Brain Tumor Detection Using Deep Learning

ABSTRACT :

The scientific world understands brain tumors as the growth of abnormal cells in the brain, so
me of which lead to cancer. The most common method of detecting brain tumors is magnetic
resonance imaging (MRI). With MRI, it is possible to detect information about the uncontroll
ed growth of brain tissue. In many research articles, brain tumor detection is done using mach
ine learning and deep learning algorithms. When these machines are used for MRI images, br
ain tumor prediction will help deliver treatment to patients much faster and more accurately.
These estimates can also help electricians make quick decisions. In the planning study, artific
ial neural networks (ANN) technique was used to detect the presence of brain tumors and thei
r performance was analyzed with different indicators.
The brain is one of the most important parts of the human body; It is responsible for our thou
ghts, autonomous movements, language, judgment and understanding. The role is played by
movement, balance and posture. Without it, we would be like "walking puppets". The word c
erebellum derives from the Latin word meaning "little brain". Brain tumors are characterized
by tumors that grow in the brain and are classified as benign (non-cancerous) or malignant (c
ancerous).

Note: While this mental report itself or the investigation may be confirmed by a physiotherapi
st and referred to as change of exchange, it should not be used to guide further treatment. br>
medRxiv preprint doi: https://doi.org/10.1101/2022.01.19.22269457 This article was publishe
d on January 25, 2022. It is no longer restricted by law. Anyone may legally distribute, reuse,
reproduce or modify this material for any purpose without attribution to the original author. S
econdary brain tumors occur frequently. Approximately half of lung or breast cancers worldw
ide affect approximately 250,000 people each year; This corresponds to less than 2% of all ca
ncers. Brain tumors are the second most common cause of cancer in young people under 15 y
ears of age. That's why radiologists commonly use magnetic resonance imaging (MRI) to exa
mine brain tumors. In this paper, neural networks are used to classify normal cells and tumor
cells. Artificial neural networks consist of many machines and their operations are quite simp
le. These units are usually connected by a communication link with some weight. A unit oper
ates only with local information, which is the input it receives from the connection. [2] Intelli
gent behavior of neural networks through the interaction of network functions. There will be i
nput and output processes and there may also be hidden processes. During the learning proces
s, weights and biases are added to the neurons of each layer based on the input and previous p
rocesses (for hidden and output processes). [3] Show the model as a dynamic use of ideas and
latent processes, further studies are carried out to achieve the desired results. Classification of
Health dataset

This data is from the Kaggle website (Public dataset - hundreds of organizations use CC0 to s
upport public domain work) (https://www.kaggle.com/ ) ). This file contains MRI images of b
rain tumors provided by researcher Jakeshbohaju Jakesh Bohaju. There are two folders, one r
epresenting normal brain images and the other representing tumor images. There are a total of
3,762 images in these two folders. Figure 1 shows an example of the brain and brain tumors.
A total of 1683 tumor images and 2079 non-tumor images were captured. legally share,
reuse, remix, or adapt this material for any purpose without crediting the original authors.
 Figure 1 shows the sample normal and brain tumor image.

Deep learning is one of the artificial intelligence technologies that is changing the way busine
ss and technology work. Thanks to this innovation, we have achieved many applications in o
ur daily lives that point to the future of our interaction with machines and services. Rules and
procedures that allow machines to operate and make decisions based on data, rather than spec
ific processes to perform specific tasks. They can recognize patterns and make decisions with
the help of patterns. This allows machines to make predictions by processing data. Import our
library

2. Learn more about our collection

3. Prepare our documents

4. Draw some examples of the information set

5. Preprocess our dataset before using MobileNet

6. Divide the dataset into training set (80%) and testing set (20%)

It is no longer prohibited by law. Anyone may legally distribute, reuse, reproduce or modify t
his material for any purpose without attribution to the original author. Convolutional neural n
etwork developed by Google researchers. They were originally designed to make it easier for
you to use the resources of your mobile phone, some data is processed in the cloud, thus savi
ng the resources of your mobile phone. Convolutional neural network architecture designed t
o work efficiently on mobile devices. It is based on the inverse residual method, where the res
idual connection is between layers. The intermediate expansion layer uses deep convolutions
to filter out features that are sources of nonlinearity. In Figure 2 we describe the number of le
arning parameters (protocols).

In more detail, MobileNets are a convolution class of neural networks designed by

Google researchers. They were initially designed to facilitate the use of your mobile device's
resources, where part of the information is processed in the cloud, thus saving your cell
phone's computing resources.

In this work, MobileNetV2 [6] was used, which is a convolutional neural network
architecture that seeks to perform well on mobile devices. It is based on an inverted residual
structure where the residual connections are between the bottleneck layers. The intermediate
expansion layer uses lightweight depthwise convolutions to filter features as a source of non-
linearity. In this Figure 2, we detail the number of parameters that
The image class labels are also generated and stored in a variable target_classes

= ['No Tumor','Tumor'] which is also an ndarray. The images are then added inside the
dataframe. The image dataset is divided into training, validation, and testing datasets. Figure
3 represents the accuracy and loss obtained when the ANN model is applied to the training
and test dataset when the ANN model is applied to the training data for fifty epochs: loss:

Figure 3. Comparing training/validation accuracy and loss of ANN model.

The model is evaluated by applying the test image dataset. The confusion
matrix for the predicted output is given as in the following Figure 4. The output
of making predictions for the testing and validation given below.

array([[364, 36],
[ 45, 308]])
Figure 4. Confusion matrix for RNN model.

In the table above, we have the comparison of the predicted value by the model
with the real value of an observation. As the names suggest:

True Positives (TP) are observations whose actual value is positive and the predicted value is
positive, i.e., the model got it right. In this work 364 images were evaluated.

True Negatives (TN) are observations whose actual value is negative and the predicted value
is negative, that is, the model got it right. In this work 308 images were evaluated.

False Positives (FP) are cases in which the correct result is negative, however, the result
obtained is positive, that is, the model was wrong. In this work 36 images were evaluated.

False Negatives (FN) are cases in which the correct result is positive but the result obtained is
negative, that is, the model was wrong. In this work 45 images were evaluated.

Figure 5. Brain Tumor Detection Using Deep Learning

Conclusion

Deep learning is considered one of the best techniques in data analysis and intelligence, even
without translation. This network makes predictions by reducing the size of the image withou
t losing the information needed to make the prediction. The accuracy of the model developed
here is 89%, and this accuracy can be increased by providing additional image data. This can
be done by using image enhancement techniques and analyzing the effectiveness of other arti
ficial intelligence techniques. We will soon see many companies and business models being p
repared to help radiologists and doctors diagnose brain tumors quickly and flexibly with the h
elp of artificial intelligence