Automated Intracranial Neoplasm Detection Using Convolutional Neural Networks

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 10 Issue: 01 | Jan 2023 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 546
Automated Intracranial Neoplasm Detection Using Convolutional
Neural Networks
Mr. P. Veeresh Kumar1, N. Geetha Gayathri2, G. Harika3, Ch. Sai Poojitha4, G. Vyshnavi5
T. Lakshmi6
1 Associate Professor Department of Information Technology, KKR & KSR Institute Of Technology And
Sciences(A), Guntur, India
2,3,4,5,6 Undergraduate Students ,Department of Information Technology , KKR & KSR Institute Of Technology And
Sciences(A), Guntur, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - A brain tumor is nothing but an abnormal
collection of cells in the intracranial. The skull that surrounds
the brain is very hard, so when a tumor grows insidethebrain,
it puts pressure on the skull and can cause serious damage. It
is regarded as one of the most dangerous diseases in both
children and adults. About 11,700 people are diagnosed with
brain tumors each year. It accounts for 85–90% of all central
nervous system tumors. The 5-yearsurvivalrateformalignant
brain tumors is approximately 34% for men and 36% for
women. Accurate diagnosis, detection of brain tumors , and
early initiation of appropriate treatment are essential to
prolonging a patient's life expectancy. Brain tumor
segmentation is one of the most importantandlaborioustasks
in the field of medical imaging. Manual classification with
human assistance can lead to inaccurate predictions and
diagnoses. Additionally, this is a daunting task when there isa
large amount of data that needstobesupported. Braintumors
are highly variable in appearance, and there are similarities
between tumor tissue and normal tissue, thus making the
extraction of tumor regions from images difficult. There are
different types of imaging tests, such as X-rays, MRIs, and CT
scans, used to detect brain tumors. In this study, computed
tomography (CT) scan images are used to identify brain
tumors. Deep learning (DL) is the latest technology that
provides more efficient results in detection and classification.
In this paper, a model is developed by detectingtumorsusinga
convolutional neural network. In our study, CNN achieved
97.87% accuracy.
Key Words: Brain Tumor, Magnetic Resonance Image
(MRI), Convolutional Neural Networks (CNN), Deep
Learning, Tensor Flow, Keras.
1. INTRODUCTION
We use medical imaging technology to get a glimpse inside
the human body. The most difficult task is to diagnose and
classify tumors or cancers. According to statistics, brain
tumors are one of the most important types of cancer in
terms of mortality. The International AgencyforResearchon
Cancer (IARC) reports that worldwide, more than one
million people are diagnosed with 4,444 brain tumors each
year, and the mortality rate is steadily increasing year by
year. Those under the age of 34 To date, doctors have
implemented over 4,444 advancedmethodstoidentifymore
painful tumorsinpatients.UsingCT(computedtomography)
and MRI (magnetic resonance imaging) scans, doctors can
now dissect abnormalities in different parts of the body.
Brain tumors have recently gained momentum due to the
increased demand for labour savings and the clinical
evaluation of large amounts of medical data. This type of
image processing and analysisinvolvescomplex calculations
of data as well as visualizing and understanding that data.
Brain tumors can originate from abnormal, precancerous,
cancerous, harmful, harmless, or non-cancerous cells that
form in the brain. He divides malignant tumors into two
types: those that originate in the brain and subsequent
cancerous growths that can spread to different areas of the
human body. Under these conditions, cancer metastasizes
within the patient's body and is said to have a very low
survival rate and a high mortality rate.
1.1 Brain Tumor
The brain is a soft tissue mass that resembles a sponge that
is covered in bone. The brain is lined with a delicate layer of
tissue, including cerebrospinal fluid. Between the brain and
the ventricles, there is cerebrospinal fluid. The cerebrum,
cerebellum, and brainstem are the three components that
make up the brain. The brain reacts to external stimuli. The
cerebellum maintains the body's equilibrium. The brain and
spinal cord are linked by the brainstem. A brain tumor is
recognized as an overgrowth of the numerous cells that
make up the brain. Primary and secondary cells are both
present. While secondarycellsaresparedbymetastasisfrom
other parts of the body to the brain, primary cells comefrom
the brain itself. Some metastases spread rapidly. Some
metastases spread fairly quickly and exhibit symptoms like
seizures, speech difficulties, nausea, and blurred vision. It
can be fatal if not treated in a timely manner. This is due to
EDA, which is brought on by decreased blood flow and the
subsequent deterioration of healthy tissue as a result of the
skull's constrained interior and growing intracranial
pressure. Electronic healthcare systems and magnetic field
information technology aid clinical professionals in giving
patients better care. A brain tumor can affect other nearby

structures and is distinguished by its position, image
intensity, shape, and size. If the tumor is accompanied by
emphysema, the intensity of other nearby areas is also
changed. Benign tumor and malignant tumor are the two
tumors . Due to their fixed size and slow growth, benign
tumors are not regarded as harmful. The tissues around
benign tumors are unaffected by them. A birthmark is an
illustration of a benign tumor. Cancer can be treated when it
is in the premalignant stage, whereas ignorance can cause
cancer. Cancer is the name for a cancerous tumor. They
begin to grow around nearby tissue after some time has
passed since their generation. Malignant tumors can befatal
and grow quickly. For patients, early detection of malignant
tumors is very beneficial. To achieve the highest level of
accuracy, various techniques have been investigated for the
automated detection and classification of malignancies.
Furthermore, hospitals are able to achieve extremely low
false-negative rates, which raises the demand for computer
support. Most observers can identify the presenceofspecific
characteristics that identify malignancy. MRI is frequently
used to recognize and highlight specific internal body
structures. represents magnetism. imagingusingresonance.
A strong magnetic field and body radio waves are used in an
MRI to produce precise images. Due to its ability to achieve
high spatial resolution and better soft-tissue contrast, it is a
crucial and useful tool for the accurate diagnosis of disease
as well as for the treatment and monitoring of disease using
imaging modalities.
Segmentation of magnetic resonance (MR) images of the
brain is key to diagnosing malignancies and other
neurological abnormalities. Image segmentation is a
preprocessing step that helps divide an image into smaller
segments. Images are divided by classes or subsets. This is
useful because the segmentation allows us to focus on
regions of interest. Segmentation of MRI images can also be
performed manually, but it is time-consuming and less
accurate.
Glioma Meningioma Pituitary
No Tumor
2. LITERATURE REVIEW
[1] K. B. Vaishnavee and K. Amshakala proposed an
Automated MRI brain image segmentation and tumor
detection using SOM-clusteringandProximal SupportVector
Machine classifier. Photograph segmentation is a critical
approach noticeably used to extract suspicious parts from
clinical photographs, which include MRI, CT scan,
mammography, and so forth. With this motivation, SOM
clustering is proposed in this paper for image segmentation
of brain MRI scans. This proposed work proposes effective
segmentation and classification using HFS-SOM and PSVM.
After segmentation, the resulting image is given as input to
his PSVM classifier, followed by feature extraction and
selection using GLCM-PCA. Experimental results show that
the proposed system has a high accuracy rate and a low
error rate. In the future, the system could be improved to
support other types of cancer images with a few changes at
both the segmentation and classification stages. We need to
support a large number of inputs, and we need to improve
the accuracy rate. To achieve this, swarm-based feature
selection can add more features to improve tumor detection
and classification results.
[2] G. Birare and V. A. Chakkarwar developed an Automated
Detection of Brain Tumor Cells Using Support Vector
Machine. Thousands of people receive a brain tumor
diagnosis each year. While it is of significant importance to
separate, find, and extract contaminatedtumorregionsfrom
magnetic resonance imaging for diagnostic evaluation, it is
time-consuming and prone to human error. To get around
these restrictions, segmentation and classification
techniques that are automated and semi-automatic are now
being deployed. As a result, we created automatic
segmentation using the K-means technique and a brain
tumor identification module that uses a support vector
machine classifier to identify tumors while also extracting
Otsu-threshold characteristics from the output segmented
image using a grayscale occurrence matrix. The accuracy of
the experimental results was 98.51%. Overall,thisapproach
produces excellent outcomes. Future work on medical

diagnostics may involve the development of additional
algorithms, like multiclass SVM.
[3]A. Kumar, A. Ashok and M. A. Ansari introduced Brain
Tumor Classification Using Hybrid Model Of PSO And SVM
Classifier. Medical imaging is a very important process for
the diagnosis of any disease today. MRI is usually used to
detect the presence and type of tumor. The process of
classifying brain tumors is very complicated. There are
various procedures in medical image processing, such as
image segmentation, image extraction, and image
classification. Different types of features are extracted from
segmented data. MRI images, including intensity, shape,and
texture-based features. In this paper, machine learning was
used to create an online database of MRI images of brain
tumors. The particle swarm algorithmwasusedtocreatethe
model optimization (PSO) algorithm for feature selection,
and then a support vector machine (SVM) classifier was
used. used to classify tumor types on current brain MRI
images .Using the PSO-SVM classification model gives
95.23% accuracy when all 14 features are included,
compared to 86.82% for the naive SVM classifier. Similarly,
the specificity is 94.8% and the sensitivity is 100%. In the
future, we hope to apply this concept to bio-inspired
algorithms and compare them all to different medical
images.
[4]R. Ahmmed, A. S. Swakshar, M. F. Hossain and M. A. Rafiq
developed Classification of tumors and it stagesinbrain MRI
using support vector machine and artificial neural network.
Cells are the smallest unit of tissue, and their abnormal
growth causes tumors in the brain. In this study, support
vector machine (SVM) and artificial neural network (ANN)-
based classification of tumors and stage in brain MRIimages
is presented. An integration of temperature-based K-Means
and modified fuzzy C-Means (TKFCM) clustering algorithms
is used to segment MRI images based on the grayscale
intensities of small portions of brain images. The K value of
the temperature-based K-Means algorithm is higher than
that of conventional K-Means,andtheautomaticallyupdated
membership of FCM eliminates the contour problem of
tumor area detection. There is a second method used to
classify tumors into benign and four malignant stages using
ANN. The accuracy of classification of normal and tumor
brains by this proposed method is up to 97.37% with a bit
error rate (BER) of 0.0294, which is superior to other
methods.
[5]C. Saha and M. F. Hossain designed MRI brain tumor
images classification using K-means clustering, NSCT and
SVM. Magnetic resonance imaging (MRI) has justified its
potential as a novel imaging tool for the anatomical study of
brain tissue. Manual classification of a large number of
patient MRI brain images is a very time-consuming process
and prone to error; therefore, there is a need to develop a
fast and accurate automatic schemeforclassifyingMRIbrain
images. In this article, we use K-means clustering, the non-
subsampled contourlet transform (NSCT), and support
vector machines (SVM) to automatically classify MRI brain
images as normal if no tumor is present and abnormal if a
tumor is present. We present a scheme for categorizing The
preprocessing stage of this proposed scheme uses a median
filter to remove noise and improve the resolution of MRI
brain images. Then, using K-means clustering, segment the
MRI image of the brain. Due to the salient properties of
NSCT, such as multi-scale, multi-orientation, and shift
invariance, NSCT is applied to segmented images. Seven
features are then extracted from the NSCT sub-band
coefficients, and these features are applied to a support
vector machine for the classification of MRI brain images.
The effectiveness of this method is evaluated in terms of
sensitivity, specificity, and accuracy, and this method
achieves 98.86% classification accuracy on GRBkernel SVM.
This scheme has been compared with other classification
schemes for MRI brain images and provides satisfactory
results, demonstrating the efficacy of our scheme. However,
further validation of the proposed hybrid scheme is needed
with further testing and training on MRI brain images from
different databases and consideration of other performance
evaluation parameters to be addressed in future work.
3. SYSTEM REQUIREMENTS SPECIFICATION
3.1 Software Requirements
 Operating System: any Windows OS
 Libraries: Keras, Tensor Flow, Numpy, Scikit-Learn
Matplotlib, OpenCV
 Editor: Collab Notebook
 Technologies: Python
3.2 Hardware Requirements
 Processor: i3
 RAM: 4 GB
 Hard disk: 512GB
3.2.1 Hardware Configuration
 Processor: Intel core i5 or above.64-bit, quad-core,2.5GHz
minimum per core .
 Ram: 4 GB or more .
 Hard disk: 10 GB of available space or more.
 Display: Dual XGA (1024 x 768) or higher resolution
monitors .
 Operating system: Windows
3.2 Functional Requirements
 Python 3.6.2 or later, PIP, and NumPy 1.13.1 for Windows.
 Pip
 NumPy
 Pandas

 Anaconda
 Jupyter Notebook
 Tensor Flow
 Keras
4. METHODOLOGY
The proposed procedure is divided into various stages and
each stage is explained in detail.
 Image input
 Image Preprocessing
 Splitting Data for Training and Testing
 Classifying Brain Tumor Images Using CNNs
 Obtaining Outputs and Validating Models
Fig - 1 : Flowchart of proposed method
Image Input
The input is MRI image. The images used here are in .jpg
format. The MRI image dataset is obtained from publicly
available sources. These images are dividedintotwoyesand
no folders, each containing images with and without brain
tumors.
The training and testing both datasets containsimages with
and without tumors. Later we train our system using
Training dataset and we test our model on test dataset we
must not test on the same train dataset if we do sothere may
be chances of getting inaccurate results so we test on testing
dataset
The Dataset of brain with tumor:
The Dataset of brain without tumor:
Image Preprocessing
MRI images acquired from a public sourcecannotbedirectly
supplied for processing. These images contain noise,suchas
external noise and noise generated by the patient’s
movement during the MRI scan. For efficient brain tumor
detection, they must be removed and images enhanced.
• Grayscale conversion: imported images in jpg format are
first converted from RGB models to grayscale images.

• Image resizing: Grayscale images are resized from their
current size to a convenient 200 x 200 size for consistency.
• Median Filtering: The resized grayscale image is median
filtered to remove existing noise and reduce edge blurring
effects.
• High-pass filtering: The median filtered image is high-pass
filtered and enhanced by adding a resized image to the
resulting image.
Splitting Data for Training and Testing
Testing the model on the same data it was trained on
will result in overfitting and poor performance in real-
world scenarios.To avoid this, split the data into twoparts:a
training set and a test set. Normal practice is an 80/20 split.
In this, we split the dataset into training and testing sets
using the train_test_split() function from the scikit-learn
package. The sample size chosen is 0.2 (i.e., 20% for testing
and 80% for training). Classification,likeanykindofmedical
image, is the best way to identify an image. All classification
algorithms are based on predictions of images in which one
or more features are present, and each of these features
belongs to one of several classes. An automatic and reliable
convolutional neural network (CNN)classificationmethodis
used. This makes the structure robust and helps identify
details. Convolutional Neural Network (ConvNetorCNN)isa
deep learning algorithm that takes in an input image and
assigns importance to different aspects or objects in the
image to distinguishthemfromeachother.ConvNetrequires
less preprocessing when compared to other classification
algorithms. While the primitive method requires manual
development of filters, ConvNet has the ability tolearnthese
filters and properties with sufficient training. ConvNet can
successfully capture the spatial and temporal dependencies
of images by applying relevant filters. This architecture
improves matching with imagedatasetsbecausethe number
of relevant parameters is reduced and the weights are
reusable. In other words, the network can be trained to
better understand image sophistication. The role of the
ConvNet is to reduce the image to a manageable form
without losing important features necessary to get good
predictions. In this step, we need to import Keras and other
packages that we will use when building CNN.
Sequential
 To initialize the neural network, wecreateanobject
of the Sequential class.
 classifier = Sequential ()
Convolution
 To add a convolutional layer, call the Add function
on the classifier object and pass the parameters to
Convolution2D. The first argument for a feature
detector is the number of feature detectors to
create. The second and third parameters are the
dimensions of the feature detector matrix. We used
256 feature detectors for the CNN. The next
parameter is the input shape, which is the shape of
the input image. Images are converted to this
format during preprocessing.
 If the image is black and white, it is converted to a
2D array; if the image is colored, it is converted to a
3D array. In this case, we assume that you are
working with color. The input shape is a tuple that
contains the number of channels (three for colour
images) as well as the dimensions of each channel's
2D array.
 If you are not using a GPU, it is recommended to use
lower dimensionality to reduce computation time.
The last parameter is the activation function. The
classification of the problem of images is nonlinear.
Therefore, we usea commutationfunctionto ensure
that we do not have negative pixel values during
computation. This is how the nonlinearity is
achieved.
Pooling
Pooling Layer: This is another layer where the matrix starts.
This layer is added to reduce the number of parameters in
the matrix. This prevents overfitting the model. There are
two types of pooling.
MAX_POOLING: Get the maximum value of the patch here.
AVG_POOLING: Here we get the averagevalueofthepatches.
Flattening
 All pooled feature maps are collectedandcombined
in this step and placed in a single vectorfor input to
the next layer.
 The flatten function flattens all feature maps into
one long column.
 classifier.add(Flatten())
Dropout Layer
This is an optional layer that can be added to the model.This
layer prevents overfitting by omitting some units.
Fully Connection
 The next step is to use the vectors obtained above
as inputs to the neural network using the Keras
Dense function. The first parameter is printed. This
is the number of nodes in the hidden layer.Through
experimentation, we can determine what is the

most appropriate number. The higher the number
of dimensions, the more computational resources
are required to fit the model. A common method is
to choose the number of nodes in powers of two.
 classifier. add(Dense(output=64))
 The next layer we need to add is the output layer. In
this case, we expect a binary result, so we use the
sigmoid activation function. Ifyouexpectmorethan
one result, use the SoftMax function.
 The output here is 1. This is because we only expect
the predicted probabilities of the classes.
 classifier.add (high density (output = 1, activation=
'sigmoid'))
Obtaining outputs and validating models
This will give you the output when the image is given as
input, along with a percentageofconfidenceandinformation
about whether a tumor is present or not.
5. RESULTS
In methodology, we are having various stages inthissection,
and the output of each stage will be described.
In first stage we successfully converted the input image (an
OpenCV image) to grayscale using edge detection after
applying pre-processing. Edge-based segmentation on an
image classifies pixels as edge or non-edge dependingonthe
filter output. In region-based segmentation, the image was
subjected to regional maxima in order to group adjacent
pixels with similar values. The CNN model was trained and
tested before the K-meanstechniquewasusedtoidentify the
region of interest (ROI), which was then examined for the
presence of a tumor.
Fig - 2 : Image Preprocessing
Original Image Cropped Image
Fig - 3: CNN model architecture
Fig - 4: Training model

Fig - 5: Training and Validation Accuracy
Fig - 6: Output with confidence percentage for no tumor
Fig - 7: Output with confidence percentage for image
having tumor
Fig - 8:Accuracy comparison with other methods
6. CONCLUSIONS
In this paper, we used CNNS to detect brain tumors, and we
successfully implemented all features and testeditacross all
possible test cases with satisfactory results.Themaingoal of
this project is to improve the accuracy with minimal
computation time and less complexity, to achieve high
validation accuracy and very low validation loss, and to
reduce the time required to detect brain tumors. Compared
to previously proposed algorithms ithasmoreaccuracyand
detects tumor in less time. It is about 97.15% accurate,
whereas SVM is only 88% accurate. Finally, the goal of this
project is achieved by using CNN.
7. REFERENCES
[1] K. B. Vaishnavee and K. Amshakala, "An automated MRI
brain image segmentation and t tumor detection using
SOM-clustering and Proximal Support Vector Machine
classifier," 2015 IEEE International Conference on
Engineering and Technology (ICETECH), Coimbatore, India,
2015, pp. 1-6, doi: 10.1109/ICETECH.2015.7275030.
[2] G. Birare and V. A. Chakkarwar, "AutomatedDetectionof
Brain Tumor Cells Using Support Vector Machine," 2018 9th
International Conferenceon Computing,Communication and
Networking Technologies (ICCCNT), Bengaluru,India,2018,
pp. 1-4, doi: 10.1109/ICCCNT.2018.8494133.
[3] A. Kumar, A. Ashok and M. A. Ansari, "Brain Tumor
Classification Using Hybrid Model Of PSO And SVM
Classifier," 2018 International Conference on Advances in
Computing, Communication Control and Networking
(ICACCCN), Greater Noida, India, 2018, pp. 1022-1026, doi:
10.1109/ICACCCN.2018.8748787.
[4] R. Ahmmed, A. S. Swakshar, M. F. Hossain and M. A.
Rafiq, "Classification of tumors and it stages in brain MRI
using support vector machine and artificial neural network,"
2017 International Conference on Electrical, Computer and
Communication Engineering (ECCE), Cox's Bazar,
Bangladesh, 2017, pp. 229-234, doi:
10.1109/ECACE.2017.7912909.

[5] C. Saha and M. F. Hossain, "MRI brain tumor images
classification using K-means clustering, NSCT and SVM,"
2017 4th IEEE Uttar Pradesh Section International
Conference on Electrical, Computer and Electronics
(UPCON), Mathura, India, 2017, pp. 329-333, doi:
10.1109/UPCON.2017.8251069.

Automated Intracranial Neoplasm Detection Using Convolutional Neural Networks

Recommended

More Related Content

Similar to Automated Intracranial Neoplasm Detection Using Convolutional Neural Networks (20)

More from IRJET Journal (20)

Recently uploaded (20)

Automated Intracranial Neoplasm Detection Using Convolutional Neural Networks