See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/382185582

Utilising Deep Learning Convolutional Models for the Classification of Brain

Tumors in Human Through Image Analysis

Conference Paper · May 2024

DOI: 10.1109/ICCICA60014.2024.10585230

CITATIONS READS

0 47

4 authors, including:

Nishith Thakkar Shanti Verma

Lok Jagruti Kendra Institutes Lokmanya Collge
1 PUBLICATION 0 CITATIONS 63 PUBLICATIONS 113 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Shanti Verma on 26 July 2024.

The user has requested enhancement of the downloaded file.

 2024 International Conference on Computational Intelligence and Computing Applications (ICCICA)

Utilising Deep Learning Convolutional Models for the

2024 International Conference on Computational Intelligence and Computing Applications (ICCICA) | 979-8-3503-0644-6/24/$31.00 ©2024 IEEE | DOI: 10.1109/ICCICA60014.2024.10585230

Classification of Brain Tumors in Human Through

Image Analysis
Nishith Thakkar Devanshu Pathak
Department of Computer Applications Department of Computer Applications
L J University, Ahmedabad, India L J University, Ahmedabad, India
[email protected] [email protected]

Harin Thakkar Shanti Verma

Department of Computer Applications Department of Computer Applications
L J University, Ahmedabad, India L J University, Ahmedabad, India
[email protected] [email protected]

Abstract - Brain tumor identification at the initial stage is an pattern to recognize it and give output on the bases of feed data.
uphill battle. In a present cut-throat technology era, various This neural network work on different layers and the main part
medical fields fluke the world by advancing and upgrading is hidden layer where all the processing and image enhancing is
technology. Furthermore, by advancing in technics many done [12]. With the CNN model image classification done as
imaginary things became possible like prediction of brain tumor this model use weighted method and if error occurs equilibrium
is possible with MRI reports by using different AI ML algorithms propagation is done by which after the processing in hidden
like CNN, R-CNN and many more. Brain tumor occurs because layers classification is done. As this model can process and
of growth of abnormal tissues in anatomical structure. In this classify images, MRI is used for classification. To deal with
paper authors used MRI image dataset available at github link
complex geometry, a convolutional neural network is more
and apply various deep learning classification algorithms like
CNN, PNN and RNN. The results of the experiment conclude that
beneficial and for the cure of medical images which use CNN
CNN algorithms have highest accuracy 95% with respect to PNN as the main source in the research of computer-assisted
and RNN. All algorithms are tested with 25 hidden layers. This diagnosis. [4]. Because of brain tumors in 2023,10 million
study makes easier for doctors and helps in diagnosis. deaths entirely in the world and approx. 20 million new cancer
cases are reported. By this process of classification doctors can
Keywords – MRI, image processing, convention models, brain identify brain tumors by using just MRI. There are many kind
tumor, CNN model, cut-throat, anatomical structure of brain tumors some are cancerous and some are noncancerous
, As noncancerous do not show rapid growth and cancerous
I. INTRODUCTION shows rapid growth [5].
Brain tumor is a core issue in terms of classification, after
the medical evolution we can classify brain tumor by
convention models. As brain tumor is mainly of three types
glioma, Meningioma, pituitary. MRI have a major impact in
detection of tumor , with this MRI image recognition is held by
doing max pooling and image extraction [1].1/3 of tumor found
is glioma and 2/3 of tumor found is malignant in the brain [2].
Glioma is a cancerous tumor which generally grows and makes
the collection of cells. Meningioma grows inside the skull but
outside of the brain which makes it sometimes life threatening
also. Pituitary tumor grows in the pituitary gland, but this is
non-cancerous. The aim is to prove the normal – abnormal
relevant classification of tumor. With the help of MRI
(Magnetic resonance imaging) the detection is possible by
fitting algorithm of neural network as detects tumor by running
pulses in nuclei [3].
CNN generally used in image recognition , trend analysis Fig. 1. Illustration of suggested technique
etc. [16]. ]. It reads or analysis the trend pattern or image

Copyright © IEEE–2024 ISBN: 979-8-3503-0644-6 553

Authorized licensed use limited to: Charotar University of Science and Technology. Downloaded on July 26,2024 at 09:15:39 UTC from IEEE Xplore. Restrictions apply.
 2024 International Conference on Computational Intelligence and Computing Applications (ICCICA)

In this image shown an MRI report is taken and goes II. LITERATURE REVIEW
through the image processing. Image processing is used for
making an image into digital form to get useful information A. Image processing
from that image. After image processing, data augmentation The process of preprocessing contains some methods they
will come into picture, augmentation means to make something are required to give importance for the end result, data
enlarge or clear. After image processing the imageundergoes extraction, reproducing image and geometric modification of
the process of data augmentation in which the image will be Initial image [10]. Modification is done because the images are
clearer and enlarged and make a variation of image. After the free of inconsistency, the areas which do not show brain are
data augmentation convention model will be used for discarded, unwanted area noise. After these we have to convert
classification of the image, many different models can be used these images into grayscale. For large amounts of noise, we use
as if we use CNN model the weighted pixel method is used in multiple separation techniques. First, preprocessing involves
which after taking the input it goesto the hidden layer where the transforming the raw MRI file into a format that may be used
process is done.With the accuracy of 95% the detection of brain for additional processing. This level of pre-processing reduces
tumor can be achieved by CNN model [3]. image noise, highlights important areas, and displays
observable parts of digital images.
To deal with complex geometry, a convolutional neural
network is more beneficial and for the cure of medical images B. Brain Tumor
which use CNN is the main source in the research of computer- There are many different types of Gliomas such as
assisted diagnosis. [4]. Our brain is divided in 3 parts – Astrocytoma- This type of tumor mostly emerges in the brain
Midbrain, hindbrain and Forebrain and all have different types part called cerebrum and as well as in cerebellum, this part
of sub parts. Forebrain is important part of brain and it function helps for making balance in walking, running etc. Some parts of
the emotional and behaviour of human, cerebrumis the biggest the brain like Brain stem are not surgical for removing tumors.
part of the human brain, and higher brain functions like This type of gliomas in the brain stem is also known as Diffuse
cognition and action are linked to it. Midbrain is an important infiltrating brainstem gliomas.
part of our brain as it performs the transmitting task for hearing
and vision messages or information, without midbrain the most Ependymomas - This tumor is mostly found in children of
useful senses of hearing and vision will not work properly. It age group of 10 years about 8% to 10% and this tumor is rare to
contains numerous numbers of dopamine producing neurons in find. This tumor is generally formed in spinal cords and brain,
it. Midbrain is located above the hindbrain and lower side of this tumor starts in ependymal cells because of extension or
Forebrain nearer to the pituitary gland. It consists of cerebral growth of cells.
aqueduct, tegmentum, cerebral peduncles, cerebral aqueduct. Mixed gliomas - Mixed gliomas is unlike Ependymomas
Usually glioma, meningioma and pituitary cancer are seen in tumor this tumor is common in adults compared to children and
large numbers of midbrain can also found in the large, outer it is amalgamation of different types of glial cells. Cerebrum is
part of the brain but also in the cerebellum. The most common a hotspot for mixed gliomas as most of the mixed glioma tumor
type of tumor is glioma which emergesin the brain, of all brain arises in this part.
tumors around 1/3rd brain tumor are gliomas which arise by the
supporting of glial cells. The growth inside the part of the brain Oligodendrogliomas – This kind of brain tumor are found in
which often fuses with normal brain tissue are called intra-axial adults and young ones and in the perspective of gender most
brain tumors. Maximum gliomas located in the brainstem commonly found in men. And as another tumor
which is also known as midbrain can also found in broader and Oligodendrogliomas also mostly emerges in the cerebrum.
outer part of brain and also in cerebellum.
Optic pathway gliomas - Optic pathway gliomas are mostly
found in neurofibromatosis as this is a genetic disorder, it may
cause vision loss. This cause in the pathway of optic nerve, the
message do not supply properly as eye see the object and give
message to brain and inreturn brain responds but the object
message do not reach properly to brain.
A meningioma is a tumor that develops from the
membranes that surround the brain and spinal cord known as
meninges. Despite not being a brain tumor strictly speaking, it
falls under this category since it has the potential to compress or
pressure nearby brain tissue, nerves, and blood vessels. The
most frequent kind of tumor to develop in the head is
meningioma. The majority of meningiomas grow extremely
slowly, frequently spanning years without showing any
symptoms.
Unusual growths in the pituitary gland are called pituitary
Fig. 2. Parts of Brain [13] tumors. The size of this gland is comparable to that of a pea. It
is situated near the base of the brain, behind the nose. A subset

Copyright © IEEE–2024 ISBN: 979-8-3503-0644-6 554

Authorized licensed use limited to: Charotar University of Science and Technology. Downloaded on July 26,2024 at 09:15:39 UTC from IEEE Xplore. Restrictions apply.
 2024 International Conference on Computational Intelligence and Computing Applications (ICCICA)

of these tumors induces the pituitary gland to produce excessive transformations to existing images [14]. We are aware that a
amounts of certain hormones that regulate vital bodily large amount of data is needed to fine-tune a deep learning
processes. The pituitary gland may produce too little of certain model’s parameter. Since our dataset is still very small, we used
hormones due to other factors. Benign pituitary tumors the data preprocessing technique to our training sample and
predominate. Thus, they are not cancerous. Pituitary adenomas made minor picture alterations, such as fips, rotations,
are another term for these noncancerous tumors. Pituitary gland intensities. This will increase the size of our dataset and enable
is responsible for metabolism and puberty stage in our body the model to learn and perform effectively by treating each of
because of release of hormones from pituitary gland. It plays an these minute changes as a new picture. [3].This allows us to
important role for making pathways between the endocrine and generate robust data, which is valuable for classification.
nervous system. Size of this gland is equal to pea and located in
the centre of the brain in the middle of the skull exactly at the
backside of the nose bridge.
C. Deep Learning
While detecting a tumor it is labelled on the image. To
resolve these obstacles various image processing techniques
and machine learning algorithms are applied to training and
testing datasets. Among these some common techniques are
Image restoration ,Color image processing etc [6]. CNN based
models are more reliable and accurate than traditional methods
for classification and have robust ability in feature presentation
[7]. They developed CNN based image and video recognition
system [8] it is done to enhance the function of model and 2
models were built for better performance of classification.
III. PROPOSED METHODOLOGY
In all over process, we used 2000 MRI imagesfor image
processing algorithms [3]. We are using three different models
CNN, PNN and R-CNN to analyse their accuracy and loss on
varied datasets and to find leading models. Inclusive we find
CNN as leading with 95% accuracy where PNN and R-CNN Fig. 3. Work Flow chart
with 94% and 42% roughly. In CNN we trained neurons using a
CNN - In our CNN model, we diligently designed a multi-
total 25-layer in detail architecture. Convolutional Layers
layered architecture to extract subtle features critical for
Utilised to detect patterns and features within input images
distinguishing between different tumor kinds. The design
implemented with 10 Conv2D layers, each applying 3x3 filters
consists of ten convolutional layers with 3x3 filter sizes. These
with Rectified Linear Unit (ReLU)activation functions. Max-
layers extract hierarchical characteristics from input photos,
Pooling Layers applied to down sample feature maps, reducing
collecting patterns such as edges, textures, and forms using a
spatial dimensions and mitigating overfitting comprises 5
collection of learnable filters. Deeper layers of the network
MaxPooling2D layers. Dropout Layers employed for
capture higher-level characteristics required for distinguishing
regularisation to prevent overfitting by randomly deactivating
between tumor kinds. To handle the spatial complexity of the
neurons during training Consists of 6 Dropout layers, set at a
feature maps and avoid overfitting, we used max-pooling layers
30% dropout rate. Flatten Layer converts multi-dimensional
with a kernel size of (2,2). These layers downscale the
feature maps into one-dimensional vectors for input into fully
representations, reducing spatial dimensions while preserving
connected layers containing 1 Flatten layer. Dense Layers fully
important properties. This reduces computing complexity and
connected layers responsible for classification based on
increases the network's capacity to learn translation-invariant
extracted features comprises 3 Dense layers, including the
characteristics, increasing resilience to changes in image
output layer with SoftMax activation for multi-class
orientation and position. Furthermore, we included six dropout
classification. The collection contains 400 images of Glioma,
layers with a 0.3 dropout rate to reduce overfitting and increase
300 images of Meningioma and 350 images of Pituitary and
generalization. During training, these layers randomly
leftover shots of noncancerous tumors. Our data were
deactivate a subset of neurons, preventing the network from
categorised into three groups: Data processing, import images
becoming unduly reliant on specific properties or memorizing
and Resize images. Test datasets are used to asses models while
the training data. This improves performance on previously
machine learning techniques are used to train models. Lastly,
unseen data and the model's capacity to generalize to new
the test findings would be applied to evaluate the models. [3].
samples. Following the convolutional and dropout layers, we
Our CNN model includes max pooling, input layer, output
used a flattened layer to turn the multidimensional feature maps
layer, convolution layer [11].
into one-dimensional vectors, preparing the data for input into
Image augmentation helps CNN to reduce overfitting and the fully connected dense layers. Finally, we added three dense
artificially increase its training set size by making different layers each with 512 neurons and ReLU activation algorithms.
These layers complete the last phases of Feature aggregation

Copyright © IEEE–2024 ISBN: 979-8-3503-0644-6 555

Authorized licensed use limited to: Charotar University of Science and Technology. Downloaded on July 26,2024 at 09:15:39 UTC from IEEE Xplore. Restrictions apply.
 2024 International Conference on Computational Intelligence and Computing Applications (ICCICA)

and classification, which translate extracted features to output RNN - The study leveraged MRI images from the 'archive'
classes and allow the network to make predictions based on dataset, categorized into glioma tumor, meningioma tumor, no
learned representations. By using numerous thick layers, the tumor, and pituitary tumor. Each image was standardized to
network may learn complex decision limits and capture 150x150 pixels and split into training and testing sets. To enrich
nuanced correlations between features and class labels, thus the dataset, advanced data augmentation techniques like
improving classification accuracy. rotation, flipping, and zooming were applied. The model
architecture featured an input layer for processing the flattened
PNN - In the PNN model, each image has been enlarged image data, followed by a GRU layer with 64 units to capture
to a uniform dimension of 150x150 pixels, a standard the sequential nature of pixel data. To mitigate overfitting, a
preprocessing step that ensures consistent input to the neural 0.5-rate dropout layer was integrated. This was succeeded by
network and reduces computational complexity while dense layers with 64 units and ReLU activation functions,
preserving essential classification features. A one-hot encoding culminating in an output layer with a SoftMax activation
scheme was implemented to handle class labels, converting function to classify images into the four categories. The model
categorical labels into a binary matrix representation, which is underwent training over 20 epochs with a 0.1 validation split.
critical for training the neural network with categorical cross- Training accuracy, validation accuracy, and loss were vigilantly
entropy loss function. The neural network used in this project is
monitored to assess the model's performance. The model
a sequential model built using the keras library consisting of achieved an overall accuracy of 39%, with the 'glioma_tumor'
numerous dense (completely connected) layers with ReLU class exhibiting the highest recall and the 'meningioma_tumor'
activation functions, which are well-known in deep learning class displaying the lowest performance.
models for their ability to alleviate the vanishing gradient
problem. The model begins with a flattening layer that converts IV. RESULT AND DISCUSSION
the 2D picture input into a 1D array, then various dense layers
with varying numbers of neurons (256, 128, 64, 32, and 16), To F1 score are calculated by CNN (Convolutional Neural
and finally a dense layer of four neurons with a SoftMax Network) model [17-18]. Analysis is performed by CNN can be
activation function that corresponds to the four classes. This depicted by equation also
setup ensures that the model can detect complicated patterns
and relationships in the dataset. The model was trained across
20 epochs, with a batch size specified by the data size and
computational limitations, using the Adam optimizer, which is
known for its efficiency.

Same process in applied to other model PNN and RNN.As

we can see the comparison between 3 models below –

TABLE I. COMPARISON OF CLASSIFICATION MODELS PERFORMANCE

Fig. 4. Proposed CNN model Fig. 5. Comparison of classification models

Copyright © IEEE–2024 ISBN: 979-8-3503-0644-6 556

Authorized licensed use limited to: Charotar University of Science and Technology. Downloaded on July 26,2024 at 09:15:39 UTC from IEEE Xplore. Restrictions apply.
 2024 International Conference on Computational Intelligence and Computing Applications (ICCICA)

As we can see PNN model is giving 94% accuracy and Iranian Journal of Science and Technology, Transactions of Electrical
Engineering 45.3 (2021): 1015-1036.
RNN is giving 41% accuracy that is not enough and CNN
giving 95% accuracy.PNN is better than RNN model as IN [12] Reszke, Maria, and Łukasz Smaga. "Machine learning methods in the
detection of brain tumors." Biometrical Letters 60.2 (2023): 125-148.
PNN total 12 layers are used including 1 Flatten layer and 1
[13] Chandrakar, Mukesh Kumar, Anup Mishra, and Arun Kumar. "Multipath
output layer in increasing process layer accuracy decreases on Convolutional Neural Network for Brain Tumor Detection (CNN)."
the result 5 layers is used in this model. With the low accuracy Handbook of Research on Computer Vision and Image Processing in the
of brain detection system that is not acceptable in medical field, Deep Learning Era. IGI Global, 2023. 293-308.
precision and sensitivity is the main factors to be seen in this [14] Kishor, Rohan, and Praveen Bhojane. "Enhanced Brain Tumor Detection
systems S[9]. CNN model is more accurate to detect brain and Stage Prediction through a Unified CNN-Based Framework."
tumor. This model is efficient which can help in the medical [15] Swarup, Chetan, et al. "Brain tumor detection using CNN,
field to detect brain tumors more precisely. AlexNet&GoogLeNetensembling learning approaches." Electronic
Research Archive 31.5 (2023): 2900-2924.
V. CONCLUSION [16] Y. R. Prajapati, D. G. Hihoriya and S. Verma, "Early Detection and
Prediction of Diabetes Using Ensemble Classifier,"
We have used different categorization strategies for the [17] 2023 14th International Conference on Computing Communication and
classification of brain tumor. Our first step was to detect the Networking Technologies (ICCCNT), Delhi, India, 2023,
region of interest in MRI scans and then we trimmed them. pp. 10.1109/ICCCNT56998.2023.10306942.
After that we prepared and enlarged the size of raw data for [18] Sadariya, T., Verma, S. (2023). Early Prediction and Detection of
further processing and analysis. For the classification of images Anxiety Level Using Support Vector Machine. In: Swaroop, A.,
to detect brain tumors we developed a simple CNN model [3]. Polkowski, Z., Correia, S.D., Virdee, B. (eds) Proceedings of Data
Analytics and Management. ICDAM 2023. Lecture Notes in Networks
Brain tumors can now be detected early thanks to this and Systems, vol 787. Springer, Singapore. https://doi.org/10.1007/978-
technique. It might be possible to identify brain tumors more 981-99-6550-2_22
rapidly and precisely. It facilitates medical science's better [19] B. S. Rao, A. D, R. R, S. Verma, P. V. Nandankar and R. P, "Machine
identification of brain structures. [10].In the ground reality of Learning and Deep Transfer Learning approaches were used to create a
medical diagnosis application, it must join different real time Face Mask Identification model for COVID-19," 2022 International
diagnostic systems to collaborate with doctors [4].In the future, Conference on Innovative Computing, Intelligent Communication and
Smart Electrical Systems (ICSES), Chennai, India, 2022, pp. 1-6, doi:
proposed strategies can be used for apparent characterization 10.1109/ICSES55317.2022.9914141.
problems, such as recognizable proof of tumor types like
Glioma, meningioma, and pituitary [15].
REFERENCES
[1] Nadeem, Muhammad Waqas, et al. "Brain tumor analysis empowered
with deep learning: A review, taxonomy, and future challenges." Brain
sciences 10.2 (2020): 118.
[2] Goodenberger, McKinsey L., and Robert B. Jenkins. "Genetics of adult
glioma." Cancer genetics 205.12 (2012): 613-621.
[3] Gupta, Manali, Sanjay Kumar Sharma, and G. C. Sampada.
"Classification of Brain Tumor Images Using
CNN." Computational Intelligence and Neuroscience 2023 (2023).
[4] Zhang, Yamei, and Liting Jing. "Convolution neural network application
for glioma image processing." 2018 2nd IEEE Advanced Information
Management, Communicates, Electronic and Automation Control
Conference (IMCEC). IEEE, 2018.
[5] Hapsari, Peni Agustin Tjahyaningtijas, et al. "Brain tumor classification
in MRI images using en-CNN." International Journal of Intelligent
Engineering and Systems 14.4 (2021): 437-451.
[6] Salçin, Kerem. "Detection and classification of brain tumors from MRI
images using faster R-CNN." Tehničkiglasnik 13.4 (2019): 337-342.
[7] Zhou, Xiaokang, Yue Li, and Wei Liang. "CNN-RNN based intelligent
recommendation for online medical pre-diagnosis support." IEEE/ACM
Transactions on Computational Biology and Bioinformatics 18.3 (2020):
912-921.
[8] Xie, Guo-Sen, et al. "Hybrid CNN and dictionary-based models for scene
recognition and domain adaptation." IEEE Transactions on Circuits and
Systems for Video Technology 27.6 (2015): 1263-1274.
[9] Zahoor, Mirza Mumtaz, and Saddam Hussain Khan. "Brain tumor MRI
Classification using a Novel Deep Residual and Regional CNN." arXiv
preprint arXiv:2211.16571 (2022).
[10] Varshney, Shivam, et al. "Image Processing based Brain Tumor
Detection." 2022 International Conference on Fourth Industrial
Revolution Based Technology and Practices (ICFIRTP). IEEE, 2022.
[11] Irmak, Emrah. "Multi-classification of brain tumor MRI images using
deep convolutional neural network with fully optimized framework."

Copyright © IEEE–2024 ISBN: 979-8-3503-0644-6 557

Authorized licensed use limited to: Charotar University of Science and Technology. Downloaded on July 26,2024 at 09:15:39 UTC from IEEE Xplore. Restrictions apply.

View publication stats