Journal of Physics: Conference

Series

PAPER • OPEN ACCESS You may also like

- Brain Tumor Image Segmentation in MRI
Brain tumour classification using machine learning Image
Hapsari Peni Agustin Tjahyaningtijas
algorithm - Adaptive cascaded transformer U-Net for
MRI brain tumor segmentation
Bonian Chen, Qiule Sun, Yutong Han et
To cite this article: A B Malarvizhi et al 2022 J. Phys.: Conf. Ser. 2318 012042 al.

- Hybrid Clustering And Boundary Value

Refinement for Tumor Segmentation using
Brain MRI
Anjali Gupta and Gunjan Pahuja
View the article online for updates and enhancements.

This content was downloaded from IP address 103.105.225.66 on 27/07/2024 at 06:23

ICBSII-2022 IOP Publishing
Journal of Physics: Conference Series 2318 (2022) 012042 doi:10.1088/1742-6596/2318/1/012042

Brain tumour classification using machine learning algorithm

A B Malarvizhi, A Mofika, M Monapreetha and A M Arunnagiri

Department of ECE, College of Engineering, Guindy, Anna University (CEG Campus)

Email: [email protected]

Abstract. A Brain tumour is formed by a gradual addition of abnormal cells, and this is one of
the major causes of death among other sorts of cancers. It is necessary to classify brain tumor
using Magnetic Resonance Imaging (MRI) brain tumor image for treatment because MRI images
assist as to detect the smallest defect of the body. This paper aimed to automatically classify
brain tumours using a machine learning algorithm. In this work, the input image of the brain was
pre-processed using median filter, segmented from the background using thresholding and K-
means clustering algorithm and its features were extracted using GLCM. Using the SVM
classifier, the brain tumour in the image was detected as either benign or malignant. This image
classification process helps the doctors and research scientists to detect the tumour during its
early stages, thereby controlling the spread of cancerous cells.

Keywords. Brain tumour, Image processing, Feature extraction, Machine learning,

Classification.

1. Introduction
Brain tumour known as intracranial neoplasm, is an irregular tissue in this cell rapidly grow and
differentiate unmanageably. This condition is naturally serious and critical due to its interfering and
proliferative behaviour in the limited space of the intracranial cavity. Brain tumours can be malignant
or benign. The elements like the type of tumour, its location, its size and its state of development are
depending on its threat level. Classification of Brain tumours for treatment is important because MRI
images assist us to detect the smallest defect in the body.
The aim of this paper is to automatically classify brain tumours using a machine learning algorithm. In
this work, the input image of the brain was pre-processed, segmented from the background and its
features were extracted. This image classification process helps the doctors and research scientists to
detect the tumour during its early stages, thereby controlling the spread of cancerous cells.

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
ICBSII-2022 IOP Publishing
Journal of Physics: Conference Series 2318 (2022) 012042 doi:10.1088/1742-6596/2318/1/012042

1.1 Literature Survey

Nandita Goyal and Bharti Sharma’s Image processing techniques for brain tumour detection. This paper
describes about requirement of clinical pictures examinations used for treatment of patients and is
becoming a critical field.MRI pictures recognizing the region of cerebrum tumours by extraordinary
areas like preplanning, dividing the pictures, incorporate extraction and picture collection.
Shraddha Vyas, Hardik S.Jayswal and Amit P Ganatra’s Brain tumour detection and classification using
Image processing and machine learning. This paper briefs as, Brain Tumour detection is one of the most
critical and arduous tasks in the area of medical image processing. Detection of tumour from brain is
very difficult due to variance and complexity of tumour and dense brain tissues. Brain Tumour is
detected and classified using different image processing and machine leaning techniques. This paper,
shows the different techniques of diagnosis of Brain Tumour.
Alok Sarkar, Md. Shamim and Ahsan’s Identification of Brain tumour from MRI with Feature extraction
by Support Vector Machine. The method presented in this paper is extracting the tumour features using
feature extraction. By using a SVM classifier, the part of human brain having the tumour and it’s type
benign or malignant can be found.
The paper named Brain tumour extraction from MRI image using image processing by Rina Bopche,
Prof. Amit Kolhe and Dr. Pankaj Kumar Mishra’s focuses mainly to solve the disadvantage of region-
based segmentation. The tumour is abnormal growth of tissue and MRI can’t tell the precise location of
tumour. Image segmentation using MATLAB gives the complete information about brain tumour. This
project includes denoising technique (median filter), grayscale conversion, thresholding and
morphological operation. Brain tumour detection and extraction from MRI scan images is done by using
MATLAB software (programming).
The paper named Brain tumour detection based on segmentation using MATLAB by Animesh Hazra,
Ankit Dey and Sujit Kumar Gupta briefs as detecting the tumour region in the brain using patient's MRI
images. The method presented in this paper is about pre-processing and segmentation. Pre-processing
stage involves grayscale transformation and noise removal. To display the tumour region in the MRI
images, Segmentation process is applied. Then, the k-means clustering algorithm is used to clustering
the image. The work is done by using MATLAB.

2. Methodology
Generally, diagnosing a brain tumour usually starts with magnetic resonance imaging (MRI). MRI
Imaging plays an important role in brain tumour for analysis, diagnosis and treatment planning. The
steps involved in detecting the brain tumour from MRI Images are given below.

2
ICBSII-2022 IOP Publishing
Journal of Physics: Conference Series 2318 (2022) 012042 doi:10.1088/1742-6596/2318/1/012042

IMAGE DATA
ACQUISITION

PREPROCESSING
(MEDIAN FILTER)

SEGMENTATION
(OTSU, LOCAL THRESHOLDING, K MEANS
CLUSTERING)

FEATURE EXTRACTION

CLASSIFICATION
(SVM)

Figure 1. Block Diagram

The process begins with acquiring Image dataset. The transformations applied to the info before feeding
it to the algorithm is understood as pre-processing. Objects and boundaries in images can be located by
Segmentation process. The technique of dividing an image into regions is known as Image
Segmentation. The division takes place on the basis of property of pixels of image. Feature extraction
uses specialized algorithm to extract features automatically from signals or images without the need for
human intervention. This technique will be very useful for quickly moving from raw data to the stage
of developing machine learning algorithms. The process of classifying or predicting the class of a
particular object in an image based on certain rules is known as image classification.

2.1 Image Dataset Acquisition

A data set is a perfect collection of data. The data are fundamentally arranged to a certain model that
helps to process the required information. Kaggle functions as a community for data scientists and

3
ICBSII-2022 IOP Publishing
Journal of Physics: Conference Series 2318 (2022) 012042 doi:10.1088/1742-6596/2318/1/012042

machine learning practitioners. For this work, we downloaded the dataset from Kaggle. In the dataset of
253 images of brain, we have used 15 images.

2.2 Pre-processing
The goal of pre-processing techniques is improving the image quality by suppressing unnecessary things
or enhancing useful image features that is important for further processing. These techniques make the
image acceptable for further processing and improves the quality of the image and finally removes the
noise existing in the image. The pre-processing steps used were Grayscale transformation and Noise
removal using Median filter. Grayscale transformation can significantly improve the viewability of an
image as shown in Figure 2. Removal of noise is an important task in image processing for the further
analysis of images. To reduce salt and pepper noise or impulsive noise, the median filtering process is
useful due to its better denoising ability and computing power [1] as shown in Figure 3.

Figure 2. Grayscale Image Figure 3. Filtered Image using Median Filter

2.3 Segmentation
As we already know, the technique of dividing an image into multiples of segment is called Image
Segmentation. Changing the representation of a picture into more meaningful and very easy to analyse
is the goal of segmentation. The segmentation algorithms used were Otsu Thresholding, Local
Thresholding and K-Means Clustering Algorithm.
Otsu is a type of global thresholding which is based only on gray value of the image. According to Otsu
binarization, for an image, approximate value in middle of peaks is taken as threshold value. As it
automatically calculates the threshold value, this method is also known as an automatic threshold
selection region-based segmentation. On the basis of interclass variance maximization, it is expected to
detect optimal value for global threshold [2] [3]. The class that has larger variance determines the
threshold, be it the background or the foreground as shown in Figure 4.

4
ICBSII-2022 IOP Publishing
Journal of Physics: Conference Series 2318 (2022) 012042 doi:10.1088/1742-6596/2318/1/012042

Figure 4. Segmented Image Using Otsu Thresholding

Local thresholding, which is a binarization process can convert the coloured image into binary image as
shown in Figure 5. In this technique, based on the statistics such as variance, range of the neighbourhood
pixels, threshold value is calculated for each pixel. This can be used for uneven illumination. Unique or
specific threshold values is used for the partitioned sub images obtained from full image.

Figure 5. Segmented Image Using Local Thresholding

K Means is an unsupervised and clustering algorithm which implies that there is no labelled data present.
It is used to identify different clusters in the given data based on similarity of data [2] [4] [6].
Furthermore, it is an iterative process that partitions the dataset according to their features into K number
of clusters as shown in Figure 6.

Figure 6. Segmented Image Using K-Means Clustering Algorithm

5
ICBSII-2022 IOP Publishing
Journal of Physics: Conference Series 2318 (2022) 012042 doi:10.1088/1742-6596/2318/1/012042

2.4 Significance
Making the image analysis easy and meaningful is the main aim of segmentation. Because of its simple
calculation, Otsu method is most successful. Regions of interest in images that exhibit non-uniform
lighting changes, can be isolated using Local Thresholding. K-means algorithm of image segmentation
segments the interest region from the background. The image is also contrast stretched so that the good
quality image for segmentation can be obtained and is relatively simple to implement.

2.5 Feature Extraction

The segmented image is used for Feature Extraction. Feature extraction is employed to efficiently
represent the image by capturing the important parts of the image. In this paper, we have extracted the
features using DWT and GLCM.
The DWT Wavelet transform is an effective one to represent the image. In this DWT process, analysis
filter bank and decimation techniques analyse the image. This filter bank contains low and high pass
filters pairs. The detailed information of the image is extracted by low pass filter and the information
about edges are extracted by high pass filter. The multi-level determination of the image is permitted by
DWT. The goal is to extract relevant data from an image. DWT Wavelet transform samples the wavelets
at discrete intervals. At the same time, it provides information about spatial and frequency of the image.
GLCM means gray level cooccurrence matrix. This function distinguishes the texture of the image by
determining how frequently a pair of pixels with some values and in a defined spatial relationship exist
in an image. In this, GLCM is created using Gray-comatrix and the statistical measures are extracted
using Graycoprops. The extracted features include Contrast, Energy, Correlation, Smoothness, Standard
Deviation, Homogeneity, Entropy and Mean.

3. Classification
Image classification is the technique of classifying or predicting the class of a particular object in an
image based on certain rules. The aim of this process is to correctly detect the features in an image.
Supervised and unsupervised methods are the two common methods in Classification. SVM is a
supervised machine learning algorithm and is well suited for classification techniques. SVM means
Support Vector Machine. The aim is to transform non-linear into linear transformation using SVM
kernel function [5] [7]. In this work, we have used the linear kernel function for classification. To make
the classification process convenient, non-linear samples can be converted into a high dimensional
future space. SVM is made as the default choice for classifying brain tumour by the features selected
with kernel class separability. The confusion matrix denoting the terms True Positive, False Positive,
True Negative and False Negative from the expected results and ground truth result is created for
accuracy calculation.

Accuracy = TP + TN
TP + TN + FP + FN

6
ICBSII-2022 IOP Publishing
Journal of Physics: Conference Series 2318 (2022) 012042 doi:10.1088/1742-6596/2318/1/012042

The whole process of classifying brain tumour image is distinguished into two stages namely training
and testing. From the image dataset consists of 15 images, 10 images are chosen for training the
classifiers and 5 images are chosen to test the performance of classifiers. For training purpose, the feature
extraction method Gray Level Cooccurrence Matrix (GLCM) is used.
In testing stage, the statistical values of trained images are used for brain tumour classification. From
this image dataset, another set of brain tumour images is chosen for testing. Using the above specified
method GLCM, statistical features are extracted for the testing images. From the set of trained and test
values, the SVM classifier classifies the brain tumour with images as benign and malignant with the
accuracy rate of 80%.

4. Results and discussion

For Segmentation, tumour detection method based on Threshold and K-means clustering techniques
were proposed. Thresholding techniques segment the tumor area by its intensity value in MRI brain
images. Otsu method is excellent for image which histogram is bimodal distribution and, in this
technique, single or one threshold value is used for the full image. But in local thresholding technique,
unique or specific threshold values is used for the partitioned sub images obtained from full image. K-
means clustering algorithm evaluates different pixels and groups the similar pixels together. Since K-
means clustering is convenient for large datasets and it works well on spherical cluster and efficient to
make change. Therefore, K-means clustering algorithm is a suitable method to segment brain MR
images.
The goal of SVM algorithm is to find the hyperplane which distinguishes the data points. The dimension
of this hyperplane is based upon the number of features. This technique is done on the basis of margin
maximization principle. In the goal of achieving perfect generalization performance, it performs
structural risk minimization. With an accuracy rate of 80% the SVM classifier classified the brain
tumour images as Benign and Malignant as shown in Figure 7 and 8.

Figure 7. Image detected as benign tumour using SVM Classification

7
ICBSII-2022 IOP Publishing
Journal of Physics: Conference Series 2318 (2022) 012042 doi:10.1088/1742-6596/2318/1/012042

Figure 8. Image detected as malignant tumour using SVM Classification

5. Conclusion
There are different types of tumours. It may be benign or malignant. K-means clustering algorithm plays
a significant role in extracting the tumour from the brain tumour image. The dataset consists of MRI
(Magnetic Resonance Imaging) brain tumour images. The brain tumour images, used in this image
processing techniques were taken from publicly available sources. The dataset was divided into training
dataset and testing dataset. In this paper, we have automated the analysing procedure for brain tumour
detection through image processing techniques. Apart from existing brain tumour classification
methodology present, our work has proved to provide an overall accuracy rate by upto 80%. All the
steps for detecting brain tumour have been discussed starting from MRI image dataset, pre-processing
to classification of the tumour using three segmentation techniques. Pre-processing involving operations
like grayscale transformation and noise removal using median filter have been done. After the image
quality improvement and noise reduction discussed here, segmentation methodology for MRI image of
brain tumour has been used. The segmentation methods used include, Otsu binary segmentation, Local
thresholding and K means clustering algorithm. Otsu method is excellent for image which histogram is
bimodal distribution and, in this technique, single or one threshold value is used for the whole image. In
local thresholding technique, unique or specific threshold values is used for the partitioned sub images
obtained from full image. K-means clustering algorithm evaluates different pixels and groups the similar
pixels together. Since K-means clustering is convenient for large datasets and it works well on spherical
cluster and efficient to make change, we conclude that K means is suitable method to segment MRI
images. Feature extraction is a process of dimensionality reduction and is followed by segmentation.
The extracted feature includes Contrast, Energy, Correlation, Homogeneity, Mean, Standard deviation,
Entropy, Smoothness and Variance. Classification was done by using SVM. SVM classification method
was able to detect the presence or absence of tumour, and if the presence of tumour was detected, it was
able to determine whether the type is benign or malignant.

References
[1] Ali HM. A new method to remove salt & pepper noise in Magnetic Resonance Images.
In2016 11th International Conference on Computer Engineering & Systems (ICCES) 2016
Dec 20 (pp. 155-160). IEEE.

8
ICBSII-2022 IOP Publishing
Journal of Physics: Conference Series 2318 (2022) 012042 doi:10.1088/1742-6596/2318/1/012042

[2] Ravikumar Gurusamy, Dr Vijayan Subramaniam. A Machine Learning Approach for MRI
Brain Tumor Classification. In2017 Tech Science Press CMC, vol.53, no.2, pp.91-108, 2017.

[3] Amin J, Sharif M, Haldorai A, Yasmin M, Nayak RS. Brain tumor detection and classification
using machine learning: a comprehensive survey. Complex & Intelligent Systems. 2021 Nov
8:1-23.

[4] Islam MK, Ali MS, Miah MS, Rahman MM, Alam MS, Hossain MA. Brain tumor detection in
MR image using superpixels, principal component analysis and template-based K-means
clustering algorithm. Machine Learning with Applications. 2021 Sep 15; 5:100044.

[5] Rao CS, Karunakara K. Efficient Detection and Classification of Brain Tumor using Kernel
based SVM for MRI. Multimedia Tools and Applications. 2022 Jan 26:1-25.

[6] Ganesh PS, Kumar TS, Kumar M, Kumar MS. Brain Tumor Detection and Classification
Using Image Processing Techniques. Brain. 2021 Apr;4(3).

[7] Warunkar I, Darzi S, Shetty P, Shaikh S, Ghadge S. Detection of Brain Tumor Using Image
Processing and Machine Learning. EasyChair; 2021 Jun 16.