INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 9, ISSUE 04, APRIL 2020 ISSN22778616

Brain Tumor Detection Using Machine

Learning
M. Aarthilakshmi, S. Meenakshi, A. Poorna Pushkala, V. Rama@Ramalakshmi, N.B. Prakash.
Abstract : Medical imaging is gaining importance with an increase in the demand for automated, reliable, fast and efficient diagnosis which can provide
insight into the image better than human eyes. The brain tumor is the second leading cause for cancer-related deaths in men age 20 to 39 and leading
cause cancer among women in the same age group. Brain tumors are painful and should end in various diseases if not cured properly. The diagnosis of
the tumor is a very important part of its treatment. Identification plays an important part in the diagnosis of benign and malignant tumors. A prime reason
behind a rise in the number of cancer patients worldwide is the ignorance towards the treatment of a tumor in its early stages. This paper discusses such
a machine learning algorithm that can write the user about the details of the tumor using brain MRI. These methods include noise removal and
sharpening of the image along with basic morphological functions, erosion, and dilation, to obtain the background. Subtractions of background and its
negative from different sets of images result in extracted in age. Plotting contour and c-label of the tumor and its boundary provides us with information
related to the tumor that can help in a better visualization in diagnosing cases. This process helps in identifying the size, shape, and position of the
tumor. It helps the medical staff as well the patient to understand the seriousness of the tumor with the help of different color-labeling for different levels
of elevation. A GUI for the contour of the tumor and its boundary can provide information to the medical staff on the click of user choice buttons.
Keywords: classification, convolutional neural network, feature extraction, machine learning, magnetic resonance imaging, segmentation, texture
features.
——————————  ——————————
1 INTRODUCTION the brain MR image and then a modified FCM algorithm is
The physical body consists of many sorts of cells. Each cell used for brain tumor detection. The average speed-ups of as
features a specific function. The cells in the body grow and much as 80 times a traditional FCM algorithm are obtained
divide in an orderly manner and form some new cells. These using the modified FCM algorithm. The modified FCM
new cells help to keep the human body healthy and properly algorithm is a fast alternative to the traditional FCM technique.
working. When some cells lose their capability to regulate their Rajesh and malar proposed brain MR image classification
growth, they grow with none order. The extra cells formed form based on rough set theory and feed-forward neural network
a mass of tissue that is named as the tumor. The tumors can classifier. The features are extracted from MR images using
be benign or malignant. Malignant tumors lead to cancer while rough set theory. The selected features are fed as input to
benign tumor is not Cancerous. The important think about the feed forward neural network classifier which differentiates
diagnosis includes the medical image data obtained from between the normal and abnormal brain and the accuracy of
various biomedical devices that use different imaging about 90% is obtained. Ramteke and monali proposed
techniques like x-ray, CT scan, MRI. Magnetic resonance automatic classification of brain MR images in two classes
imaging (MRI) may be a technique that depends on the normal and abnormal based on image features and automatic
measurement of magnetic flux vectors that are generated after abnormality detection. The statistical texture feature set is
an appropriate excitation of strong magnetic fields and obtained from normal and abnormal images and then the KNN
radiofrequency pulses in the nuclei of hydrogen atoms present classifier is used for classifying an image. The KNN obtain an
in the water molecules of a patient's body. The MRI scan is 80% classification rate. Othman proposed a probabilistic
much better than the CT scan for diagnosis as it doesn't use neural network technique for brain tumor classification. Firstly,
any radiation. The radiologists can evaluate the brain using the features are extracted using the principal component
MRI. The MRI technique can determine the presence of analysis (PCA) and the classification is performed using
tumors within the brain. The MRI also contains noise caused probabilistic neural network (PNN). Jafari and shafaghi
thanks to operator intervention which may cause inaccurate proposed a hybrid approach for brain tumor detection in MR
classification. The large volume of MRI is to analyze; thus, images based on support vector machines (SVM). The texture
automated systems are needed because they're less and intensity features are used. The accuracy of about
expensive -. Automated detection of tumors in MR images is 83.22% is achieved and is more robust. Thus, from an
important as high accuracy is required when handling human extensive literature survey, we found that most of the current
life. The supervised and unsupervised machine learning brain tumor detection system uses texture, symmetry, and
algorithm technique can be employed for the classification of intensity as features. Texture features are important property
brain MR image either as normal or abnormal. During this of the brain as texture perception has a very important aspect
paper, an efficient automated classification technique for brain in the human visual system of recognition and interpretation.
MRI is proposed using machine learning algorithms. The Further, we propose the use of the ml algorithm to overcome
supervised machine learning algorithm is used for the drawbacks of traditional classifiers. We investigate the
classification of brain MR image performance of a machine learning algorithm namely CNN in
this work. Neural networks are useful as they can learn
2 RELATED WORK complex mappings between input and output. They are
Joshi proposed brain tumor detection and classification capable of solving much more complicated classification tasks.
systems in MR images by first extracting the tumor portion
from brain image, then extracting the texture features of the 3 PROPOSED WORK
detected tumor using gray level co-occurrence matrix (GLCM) As per the literature survey, it was found that automation of
and then classified using a neuro-fuzzy classifier. shasidhar brain tumor detection is very essential as high accuracy is
proposed a modified fuzzy c-means (FCM) algorithm for MRI needed when human life is involved. Automated detection of
brain tumor detection. The texture features are extracted from tumors in MR images involves feature extraction and
1976
IJSTR©2020
www.ijstr.org
 INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 9, ISSUE 04, APRIL 2020 ISSN22778616

classification using a machine learning algorithm. In this paper,

a system to automatically detect a tumor in MR images is
proposed as shown in the figure.

Fig.1.Proposed work

Fig.3.MRI Data Set

 Validation set - is the set used during the model

training to adjust the hyperparameters.
 Test set - is the small set that doesn't touch for the
whole training process at all. It's been used for final
model performance evaluation these are some
sample images of both classes

The images have different width and height and the different
sizes of "black corners". Since the image size for the vgg-16
input layer is (224,224) some wide images may look weird
after resizing. The first step of "normalization" would be to crop
the brain out of the images.

Fig.2.Block diagram

3.1 Image Acquisition

The image data that was used for this problem is brain mri
images for brain tumor detection. It consists of mri scans of
two classes: Fig.4.Normalised images
 No - no tumor, encoded as 0
 Yes - tumor encoded as 1 3.2 Preprocessing
All images are in one folder with yes and no subfolders. I Preprocessing is required because it provides an improvement
will split the data into train, val and test folders which makes it in image data which reinforces a number of the image features
easier to work with the same dimension of images. which are important for further processing.

TABLE 1. set of folders of images

No. Of images Folder directory
253 Train
25 Test
50 Validation

Fig.5.Processed images

1977
IJSTR©2020
www.ijstr.org
 INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 9, ISSUE 04, APRIL 2020 ISSN22778616

The pre-processing steps that are applied to the MR image are of ml algorithms is to automatically learn and make intelligent
as follows: the RGB MR image is converted to grayscale decisions the classification is done based on the below
image and then the median filter is applied for noise removal features:
from brain MR images. The noise is to remove for further (a) feature processing: before classification, the extracted
processing as high accuracy is needed. Then edges are feature set can be refined to make it more appropriate for
detected from a filtered image using canny edge detection as achieving high classification accuracies:
shown. The edge detected image is needed for the (b) classifier training: pixels that are labeled as normal and
segmentation of the image. The segmentation aims to change abnormal are used with the extracted features to automatically
the representation of an image into something easier to learn a classification model that predicts labels based on the
analyze. features;
(c) pixel classification: the learned classification model can
3.3 Segmentation then be used to predict the labels for pixels with unassigned
Brain tumor segmentation is the process of separating the labels, based on their extracted features;
tumor from normal brain tissues; in clinical routine, it provides (d) relaxation: since the learned classification model may be
useful information for diagnosis and treatment planning. noisy, a relaxation of the classification results which takes into
However, it's still a challenging task thanks to the irregular account dependencies in the labels (i.e. Classification) of
form and confusing boundaries of tumors. Tumor cells neighboring pixels can be used to refine the classification
thermally represent a heat source; their temperature is high predictions and yield a final segmentation.
compared to normal brain cells. The most aim of this paper is This CNN method requires only a small amount of training
to demonstrate that thermal information of brain tumors is data to estimate the parameters which are needed for
often wont to reduce false positive and false negative results classification. The time taken for training and classification is
of segmentation performed in MRI images. The obtained less. This can extract useful attributes from trained weights by
results in all patients showed a significant improvement using feeding data by levels and tune CNN for the specific task
the proposed method compared to segmentation by a level set
method with an average of 0.8% of the tumor area and 11 4 EXPERIMENTAL RESULTS
2.48% of healthy tissue was differentiated using thermal Accuracy as a metric to justify the model performance which
images only. We conclude that tumor contours delineation can be defined by accuracy (ACC), sensitivity (SE), specificity
based on tumor temperature changes can be exploited to (SP):
reinforce and enhance segmentation algorithms in MRI
diagnostics.

3.4 Feature Extraction

(a) image-based features: the extraction of features based on
the image data, potentially including intensity features, texture
features, histogram-based features, and shape-based
features;

Fig.7.Experimental Results

Fig.6.Feature Extraction

(b) coordinate-based features: the extraction of features

based on the registration to a standard coordinate system,
potentially including coordinates features, spatial prior
probabilities for structures or tissue types in the coordinate Final results look as follows:
system, and local measures of anatomic variability within the
coordinate system; (c) registration-based features: the
extraction of features based on known properties of the one or
more aligned templates, potentially including features based
on labeled regions in the template, image-based features at
corresponding locations in the template, features derived from Total ACC=94%;
the warping field, and features derived from the use of the
template's known line of symmetry.

3.5 Classification
Machine learning algorithms are used for the classification of
Val set ACC=92%; SE=93%;SP=89%
MR brain images either as normal or abnormal. The major aim
1978
IJSTR©2020
www.ijstr.org
 INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 9, ISSUE 04, APRIL 2020 ISSN22778616

of brain data. In: information processing in medical

imaging. Springer, pp. 17–29.
[9] Erihov, m., Alpert, s., Kiselev, p., Hashoul, s., 2015. A
cross saliency approach to asymmetry-based tumor
detection. In: medical image computing and computer-
Test set ACC=91%;SE=94%;SP=87%
assisted intervention–miccai 2015. Springer, pp. 636–643.
[10] Simonyan k., Zisserman a., 2014. Very deep convolutional
networks for large-scale image recognition. Arxiv preprint
arxiv:1409.1556

[11] M.Aarthilakshmi, is studying final year, Electrical and

Electronics Engineering in National Engineering College,
Kovilpatti.
a. ([email protected])
[12] S. Meenakshi, is studying final year, Electrical and
Electronics Engineering in National Engineering College,
Kovilpatti.
a. ([email protected])
[13] A.Poorna Pushkala, is studying final year, Electrical and
Fig.8.Accuracy for respective set Electronics Engineering in National Engineering College,
Kovilpatti.([email protected])
5 CONCLUSION [14] V.Rama@Ramalakshmi, is studying final year, Electrical
In this proposed work different medical images like MRI brain and Electronics Engineering in National Engineering
cancer images are taken for detecting tumor. The proposed College, Kovilpatti. ([email protected])
approach for brain tumor detection supported convolution [15] Dr.N.B.Prakash is working as Associate Professor in
neural network categorizes into multi-layer perceptron neural Electrical and Electronics Engineering at National
network. The proposed approach utilizes a mixture of this Engineering College, Kovilpatti.([email protected])
neural network technique and consists of several steps
including training the system, pre- processing, implementation
of the tensor flow, classification. In the future, we'll take an
outsized database and check out to offer more accuracy which
can work on any sort of MRI brain tumor.

6 REFERENCES
[1] Pankaj sapra, Rupinder pal Singh, Shivani Khurana,
"brain tumor detection using neural network", international
journal of science and modern engineering, ijisme, issn:
2319-6386, volume-1, issue-9, august 2013.
[2] Prachi gadpayleand, p.s. Mahajani, "detection and
classification of brain tumor in MRI images ", international
journal of emerging trends in electrical and electronics,
ijetee – issn: 2320-9569, vol. 5, issue. 1, july-2013
[3] Bakes.s., Reyes.m., menze, b.: identifying the best
machine learning algorithms for brain tumor segmentation,
progression assessment, and overall survival prediction in
the Brats challenge. In: arxiv:1811.02629 (2018)
[4] Bauer s., Nolte, Reyes.m., 2011. Fully automatic
segmentation of brain tumor images using support vector
machine classification in combination with hierarchical
conditional random field regularization., in miccai, pp.
354–361.
[5] Prastawa m., Bullitt e., Greig, g., 2004. A brain tumor
segmentation framework based on outlier detection.
Medical image analysis 8, 275–283.
[6] Goodfellow, Warde-Farley, d., mirza, m., Courville, a.,
bengio, y., 2013b. Maxout networks, in icml.
[7] Gotz m., Weber c., Blocher j., Stieltjes b., Meinzer, h.p.,
Maier-Hein, k., 2014. Extremely randomized trees-based
brain tumor segmentation, in proc of brats challenge -
miccai.
[8] Cardoso, m. J., Sudre, c. H., Modat, m., Ourselin, s.,
2015.a template-based multimodal joint generative model
1979
IJSTR©2020
www.ijstr.org