classify the brain tumor into 4 categories of images. The first being no tumor, the second being pituitary tumor,

the third is meningioma and the last one classified as glioma. As we are well aware, the training datasets for the

medical imaging scenario are very few. This is a challenging task to apply the deep learning that is obtained

from a trained CNN model to dig up the small data set to attain the result. A pre trained CNN model is used

here to solve the problem. The obtained results are good over all Performance is measured.

Key-words: - Artificial Intelligence, Machine Learning, Brain Tumor, CNN, Kaggle database, MRI.

Received: May 15, 2021. Revised: February 23, 2022. Accepted: March 27, 2022. Published: April 29, 2022.

conducted by national Brain tumor society, 7,00,000

people live with brain tumors, the chances of

increasing the cases might be beyond 7.8 lakhs by

the end of 2021[1]. Breast cancers and lung cancer

are the most widely found cancers on the planet.

Even though the brain tumors are little uncommon,

but still, it ranks as a 10th leading cause for death.

Brain is the organ that controls the activity of a

human, If the patient is suffering from brain tumor,

definitely it will impact on the patient’s life

scanning, EEG, CT scanning and others[8]. The

figure:1, shows the healthy brain and the brain with

a tumor disorder when taken with MRI images.

The major improvement of the MRI based imaging

when compared with the CT scan technique is the

capturing of the all-possible information in the test

sample under consideration with minimized effect of

the few weights are considered individually for the

analysis such as T1, T2 and Proton density weights

for MRI analysis.

Fig. 2: (a) Coronal image; (b) Sagittal image; (c)

Axial image [5]

Based on the classification that has been described

by World Health Organization (WHO), there are

about 120 types of brain tumors. These brain tumors

Figure 3 shows the different MRI images of tumors

are in a different place. In all the three images the

tumors are marked with the red outline.

Fig. 3: A normalized MRI image that shows

different forms of tumor. In each image, the tumor is

segmentation it is deployed with the deep learning

and AI (Artificial Intelligence). Till date an huge

amount of works are carried in the domain of

biomedical engineering.

Badza et al. [2] discuss about new architecture of

brain tumor classification based on CNN. He

proposes a simpler system rather than the existing

complicated one. The system is capable of

classifying three types of tumors. It was mainly

suitable hardware such as mobile phone, digital

camera and many other devices for the collection of

the data source from unified surface. The image

acquisition is the initial step for every image

processing system, by the acquisition of the image

Fig. 5: Proposed convolutional neural networks

A brief explanation of each layer is mentioned

below.

The output image of the size 64 X 64 pixels with the

performance of the convolution operation with 3 X 3

filters is applied on MRI image of dimension 64 X

64 X 3 , output image is represented in (4).

network.

This is an linear function, where the resultant

obtained in accordance with the applied input if and

only if it is non-negative. ReLU is a novel and

mandatory triggering feature for various types of

neural networks as a mannequin utilizes an simpler

manner for the instruction to attain improved

performance. Its operations are relatively linear for

the values non zero, which implies the training of

neural network in backward propagation. The other

d) Max pooling layer

The pooling layer is very essential in terms of

minimizing the dimension of operation. The

minimization of the numbers of neurons in the

4.1 Classification of Brain MRI Images

In this research article for the testing of the designed

architecture of classifier for the brain tumour over

the wide range of public available database it is

chosen to operate with the KAGGLE dataset, in

which 394 test images which is the association of

100 MRI images with glioma tumor,115 Images

tumor and healthy person MRI images of 105. In the

approach it is considered with the 104 MRI images

with disorder oriented dataset and 65 normal dataset

of MRI images. For the classifier oriented

be calculated using four basic building blocks.

Those are TP, FP, TN and FN. Those building

blocks are explained in Table 2 in detail. Predicted

values are taken on the X-axis and actual values are

taken on the Y-axis.

classifier utilized for training of the 394 data images,

In the present research context, it is considered with

KAGGLE dataset for the analysis of the classifier

performance. Initially, MRI image denoising was

done using Anisotropic diffusion filter. In the

training phase for CNN classifier it achieved with

aimed to identify the type of brain tumor from the

database under consideration with the classifier

approach is carried out with 100 image sets with

glioma tumor,115 images with meningioma

tumor,74 images with pituitary tumor and 105

normal images making in total of 394 image sets. A

hybrid CNN model has been presented, with feed

forward neural network as a seed, with the

incremental values of epochs, it is assured to have

higher accuracy due to higher rate of training states.

An individual Epoch is defined as the entire process

increased to 71%. Finally, by using the pre-trained

model 98% accuracy was achieved. The major gain

in the classifier proposed to the MRI images of the

brain was more accurate when a pre-trained model

was used. Other factor of improvement as a future

scope is to minimize the processing time to process

the tumor images as a design constraint.

References:

[10] Ali Mohammad Alqudah, Hiam Alquraan,

[12] Shaik Basheera, and M. Satya Sai Ram,

[14] S. Deepak, P.M. Ameer., “Brain tumor

[15] Ravikumar Gurusamy and Dr.Vijayan

[19] S Harish, GF Ali Ahammed, “An optimized