The benign tissue has cancer cells that are unrelated to the tumor and are physically equivalent. The malignant tumor contains active cancer cells that move across regions and has irregularly shaped features.

Malignant tumors are categorized on the scales from Grade I to Grade IV, which are utilized in the grading system. Tumors of grades I and II are of lesser grade, whereas those of grades III and IV are of greater grade [7]. At any age, brain tumors can affect people. It might not affect everyone in the same way. The complexity of the human brain makes it challenging to diagnose the tumor region there.

Malignant grade III and IV tumors are rapidly developing. It damages the proper brain cells, has the potential to spread to all other sections of the central nervous system (CNS), and can be dangerous if left untreated [9]. The early detection and categorization of such brain tumors is thus a critical concern in medical science. It enables a physician to monitor and follow the occurrence and progression of malignant cell regions at various stages by enhancing the most recent imaging techniques, allowing them to make reliable diagnoses by scanning these pictures.

The biggest problem was finding brain cancers in time for proper therapy. Based on these factors, the best radiation, surgery, or chemotherapy therapy can be decided. As an outcome,

it's also obvious that early diagnosis of a tumor can considerably boost a tumor-infected patient's chances of survival [26]. Using imaging methods, segmentation is used to evaluate the part of the tumor that is affected. It is the process of dividing an object into component parts that share similar properties such as color, texture, contrast, and borders.

Human brain tumors are among the primary causes of mortality. It is clear that if the tumor is recognized early and detected effectively, the chances of survival may in-crease. Invasive approaches include the usual technical procedures for identifying and classifying benign (non-cancer) and malignant (cancer) brain tumors, including biopsies, lumbar punctures, and movements of the spinal tap.

A computer-assisted diagnostic algorithm was developed to replace traditional invasive devices and time-consuming techniques in order to increase brain tumor di-agnostic accuracy and classification. This article de-scribes a high-quality tumor classification approach in which individual magnetic resonance (MR) pixels are categorized as common genius tumors, non-cancerous tumor letters, and malignant tumor letters. There are three phases in the proposed method: (1) wavelet decomposition, (2) texture extraction, and (3) classification. The MR image was initially decomposed into unique steps with defined approximation coefficients using discrete wavelet processing with Daubechies' wavelet (db4), after which structural data such as forces were acquired. The suggested technique has been applied to real MR images, with a classification accuracy of over 98% utilizing probabilistic neural networks.

LITERATURE REVIEW Image Acquisition

The datasets were created from OASIS and, the ADNI to axial T2-weighted MR pictures and 256 albums. Higher contrast images are T2 and have the best visuals between the T1 and PET (positron emission tomography) modalities; we prefer the T2 model. The unnatural data set of brain MR picture consists of the below diseases: Alzheimer's disease, glioma, meningioma, visual agnosia, Pick's disease when Alzheimer's disease stays with it, Huntington's disease, & sarcoma. All disease specimens are shown in Fig. 1. Pick 20 random images for each brain category. The reason is that there is one category of the natural brain & seven categories of the unnatural brain in the dataset, and 160 pictures are chosen to create 20 natural & 140 unnatural brain pictures.

Pre-Processing

Figure 1. MRI brain picture: (a) natural brain; (b) Glioma; (c) Meningioma; (d) Alzheimer's illness; (e) Alzheimer's illness with visual agnosia; (f) Pick's disease; (g) Sarcoma; (h) Huntington's illness.

Five different levels are created by dividing the images of the MRI by the identical coefficients of the LL and HL bands. Such sub-bands have been attained from the disarranged wavelet; GLCM has been used to extract quantitative textural features such as strength, co-relation, entropy, and homogeneity [3]. Filtering and de-noising pictures (shown in Fig. 2) are the initial stages of image pre-processing. Certain recovery approaches are used to eliminate generated noise that might enter a picture during capture, transfer, or compression. [2].

Noisy image Denoised image

Figure 2. Example of noisy image and reduction of noise.

Segmentation of Image

Image segmentation is the procedure of splitting an image into distinct segments, each with comparable features in Figure 3. depicts the many types of image segmentation [1, 24].

Figure 3. Image segmentation types.

Feature Extraction

Five different levels are created by dividing the images of the MRI by the identical coefficients of the LL and HL bands. Such sub-bands have been attained from the disarranged wavelet; The GLCM has been used to extract quantitative textural characteristics such as strength, co-relation, entropy, and uniformity [10]. Filtering and de-noising pictures are the initial stages of image pre-processing. De-noising is the technique of eliminating artificial noise from a picture before capturing, transmission, or compression utilizing specialized recovery algorithms [2]. Picture segmentation is the process of dividing an image into distinct segments, each with comparable features. As seen in figure 3, there are several forms of image segmentation [11, 17]. Feature extraction is the process of retrieving statistical data from an image that resembles color characteristics, texture, shape, and contrast [12]. We employed the discrete wavelet transform (DWT) to extract wavelet coefficients and the GLCM to obtain applied math features. Feature extraction is the process of choosing subsets from the set of variables that demonstrate the behavior of the entire set. choosing the helpful variables and discarding the inapplicable ones.

Pictures 1 to 8 display various sub-band levels up to the decomposition stage of the 5th wavelet. As an input vector, the extracted form was used to train and evaluate the output of the PNN types. The characteristics of the statistical field are shown in Table 1 [3].

Table 1. GLCM of LL and HL sub- bands of trained pictures & statistical field.

Picture Contrast Correlation Energy Homogeneity Entropy

Pic1 0.386541 0.0805 0.81568 0.900 2.3224

Pic2 0.360122 0.1725 0.81607 0.9474 0.2244

Pic3 0.265017 0.1248 0.77808 0.9383 2.8746

Pic4 0.315628 0.0928 0.79636 0.9404 2.6739

Pic5 0.365406 0.1353 0.80855 0.9447 2.5116

Pic6 0.285873 0.1143 0.75812 0.9315 2.8806

Pic7 0.299221 0.1129 0.78455 0.9387 2.9101

Pic8 0.267241 0.1246 0.77884 0.9367 2.9114

Image Classification

Image classification is the extraction of data classes from bitmap image having numerous bands. The real three types of classification are per pixel, per subpixel, and object.

Figure 4. Different types of techniques for image recognition. This study concentrated on pixel-scale image categorization [4], which is classified into three categories.: The two most common approaches are supervised classifiers (manual) and

unsupervised classifiers [5, 6], (calculated by the software), but analytical object-based images are rare and the most recent technique, as mentioned above, and the analysis is fed with highresolution images. Figure 4 depicts the various types of object classification from different points of view.

METHODOLOGY

The suggested method for the identification of brain tumors involves five major steps:

• Extraction of features using DWT

• Extraction of textures using GLCM

• Selection of features using PCA

• Detection of tumors and K-means structural analysis

• Using PNN-RBF Network identification

Extraction of features using DWT

Ж

DWT

HC

PCA

PNN

Ж

Output

MRI

■

Test Image

Train Ima ing ge

1 у

DWT

1

PCA

where Фа,ь(Р) =j= .

Figure 5. Six-stage image reference block diagram to output.

b) = f(x) * ^ab(t)dx,

1

(1)

As a function vector, the proposed system makes use of the coefficients of the Discrete Wavelet Transform (DWT). The wavelet is a powerful mathematical tool for extracting facets and determining the wave coefficient from images; where employed MR. In Figure 5, Six-stage image reference block diagram to output. Wavelets are twisted as well as resized variations of constant-moment wavelets that have localized basic properties. Wavelets have the primary

advantage of imparting localized frequency records about a signed variable, which is especially useful for ranking. A quintessential review of wavelet decomposition is delivered as below: (t) is a continuous wavelet transformation of the ax(t) signal with a squareintegrable characteristic, unlike a real-value wavelet. An indispensable evaluation of wavelet decomposition is delivered as below: A continuous wavelet transformation of an ax (t) signal, a squareintegrable feature comparative to a real-value wavelet (t), is defined as:

WT = {dj,k = ^(x(n)h * i(n - 2jk)>)

X(n) Uj,k = l(x(n)g*j(n-2jk)y Wavelets a and b are defined by translation and dilation from the mother's wavelet, wavelet, dilation factor, and b translation parameter (both of which are real, high-quality numbers). Under some mild assumptions, the base wave-let responds to the limit of having zero meaning. The equation can be debunked by means of limiting a and b to a discrete lattice, which gives the discrete wavelet [7].

Figure 6. DWT Schematically.

The technique that divides information into different wavelength elements and analyzes these elements with a precision proportional to their size is known as DWT:

Ho_D - High Pass Filter, Lo_D - Low Pass Filter.

The Daubechies wavelet is thought to have the best quality for image utility based on literature analysis, and among the other wavelets, LH's overall performance was once better than the HL sub-band characteristics, shown in Figure 6. The qualities were then extracted from the DWT sub bands LH and HL, and a five-level decomposition of the Daubechies wavelet usage was determined in this approach.

Extraction of textures using GLCMs

Examination of the texture makes it easy to distinguish between regular and anomalous tissue. It even contrasts malignant tissue to regular tissue, which can be un-der the person's experience level. Texture evaluation is one of the uses of computer-assisted pathology as a supplement to strategies for biopsy. This technique calculates a specific gray level frequency in a random photograph region and does not take the correlations between pixels into consideration. Using this strategy, the frequency is computed at a specific gray level in a random image region. It no longer takes into account pixel associations. The chance of detecting a pair of grey ranges at very vast distances and with intention throughout a full body is the base of statistical texture analysis in second-order texture recording. Using the GLCM additionally acknowledged as the Gray Level Spatial Dependence Matrix (GLSDM), statistical aspects of the MR snapshots are obtained. Haralick's GLCM is a statistical technique for explaining the connection between pixels of a particular gray level. GLCM is a two-dimensional histogram where (I, j) is the variable and the prevalence frequency is (I, j). It is an interval feature with d = 1, perspective 45, ninety, and one hundred thirty-five, and greyscales I and j calculate the frequency with which a pixel has an intensity I that is about to positively outstrip and an orientation in relation to another pixel j. The gray degree co-occurrence matrix and the statistical field elements such as contrast, interrelationship, and electricity are formed in this way. Homogeneity and entropy were obtained from the first five wavelet decomposition degrees of the LH and HL sub-bands [3].

Equation Selection of features using PCA

The element-component analysis is the best way to do dimension reduction. In PCA, the linear data seeks the lower dimensional in such a way that the replicated data variance due to a set of data is preserved. Limiting the feature vector calculated from waves based on a combined feature vector component analysis using a feature reduction system to a variable chosen by the PCA will result in an effective classification algorithm using a monitoring method, shown in Figure 7.

The purpose of PCA is to diminish the complexity of the wavelet coefficient. It operates under a classification that is much more descriptive and precise.

Figure 7. Block diagram for the extraction and reduction of the feature used.

Using the Probabilistic Neural Network (PNN) Classification

It is a neural feed-forward network derived from a Bayesian network and Kernel Fisher's discriminatory analysis statistical algorithm [19, 25, 21, 22]. D.F. implements it. Specht's operations were founded in early 1990s on a four-tiered, multilayered, feed-forward net-work based on a PNN: in layer: 1. Input, 2. Pattern /Summation, 3. Output, 4. Hidden.

Probabilistic Neural Network Architecture: PNN layers: PNN is regularly used in problems with classification. When an entry is present, the very first layer calculates the gap between the entry vector and the entry vectors of the layers. Probabilistic Neural Network (PNN) architecture used to evaluate this research for MRI image identification and classifications, the architecture shown in Figure 8. It generates a vector in which the factors indicate how close the center is to the center of the learning [7, 19]. Level 2 adds up the benefits of every in-put class and generates its internet output like a probability vector. Then, a real switch characteristic on 2D layer output selects the greatest of these chances and generates a 1 for that type and a 0 (obstructive identity).

Figure 8. Architecture of PNN.

EXPERIMENTAL RESULT AND DISCUSSION

The tests were performed with a 5.0GHz processor and 32 GB of RAM on the Intel i7 platform running Windows 10. Using the wavelet toolbox to develop the algorithm, Matlab's 2018 bio statistical toolbox (Mathworks°c) was used. Install a free SVM toolbox, expand the SVM kernel, and use the classification of the MR brain picture. On any software platform where Matlab is available, the pro-grams can be run or checked. We tested four SVMs using the most recent Di® kernels. The SVM using a linear kernel is called KSVM [14, 16]. Table 2. Confusion matrix of our DWT+ PCA+KSVM method (Kernel pick out LIN, HPOL,

IPOL).

Approach from existing study Accuracy(%)

SOM+DWT 95

SVM+ DWT with linear kernel 96

SVM+ DWT with RBF based kernel 98

PCA+KNN+ DWT 98

PCA+ANN+ DWT 97

PCA+ACPSO+FNN+ DWT 98.75

Approach from this study Accuracy(%)

PCA+KSVM+ DWT (LIN) 95.38%

PCA+KSVM+ DWT (HPOL) 96.53%

PCA+KSVM+ DWT (IPOL) 98.88%

PCA+KSVM+ DWT (GRB) 99.61%

Table 3. A comparison of the classification accuracy of eight different algorithms using the same MRI dataset and the same number of images. Parameter Base Accuracy

Lin normal [O] abnormal [0]

Normal (T) 18 3

Abnormal 5 230

HpoL NORMAL[O] ABNORMAL [O]

Normal (T) 19 1

Abnormal (T 4 232

IPOL Normal (T) Abnormal (T

GRB Normal (T) Abnormal (T)

NORMAL [O] 18 1

NORMAL [O] 20 1

ABNORMAL [O] 2 239

ABNORMAL [O] 0 239

The results reveal that the proposed DWT+PCA+KSVM technique generates outstanding results both at the training and validation datasets. S um up the overall classification performance of the LIN, HPOL, IPOL, and GRB kernels in table 2. The GRB kernel SVM architecture which provides the best the kernel's com-pared to others three SVMs kernels [15].

Furthermore, we evaluated our technique to six successful techniques disclosed in the current papers employing the same MRI datasets and image count. Table 2 dis-plays the results of the analysis. It demonstrates that our suggested DWT+PCA+KSVM system with the GRB kernel outperforms the other 8 approaches, obtaining the greatest classification performance at 99.38 percent. The next technique is the DWT+PCA+ACPSO+FNN method, which has a 98.75 percent accuracy. The fifth is our suggested DWT+PCA+KSVM IPOL kernel, which has a classification accuracy of 98.12% [13, 23].

Table 4. GLCM of LL and HL sub bands of tested images & statistic field.

Pictures Psnr Mse Area of Picture in pixel Area of the tumor region

Picl 13.011 5.121 38,240 7598

Pic2 12.82 2.116 66,824 9774

Pic3 13.12 6.068 49,508 7532

Pic4 14.86 5.77 49,388 8956

Pic5 12.79 4.84 23,964 4465

Pic6 12.82 6.79 49,429 3598

Pic7 12.99 5.92 49,298 5979

Pic8 13.004 6.35 34,040 12,932

Table 5. The declaration of the region and the success of the evaluation of tumors extraction

location of trained image.

Pictures Con Cor Ene Hom Ent

Picl 0.2755 0.106333 0.72044 0.92139 3.3209

Pic2 0.2611 0.129665 0.76435 0.93428 3.3830

Pic3 0.2611 0.129665 0.76435 0.93428 3.2821

Pic4 0.2929 0.156369 0.82164 0.94967 2.7610

Pic5 0.2469 0.07749 0.74696 0.92848 3.3895

Pic6 0.2416 0.177739 0.77415 0.9373 2.5186

Pic7 0.2611 0.12966 0.76435 0.93428 3.2820

Pic8 0.3053 0.09188 0.7674 0.9326 2.9181

The training data set objects were those for which the extracted features were classified using the PNN algorithm, whereas the test set was not trained and just statistical and textual characteristics were extracted. The comparison of the category of training and tested datasets using probabilistic neural networks, shown in Figure 9. The correctness of the qualifying and verified image was determined by classifying malignant cells as either normal or abnormal [20, 21]. Figure 10 depicts the results of the identification of malignant and non-malignant tumor tissues.

Figure 9. The categorization of training and tested datasets using probabilistic neural

networks is compared.

The efficacy of an acceptable classification proportion to the actual number of form is referred to as prediction performance or "correct level" [19]. The brain tumor classification procedure was applied to a variety of both nor-mal and abnormal MR images, and the performance of the PNN predictor was modified using the equation shown be-low:

Calculation time is also an important factor in classifier evaluation. Since the SVM parameters stayed the same after training, the SVM training time was not taken into account. All 260 images were fed into the classifier. The time spent at each stage of the DI® event was plotted in Fig. 10, along with the associated computation time, average value, and reported computation time.

Type Of Tumor MALIGNANT Contrast iiwi _ ^ Correlation «turn

50 и 90 Ene,W l"*1"

Homogeneity 0 i3Ma3

Figure 10. Segmentation and Classifying of Tumor.

In this lesson, we also created a brand-new DWT + PCA + KSVM technique to differentiate between authentic and artificial brain MRIs. We selected four different kernels for the LIN, HPOL, IPOL, and GRB [8, 24]. The study's findings show that the GRB kernel SVM outperformed the HPOL, IPOL, and GRB kernels as well as other well-known methods, achieving a classification performance of 99.38% on the 260 images from MR. For the public review and improvement of research This research papers have two preprints [26, 27].

CONCLUSION

In this research, we used brain MRI images separated between healthy (unaffected) brain tissue and aberrant (infected) tumor tissue. Preprocessing is used to eliminate noise from images and smoothing them out. This also helps to increase the signal-to-noise ratio. As a result, we applied a discrete wavelet transformation, which breaks down the image and extracts the features from the GLCM, which was detected using structural procedures. PNN could be a classifier used to perceive tumors in brain MRI images. According to research findings, diagnosing a brain tumor yourself is both faster and more accurate than having it done by a medical professional. The greatest components also demonstrate that it performs better by increasing PSNR and MSE metrics. The suggested technique concluded with precise, quick brain tumor detection and tumor site localization. By recognizing and classifying normal and abnormal tumors in the brain, MR pictures have achieved almost 100 percent for the series of qualified data due to the extraction of the statistical features of the wavelet decomposition of LL and HL sub bands and 95 percent for the proven record. We conclude with the above findings that our mentioned technique clearly differentiates the tumor from normal and abnormal, which allows medical experts to make clear diagnostic decisions.

REFERENCES

[1] Varuna Shree N., Kumar T. N. R. Identification and classification of brain tumor MRI images with feature extraction using DWT and probabilistic neural network. Brain Informatics. 2018; 5(1): 23-30. https://doi.org/10.1007/s40708-017-0075-5

[2] Sharma A., Singh J. Image denoising using spatial domain filters: A quantitative study. 2013 6th International Congress on Image and Signal Processing (CISP), Dec. 2013. https://doi.org/10.1109/cisp.2013.6744005

[3] Salem Ghahfarrokhi S., Khodadadi H. Human brain tumor diagnosis using the

combination of the complexity measures and texture features through magnetic resonance image. Biomedical Signal Processing and Control, 2020; 61: 102025. https://doi.org/10.1016/j.bspc.2020.102025

[4] Kwak N., Choi Chong-H. Input feature selection for classification problems. IEEE Transactions on Neural Networks. 2002; 13(1): 143-159. https://doi .org/10.1109/72.977291

[5] Amin J., Sharif M., Raza M., Saba T., Rehman A. Brain Tumor Classification: Feature Fusion. 2019 International Conference on Computer and Information Sciences (ICCIS), Sakaka, Saudi Arabia. 2019: 1-6. https://doi.org/10.1109/ICCISci.2019.8716449

[6] Ayadi W., Elhamzi W., Charfi I., Atri M. A hybrid feature extraction approach for brain MRI classification based on Bag-of-words. Biomedical Signal Processing and Control. 2019; 48: 144-152. https://doi.org/10.1016/j.bspc.2018.10.010

[7] Nagtode S. A., Potdukhe B. B., Morey P. Two dimensional discrete Wavelet transform and Probabilistic neural network used for brain tumor detection and classification. In 2016 Fifth International Conference on Eco-friendly Computing and Communication Systems (ICECCS). 2016. https://doi.org/10.1109/eco-friendly.2016.7893235

[8] Mathew A.R., Anto P.B. Tumor detection and classification of MRI brain image using wavelet transform and SVM. 2017 International Conference on Signal Processing and Communication (ICSPC). 2017: 75-78. https://doi.org/10.1109/CSPC.2017.8305810

[9] Mao K. Z., Tan K.-C., Ser W. Probabilistic neural-network structure determination for pattern classification. IEEE Transactions on Neural Networks. 2000; 11(4): 1009-1016. https://doi.org/10.1109/72.857781

[10] Bahadure N.B., Ray A.K., Thethi HP. Image Analysis for MRI Based Brain Tumor Detection and Feature Extraction Using Biologically Inspired BWT and SVM. International Journal of Biomedical Imaging. 2017; 2017: 1-12. https://doi.org/10.1155/2017/9749108

[11] Sabitha V., Nayak J., Reddy P. R. MRI brain tumor detection and classification using KPCA and KSVM. Materials Today: Proceedings. 2021. https://doi.org/10.1016/j.matpr.2021.01.714

[12] Rao C.S., Karunakara K. Efficient Detection and Classification of Brain Tumor using Kernel based SVM for MRI. Multimedia Tools and Applications. 2022. https://doi.org/10.1007/s11042-021-11821-z

[13] Kumar K, Devi K.T. An Efficient Method for Brain Tumor Detection Using Texture Features and SVM Classifier in MR Images. Asian Pac J Cancer Prev. 2018; 19(10): 27892794. https://doi.org/10.22034/APJCP.2018.19.10.2789

[14] Mandle A. K., Sahu S. P., Gupta G. Brain Tumor Segmentation and Classification in MRI using Clustering and Kernel-Based SVM. Biomedical and Pharmacology Journal. 2022; 15(2): 699-716. https://doi.org/10.13005/bpj/2409

[15] Jahidul I.M., Faruq O. Further Exploration of Deep Aggregation for Shadow Detection. Modern Innovations, Systems and Technologies. 2022; 2(3): 0312-0330. https://doi.org/10.47813/2782-2818-2022-2-3-0312-0330

[16] Naceur M. B., Saouli R., Akil M., Kachouri R. Fully Automatic Brain Tumor Segmentation using End-To-End Incremental Deep Neural Networks in MRI images. Computer Methods and Programs in Biomedicine. 2018; 166: 39-49. https://doi.org/10.1016/j.cmpb.2018.09.007

[17] Lavanyadevi R., Machakowsalya M., Nivethitha J., Kumar A.N. Brain tumor classification and segmentation in MRI images using PNN. 2017 IEEE International Conference on Electrical, Instrumentation and Communication Engineering (ICEICE), Apr. 2017. https://doi.org/10.1109/iceice.2017.8191888

[18] Othman M.F., Basri M.A.M. Probabilistic Neural Network for Brain Tumor Classification. 2011 Second International Conference on Intelligent Systems, Modelling and Simulation. 2011: 136-138. https://doi.org/10.1109/ISMS.2011.32

[19] Deepa B., Sumithra M. G., Kumar R. M., Suriya M. Weiner Filter based Hough Transform and Wavelet feature extraction with Neural Network for Classifying Brain Tumor. 2021 6th International Conference on Inventive Computation Technologies (ICICT), Jan. 2021. https://doi.org/10.1109/icict50816.2021.9358680

[20] Thara S., Jasmine K. Brain tumour detection in MRI images using PNN and GRNN. 2016 International Conference on Wireless Communications, Signal Processing and Networking (WiSPNET), Mar. 2016. https://doi.org/10.1109/wispnet.2016.7566388

[21] Zhang Y., Wang S., Wu L. A novel method for magnetic resonance brain image classification based on adaptive chaotic pso. Electromagn. Waves (Camb.). 2010; 109: 325343. http://dx.doi.org/10.2528/PIER10090105

[22] Das S., Siddiqui N., Kriti N., Tamang S. P. Detection and area calculation of brain tumour from MRI images using MATLAB. International Journal of Computer Engineering In Research Trends. 2017; 4(1): 37-40. Available:

https://www.semanticscholar.org/paper/d7f3b185dcc8ed23f862b98a221988418e88c53f

[23] Saraswathi D., Priya B. L., Punitha Lakshmi R. Brain tumor segmentation and classification using self-organizing map. In 2019 IEEE international conference on system,

computation, automation and networking (ICSCAN). IEEE. 2019: 1-5. http://dx.doi.org/10.1109/ICSCAN.2019.8878763

[24] Amin J., Sharif M., Raza M., Saba T., Anjum M. A. Brain tumor detection using statistical and machine learning method. Computer Methods and Programs in Biomedicine. 2019; 177: 69-79. https://doi.org/10.1016/j.cmpb.2019.05.015

[25] Pandiselvi T., Maheswaran R. Efficient Framework for Identifying, Locating, Detecting and Classifying MRI Brain Tumor in MRI Images. Journal of Medical Systems. 2019; 43(7). https://doi.org/10.1007/s10916-019-1253-1

[26] Faruq O., Jahi I., Ahmed M. S., Hossain M. S. Brain tumor MRI identification and classification using DWT, PCA, and KSVM. Springer Research Square. 2023. https://doi .org/10.36227/techrxiv.21771329.v2

[27] Faruq O., Islam M. J., Ahmed M. S., Hossain M. S. Brain tumor MRI identification and classification using DWT, PCA, and KSVM. IEEE TechRxiv. 2023. https://doi.org/10.21203/rs.3.rs-2562932/v1

ИНФОРМАЦИЯ ОБ АВТОРАХ / INI

Omar Faruq, PhD Research Fellow in Electronic and Electrical Engineering at Bangladesh University of Engineering and Technology, Dhaka, Bangladesh. He is an Assistant Professor of Computer Science and Engineering at BGIFT Institute of Science and Technology, Dhaka, Bangladesh. He received his Post-Graduate degree from School of Communication and Information Engineering, Chongqing University of Posts and Telecommunications, China. Also, He graduated from Electrical and Electronic Engineering from Daffodil International University, Dhaka, Bangladesh. He engaged with researching of Machine Learning, Deep Learning, Artificial Inteligicence, Image and Signal Processing, Computer Vision, Free-Space Optical Communication, Optical Fiber Communication, Wireless Communications and Networks, etc. e-mail: [email protected] ORCID: https://orcid.org/0000-0002-4901-2753.

)RMATION ABOUT THE AUTHORS

Islam Md. Jahidul, A post-graduate student of Software Engineering at Northeastern University, China. He graduated from school of software engineering at Chongqing University of Posts and Telecommunications, China. His interested field of research is computer vision, image processing, signal processing, artificial intelligence, machine learning, deep learning, neural networks, etc. e-mail: [email protected], ORCID: https://orcid.org/0000-0002-6847-1808

Md. Sajib Hossain, Electrical and Electronic Engineering at Green University Bangladesh, Bangladesh. His interested field of research is image processing, bio-medical, bio-informatics, etc.

Nath Narayan Chandra, A post-graduate Master's program: Control, Microsystems, Microelectronics at University of Bremen, Bremen, Germany. Also received his PostGraduate degree from School of Communication

Email: [email protected], ORCID: https://orcid.org/0000-0003-2590-5095.

Ahmed Md. Sakib, Electrical and Electronic Engineering at Green University Bangladesh, Bangladesh. His interested field of research is image processing, bio-medical, bio-informatics, etc.

e-mail: [email protected], ORCID: https://orcid.org/0000-0002-4950-4539.

and Information Engineering, Chongqing University of Posts and Telecommunications, China. His interested field of research is Control system, microelectronics, computer vision, image processing, signal processing, artificial intelligence, machine learning, deep learning, neural networks, etc. e-mail: [email protected], ORCID: https://orcid.org/0009-0009-9651-747X

Статья поступила в редакцию 17.03.2024; одобрена после рецензирования 27.03.2024; принята

к публикации 28.03.2024.

The article was submitted 17.03.2024; approved after reviewing 27.03.2024; accepted for publication

28.03.2024.