Volume 8, Issue 6, June 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Automatic Brain Cancer Detection using SVM

Kernel Trick from MRI Imaging
Azad Kumar1, Dr. Neelesh Jain2, Prateek Singhal3
Research Scholar1, Professor2, Asst. Professor3
Dept. of Computer Sceience & Engineering
SAM College of Engineering & Technology, Bhopal, Madhya Pradesh, India

Abstract:- A brain tumour is an unchecked cell and 99.35 percent of accuracy, specificity, and
development that has the potential to spread to other sensitivity, respectively, were attained by the system.
tissues. It can be identified using Magnetic Resonance
Imaging (MRI), a non-surgical method of organ research Keywords:- Support Vector Machine, Brain Tumor,
for the diagnosis of any illness associated with the Segmentation,, Malignant, Benign, MRI, Brain Cells.
symptoms. Tumours can be malignant or not, and they
can also be hazardous or not to human life. A tumour I. INTRODUCTION
can be classified as either benign or malignant, two For the purpose of examining, locating, and treating
separate categories. Benign tumours are seen as less brain tissues, magnetic resonance imaging is a common and
hazardous or non-cancerous since they do not spread to safe method. The patient's life can be saved by receiving the
other parts of the brain. It has distinct contours or limits right therapy if a brain tumour is discovered early. The
that show the specific shade of the tumour, however accurate categorization and division of the tumour region
malignant tumours are cancerous growths that may should be made feasible by this suggested framework since
migrate to other parts of the brain on their own and are the precise identification of tumours in MRI images
thus extremely hazardous. The malignant tumor's becomes a crucial task to complete. The detection of a
borders don't appear to be firm; rather, they have a tumour from an MR image is also used in division and 3D
faded look. Here, the suggested approach is capable of reproduction. To avoid wasting time, professionals only
classifying both the tumour type and the diagnosis of a conduct a limited amount of physical and visual research.
brain tumour. Here, the suggested system makes use of The SVM Classifier may be used to restrict an abnormal cell
polynomial Support Vector Machine (SVM) to handle mass in a slice of magnetic resonance imaging (MR) and
the impairments and identify the illness as necessary. divide the tumour cells to determine the size of the tumour
The clustered cluster may be classified using SVM based that is present in that sectioned area. To demonstrate the
on their patterns. The feature mapping of patterns might kind of the tumour, the sectioned part's divided components
come from both diseased and normal cells. Drawing a will be produced using a fake brain organisation. These
hyper plane for non-linear data is challenging for other features will also be used in the Classification student
classification algorithms, however SVM can classify the application to compare the accuracy of several classifiers
data by converting it to linear data, which then makes it [1]. Perhaps the most precise and sensitive organ in the
simple to design a hyper plane. It has been controlled human body is the brain. Unimaginably fake brain tumours
using a kernel approach that allows for the mapping of are to blame for the high fatality rate. The National Brain
high dimensional feature space. Utilising the polynomial Tumour Society has shown that brain tumours are a highly
feature of SVM, the optimisation problem may also be harmful contaminant for people. It is a collection or mass of
resolved. High degree of accuracy was perceived by the damaged brain cells. The cerebral cortex is located quite
system in comparison to the prior model. 97.24, 93.94, near to the brain. Brain tumours can either be non-
destructive (harmless) or cancerous (compromising).

Fig. 1: Benign Tumor MRI Image

IJISRT23JUN1813 www.ijisrt.com 1882

Volume 8, Issue 6, June 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
As dangerous or risky tumours grow, the pressure [2].Destructive tumours that progress rapidly and cause
inside of your brain increases. Undermining tumours can be mortality are referred to as risky tumours. Destructive
divided into two categories: primary and auxiliary tumours tumours grow quickly, are envious, seek for a new location,
and benign tumours. The patient's health worsens as the and spread (metastasize) as opposed to cleanse tumours. A
undermining tumour quickly spreads to numerous brain dangerous tumour is formed by damaged cells that are
structures. A mass of slowly proliferating brain cells is a growing quickly. The magnetic field fragment utilised to
brain tumour that is benign (without malignant finish radio recurrence to provide clear images of interior
development). It typically doesn't stand up and spread. The changes in the human body, including bones, sensitive
symptoms of a brain tumour vary depending on its size and tissues, and organs. The MRI approach is very sensible for
location inside the brain. A few tumours that keep on identifying brain tumours. MRI images should make it
growing progress silently at every turn. Not joking, possible to identify brain tumours. Image updating clusters
persistent brain discomfort, seizures (sufficient), perpetual utilised for clinical image diagnostics are employed in
squeamishness, vomiting, and sluggishness are typical side image processing to further improve picture quality. In order
effects. A situation where the phones mentioned in the to identify and collect brain tumours, EDGE detection,
danger of sickness are unexpectedly generated is known as histogram, and division limits are expected to play a
an earlier condition or basic condition, also known as significant role. The goal of this research is to identify
probable condition or fundamental condition. These various channels, detachment techniques, and assessments
conditions can eventually lead to illness if left untreated for the perception of brain tumours [2].

Fig. 2: Malignant Tumor MRI Image

II. RELATED WORKS of brain tumours is advised using a mixed technique. The
division and categorization of brain tumours suggested by T.
A. Literature Survey A. Jemimma et al. [4] are carried out using the Water Shed
Using CWT, DWT, and SVMs, Mircea Gurbin et al. Algorithm (WSA), Dynamic Angle Projection Pattern
[3] established a framework for classifying and detecting highlights, and these elements are arranged using CNN. The
brain tumours. The suggested technique uses several levels importance of the watershed division calculation effectively
of wavelets, with CWT being used to provide the high eliminates the tumour regions to enable competent DAPP
precision component. The CWT prevents divisions from highlight extraction. The histogram highlights are acquired
lacking edges. The result demonstrates that SVMs with the after the DAPP eliminates the surface components from the
proper information preparation arrangements can recognise fragmented tumour districts. These component vectors are
unique and typical tumour sites and properly classify them raised to play a role in the classification process as it is
as a benign tumour, malignant tumour, or normal brain. carried out by the CNN classifier. For a successful diagnosis
SVMs offer important computational advantages. The of a brain tumour, the division and categorization of the
doctor needs this categorization to accurately describe the MRI brain picture are essential. The BRATS data set, which
symptoms and recommend the best course of therapy. The achieves greater dice score proficiency of 93.5% and
obtained results demonstrate that, in comparison to DWT, awareness of 94.2%, is used to execute the trial outcomes. In
CWT provides greater calculation. Even if the computation further study, a few more varied factors may compare to
time is greater, using CWT is preferable if our main interests obtain greater accuracy in the categorization and division of
are perception, coordination, and include recognition. DWT brain tumours. It may also be extended to differentiate
is frequently more appropriate if we are interested in de- between various tumour types, including fibromas,
noising, pressure, or rebuilding. The proper resolution of the adenomas, and pancreatic tumours.
challenges surrounding the localization and categorization

IJISRT23JUN1813 www.ijisrt.com 1883

Volume 8, Issue 6, June 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig. 3: Overview of the System [3]

The suggested technique by R.Lavanyadevi et al. [5] brain tumour and identification of the tumour type was
calls for precisely detecting the semantically significant proposed by Ragib Shahariar Ayon et al. [7]. After
entire areas in a picture. The doctor or radiologist might preprocessing the picture using denoising and inclination
therefore identify hazard by connecting each and every pixel correction, which was then handled to the handling stage as
in the image together with which suggest a semantically information image, brain tumour finding is completed. We
important relevance. Brain pictures containing safe, risky, or used the spatial FCM to clip the probable tumour out of the
usual images are removed from adjoining twofold models brain MR picture. The tumour cut was then handled until the
and dark level co-events. The PNN classifier is used in post-handling step, where it is channelled via a local region.
preparation mode to prepare the semantic components and The final graphic shows the projected tumour region
the deleted highlights. Similar highlights from the test brain separately. Similar features were used to create a variety of
picture and mystery are subtracted in classification mode classifiers, and we chose the one that predicted the tumour
using preset models and a PNN classifier. When the test type most accurately. After comparing and analysing
picture differs from any preparation image, the image can be various bunching and classification computations, we can
recalled for creating set data. Regarding the relationship state that the suggested method is superior to conventional
between PNN and CNN, PNN is seen to have several strategies for tumour division and classification.
advantages. PNN benefits from gathering knowledge briefly
because of reality. PNN can modify its learning gradually III. PROBLEM IDENTIFICATION
since it has a fast learning capability. The method suggested
by Hein Tun Zaw et al. [6] can assist clinical staff members, An implementation model based on FCM (Fuzzy C-
such as specialists and radiologists, in analysing brain Means Clustering) and CNN (Convolutional Neural
malignant development from MRI images, notably for Network) techniques was developed by L. Jagjeevan Rao et
GBM, which necessitates the finding of all potentially al. [8]. The algorithm in this case employs CNN as a
harmful spreading locations. The most severe entropy edge classifier and FCM to extract the brain's characteristics and
has been used in this method to identify brain tumours using limitations. However, CNN is typically employed for feature
Naive Bayes classification. This review makes use of the extraction, whereas C-means clustering is utilised for
REMBRANDT data collection. The developed computation classification. CNN is a convolutional neural network that
can accurately identify the tumour in all possible regions of has the capacity to obtain input features and enhance the
the brain where the tumour may reside, including the features by using distinct filters, train the layers accordingly,
projection of the outside world. The computation results in a and generate a model that can be able to diagnose or
94% overall precision, an 81.25% identification rate on recognise the target object. CNN has not been designed for
tumour pictures, and a 100% discovery rate on non-tumor the classification process. SVM is regarded as the top
images. A method for the straightforward division of the classifier for diagnosing diseases.

IJISRT23JUN1813 www.ijisrt.com 1884

 Volume 8, Issue 6, June 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig. 4: Graphical Representation of Result [8]

The FCM's drawback is that it takes a long time to system's implementation has been broken down into three
converge the data, is more sensitive to noise, and has trouble categories, including normal, malignant, and benign. When
with non-linear data. The traditional CNN model performs a tumour is benign, its borders appear to be solid or fixed in
poorly when it comes to training and creating dense size and are thus less harmful than when a tumour is
networks, which directly impacts the execution time. The malignant, which is dangerous because it can spread to other
accuracy of the system was 91%, which is somewhat less parts of the brain. The malignant tumor's borders blur or
than what may be improved by utilising other techniques. fade, and it may be distinguished from the benign picture.
The normal picture of the brain is thought to be a healthy
IV. PROPOSED WORK & IMPLEMENTATION brain image since it does not include any tumours. The
suggested method in this case makes use of polynomial
The brain tumour can be accurately diagnosed with the SVM, a kernel-based classifier that can handle non-linear
planned study. The system may also categorise tumours data and produce results with more accuracy.
according to whether they are benign or malignant. The

Fig. 5: Original Brain Tumor MRI Image

The original MRI picture of the brain tumour is shown Figure 6 depicts the original image that has been
in Fig. 5. The white portion of the image is thought to be the smoothed down so that clusters may be produced more
tumour portion of the image, and according to the theory readily for segmentation. In the segmentation process,
and real idea, it belongs to the malignant class. certain clusters are separated into various groups based on
the threshold value or colour of the picture, which is then
A. Smoothening classified using a classifier to get the desired results.

IJISRT23JUN1813 www.ijisrt.com 1885

Volume 8, Issue 6, June 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig. 6: Region Smoothening the Brain Tumor MRI Image

B. Segmentation component or pixel that belongs to the same class. Take, for
Picture segmentation is a technique that divides a digital instance, a situation where a picture is required to be
picture into several subgroups known as Image segments. provided as part of a protest finding. The identification can
This approach helps to reduce the complexity of the image be inputted with a location selected by a segmentation
to make further handling or inspection of the image easier. computation rather than processing the full picture. This will
Simply said, segmentation is the process of giving pixels prevent the identify from processing the complete image,
names. A common name is assigned to each picture hence cutting down on guessing time.

Fig. 7: Brain Tumor Segmentation

The segmentation of a brain tumour picture is shown categorise the data points. The system then determines the
in Fig. 7, where the red portion depicts a cluster of damaged retrieved lesion's entropy. It determines the lesion's density
cells and the remaining portions are all thought of as the for subsequent comparison with the cutoff value. Threshold
image's background that must be removed later for better value is the relative value at which a choice can be made.
categorization. The suggested system's flowchart, shown in There are three categories of brain tumours; the first is
Fig. 8, initially loads the dataset picture as input data. The malignant; as a result, if entropy exceeds and is equivalent
pre-processing module was then started in order to improve to the higher threshold value, it will be regarded as
the photos' visibility. One of them is the histogram, which is malignant cancer. The second kind is benign; a healthy brain
in charge of balancing the system's brightness and contrast. imaging can be taken into consideration if entropy is larger
Once visibility increases, the image is smoothed to improve than the lower threshold value but less than the higher
lesion visibility before features are extracted and feature threshold value. As a result, the method also assigns a grade
selection, after which SVM classification can be started to to each brain tumour according to its density.

IJISRT23JUN1813 www.ijisrt.com 1886

Volume 8, Issue 6, June 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Start

Input Dataset
Image

Compute
Lesion Entropy
Histogram
Equalization

Is
Smoothening Entropy≥Threshold Yes Malignant
(t1) ?

No

Polynomial SVM
Classification
Is
Entropy < t1 && Yes Benign
Entropy > t2 ?

Lesion
No
Classified

Healthy Brain
Detected

End

Fig. 8: Flow Chart of Proposed System

Fig. 9: Polynomial SVM Classification [9]

IJISRT23JUN1813 www.ijisrt.com 1887

Volume 8, Issue 6, June 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Polynomial-SVM Algorithm (Kernel Trick)
Input: 2-D Image Matrix
Output: Entropy
Step 1: Input 2-D Image Matrix
Step 2: Convert Matrix to Gray Levels
Step 3: Apply Histogram Equalization
𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑃𝑖𝑥𝑒𝑙 𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦 𝑛
𝑃𝑛 = 𝑛 = 0,1 … . 𝐿 − 1
𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑖𝑥𝑒𝑙𝑠

Where 𝑃𝑛 is the affected pixel value after histogram equalization.

Step 4: Apply Gaussian Filter for Smoothening
Step 5: Collect data points
𝑦 = 𝑤0 + 𝑤1 𝑥1 + 𝑤2 𝑥2 + 𝑤3 𝑥3 + 𝑤4 𝑥4 … … …
𝑚

= 𝑤0 + ∑ 𝑤𝑖 𝑥𝑖
𝑖=1
𝑤𝑖 = 𝑤0 , 𝑤1 , 𝑤2 … … … … . 𝑤𝑚
Where wi is the vector, b is the bias and x is the variable
Step 6: Separate the data points by hyperplane
𝑤
⃗⃗ .𝑥 - b = 1
𝑤
⃗⃗ .𝑥 - b = 0
𝑤⃗⃗ .𝑥 - b = -1
where 𝑤⃗⃗ is the normal vector of the hyperplane
Step 7: Classify Datapoints
Step 8: Compute Entropy
Step 9: if E ≥ T1 then
Malignant;
elseif E < T1 && E > T2 then
Benign;
else
Normal;
end else
end if
Step 10: End

V. RESULT ANALYSIS
TP
Precision = ∗ 100 %
A. Result Simulation TP + FP
Four metrics—True Positive (TP), False Positive (FP),
True Negative (TN), and False Negative (FN)—are used to TN
Negative Prediction Rate = ∗ 100 %
measure the success of an experiment. True Positive refers FN + TN
to cases in which a system correctly identified an image as
either malignant or benign, whereas True Negative refers to FP
situations in which a system incorrectly identified an image False Positive Rate = ∗ 100 %
FP + TN
as either benign or malignant. False Positive means when a
system diagnoses a normal-class picture as either malignant FN
False Negative Rate = ∗ 100 %
or benign, while False Negative means when a system FN + TP
diagnoses a malignant or benign image as normal. In the
Kaggle benchmark, there are a total of 76 testing photos TP + TN
Accuracy = ∗ 100 %
from the benign class, 89 from the malignant class, and 98 TP + FP + TN + FN
from the normal class. 2TP
F1 = ∗ 100 %
TP 2TP + FP + FN
Sensitivity = ∗ 100 %
TP + FN TP
Recall = ∗ 100 %
TN FN + TP
Specificity = ∗ 100 %
FP + TN

IJISRT23JUN1813 www.ijisrt.com 1888

Volume 8, Issue 6, June 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig. 10: Proposed API

SEN: Sensitivity, SPE: Specificity, P: Precision, NPV: Negative Predictive Value, FPR: False Positive Rate, FNR: False
Negative Rate, ACC: Accuracy, F1: F1 Score and R: Recall.

Table 1: Result Outcomes (Confusion Matrix)

Terms & Parameters Proposed
Total Testing Class 254
TP 154
TN 93
FP 6
FN 1
SEN in % 99.35
SPE in % 93.94
P in % 96.25
NPV in % 98.94
FPR in % 6.06
FNR in % 0.65
ACC in % 97.24
F1 Score in % 97.78
R in % 99.35

100 SEN
SPE
80 P

60 NPV
FPR
40 FNR

20 ACC
F1
0 R
Result in %
Graph 1: Result Analysis

IJISRT23JUN1813 www.ijisrt.com 1889

Volume 8, Issue 6, June 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Table 2: Result Comparison
Methods Specificity in % Sensitivity in % Accuracy in %
CNN [8] 92.00 91.00 91.00
BPNN [8] 89.50 91.00 89.00
KNN [8] 86.00 88.00 87.00
P-SVM (Proposed) 93.94* 99.35* 97.24*

100

98

96

94
CNN [8]
92 BPNN [8]
KNN [8]
90 P-SVM

88

86

84
SPE SEN ACC
Graph 2: Result Comparison

VI. CONCLUSION & FUTURE SCOPE REFERENCES

The brain tumour may be identified and classified [1.] T. S. Kumar, K. Rashmi, S. Ramadoss, L. K.
according to whether it is malignant, benign, or normal Sandhya and T. J. Sangeetha, "Brain tumor detection
utilising brain tumour classification using MRI imaging using SVM classifier," 2017 Third International
using polynomial SVM. The system employs histogram Conference on Sensing, Signal Processing and
equalisation to change the image's contrast for improved Security (ICSSS), 2017, pp. 318-323, doi:
image visibility. In order to make impairments more visible 10.1109/SSPS.2017.8071613.
and create a distinct cluster that can aid in segmentation and [2.] Healthline, Brain Tumor,
classification, the system additionally applies a smoothening https://www.healthline.com/health/brain-tumor,
filter. For final classification, the system employs Accessed- 04 Jun 2022.
Polynomial SVM, which creates two separate clusters: one [3.] Gurbina, Mircea; Lascu, Mihaela; Lascu, Dan
is composed of aberrant or lesion cells, while the other is (2019). [IEEE 2019 42nd International Conference on
composed of normal cells, which may be thought of as the Telecommunications and Signal Processing (TSP) -
image's backdrop or noise. Comparing the system's accuracy Budapest, Hungary (2019.7.1-2019.7.3)] 2019 42nd
to that of CNN, BPNN, and KNN classifiers from earlier International Conference on Telecommunications and
systems, the system attained a higher level. Because Signal Processing (TSP) - Tumor Detection and
accuracy is crucial in the field of medical science, hybrid Classification of MRI Brain Image using Different
classifiers or models may be used in the future to improve Wavelet Transforms and Support Vector Machines. ,
accuracy, precision, specificity, and sensitivity. Because (), 505–508. doi:10.1109/TSP.2019.8769040
brain tumours are a potentially fatal condition, it is essential [4.] Jemimma, T. A.; Vetharaj, Y. Jacob (2018). [IEEE
to achieve the highest degree of accuracy with the fewest 2018 International Conference on Smart Systems and
possible false positive rates. Inventive Technology (ICSSIT) - Tirunelveli, India
(2018.12.13-2018.12.14)] 2018 International
Conference on Smart Systems and Inventive
Technology (ICSSIT) - Watershed Algorithm based
DAPP features for Brain Tumor Segmentation and
Classification. , (), 155–
158. doi:10.1109/ICSSIT.2018.8748436

IJISRT23JUN1813 www.ijisrt.com 1890

Volume 8, Issue 6, June 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[5.] Lavanyadevi, R.; Machakowsalya, M.; Nivethitha, J.;
Kumar, A. Niranjil (2017). [IEEE 2017 IEEE
International Conference on Electrical,
Instrumentation and Communication Engineering
(ICEICE) - Karur, Tamilnadu, India (2017.4.27-
2017.4.28)] 2017 IEEE International Conference on
Electrical, Instrumentation and Communication
Engineering (ICEICE) - Brain tumor classification
and segmentation in MRI images using PNN. , (), 1–
6. doi:10.1109/ICEICE.2017.8191888
[6.] Zaw, Hein Tun; Maneerat, Noppadol; Win, Khin
Yadanar (2019). [IEEE 2019 5th International
conference on Engineering, Applied Sciences and
Technology (ICEAST) - Luang Prabang, Laos
(2019.7.2-2019.7.5)] 2019 5th International
Conference on Engineering, Applied Sciences and
Technology (ICEAST) - Brain tumor detection based
on Naïve Bayes Classification. , (), 1–
4. doi:10.1109/ICEAST.2019.8802562
[7.] R. Ezhilarasi and P. Varalakshmi, "Tumor Detection
in the Brain using Faster R-CNN," 2018 2nd
International Conference on I-SMAC (IoT in Social,
Mobile, Analytics and Cloud) (I-SMAC)I-SMAC
(IoT in Social, Mobile, Analytics and Cloud) (I-
SMAC), 2018 2nd International Conference on,
Palladam, India, 2018, pp. 388-392.
[8.] L. J. Rao, R. Challa, D. Sudarsa, C. Naresh and C. Z.
Basha, "Enhanced Automatic Classification of Brain
Tumours with FCM and Convolution Neural
Network," 2020 Third International Conference on
Smart Systems and Inventive Technology (ICSSIT),
2020, pp. 1233-1237, doi:
10.1109/ICSSIT48917.2020.9214199.
[9.] Manisha, B. Radhakrishnan and L. P. Suresh, "Tumor
region extraction using edge detection method in
brain MRI images," 2017 International Conference on
Circuit ,Power and Computing Technologies
(ICCPCT), Kollam, 2017, pp. 1-5.
[10.] Mubaris, Support Vector Machines for Classification,
https://mubaris.com/posts/svm/, Accessed- 23 Aug
2021.
[11.] G. Singh and M. Ansari, “Efficient detection of brain
tumor from MRIs using K-means segmentation and
normalized histogram”, 2016 1st India International
Conference on Information Processing (IICIP), 2016.
[12.] T. Goh, S. Basah, H. Yazid, M. Aziz Safar and F.
Ahmad Saad, “Performance analysis of image
thresholding: Otsu technique”, Measurement, vol.
114, pp. 298- 307, 2018.

IJISRT23JUN1813 www.ijisrt.com 1891