Soft Computing

https://doi.org/10.1007/s00500-020-05096-z (0123456789().,-volV)(0123456789().,-volV)

METHODOLOGIES AND APPLICATION

ICM-BTD: improved classification model for brain tumor diagnosis

using discrete wavelet transform-based feature extraction and SVM
classifier
A. Gokulalakshmi1 • S. Karthik1 • N. Karthikeyan2 • M. S. Kavitha1

 Springer-Verlag GmbH Germany, part of Springer Nature 2020

Abstract
In medical image processing, the detection, classification and segmentation of the tumor region from MRI scans accurately
are very complicated, significant and time-consuming process. When there is a scenario occurs to handle with large amount
of images for tumor diagnosis, there is need of an efficient and adaptive classification model to handle with the anomalous
structures of human brains. The MRI brain images show the typical internal brain structure and hence help scholars and
medical practitioners in accurate disease diagnosis. With that note, this paper develops a model called improved classi-
fication model for brain tumor diagnosis for appropriate classification of tumor images from input MRI images. Initially,
filtering techniques are applied for preprocessing the acquired scan images and feature extraction is done with gray-level
co-occurrence matrix and discrete wavelet transform equations, which produces more precise results. And, classification is
done with the technique called support vector machine, in which the binary classifications are effectively done. The
proposed model is evaluated under simulation, and the obtained results outperform the results of traditional brain tumor
detection process based on precision, recall and processing time.

Keywords Brain tumor diagnosis  MRI brain images  Discrete wavelet transform  Support vector machine 
Classification  Segmentation

1 Introduction

Digital image processing with medical datasets is the area

in which the clinical images are processing using com-
puting models for appropriate disease diagnosis. A typical
digital image is made with fixed amount of image elements
Communicated by V. Loia. called pixels, having specific intensity rates and positions.
Moreover, in the domain of medical image diagnosis, the
& A. Gokulalakshmi
agokulalakshmisnsct@gmail.com disease detection and proliferation about the internal model
of the human body, magnetic resonance imaging (MRI) is
S. Karthik
profskarthik@gmail.com utilized. When compared to CT images, the details on
tissue differences provided by MRI images are more
N. Karthikeyan
profkarthikeyann@gmail.com appropriate (Mustaqeem et al. 2012; Akram and Usman
2011; Shen et al. 2003). Hence, MRI images are widely
M. S. Kavitha
mskavitha1977@gmail.com used in many researches based on brain disease detection
(Salman and Bahrani 2010; Ananda and Thomas 2012;
1
Department of Computer Science and Engineering, SNS Dubey et al. 2011).
College of Technology, Coimbatore, Tamil Nadu 641035,
The typical medical image processing includes functions
India
2
such as preprocessing, segmentation, feature extraction,
Department of Computer Science and Engineering, SNS
classification and diagnosis of diseases. Figure 1 presents
College of Engineering, Coimbatore, Tamil Nadu 641107,
India the diagrammatic representation of the basic operations

123
 A. Gokulalakshmi et al.

Fig. 1 Basic steps involved in

image processing–tumor
diagnosis

involved in medical image processing. Between the image made at earlier stage, the curing process of patients can be
acquisition and reporting results, those operations are to be simpler and their life span can also be increased.
effectively done with various computations. In recent days, several tumor detection methodologies
In clinical practices, it is a complicated and time-con- are developed. Among all, the automated system for brain
suming process to detect tumor from MRI brain images tumor diagnosis is very exigent, which is still on the focus
through manual detection process by doctors and radiolo- of myriad researchers. With that concern, the major motive
gists; hence, there is need of an automated system for of the proposed model is to build an automated system for
appropriate detection. Nevertheless, there may also pro- brain tumor diagnosis appropriately from the MRI scans.
duce some variations in results as well in analyzing the After preprocessing of MRI images, segmentation is car-
images manually (Telrandhe et al. 2015). Consequently, in ried out for identifying the affected regions with image
the present decade, image processing is found as an processing techniques. Typically, segmentation can be
effective technique for cancer diagnosis with reduced time described as the process of segregating the images
and risk factors (Cha et al. 2006). Moreover, in general, according to their textures, colors, contrasts and other
brain tumor is defined as the anomalous growth of tumor features. Following that, feature extraction is performed
cells in brain. It can be classified under benign and based on DWT and GLCM. Based on that, classification is
malignant stages, which can also be termed as primary and executed using the support vector machine-based
secondary stages of tumors, respectively. In that, malignant classification.
tumor is the most aggressive and dangerous (Kalaiselvi and The results are reported to the concern medical person
Somasundaram 2011), in which the active tumor cells have for further treatment process to the patient, in real-time
non-uniform structures that can extend to all brain parts. clinical practice applications.
According to WHO health standards, the tumor is catego- The remainder of the paper is framed as follows: Sect. 2
rized into four types: ranging from GRADE I to GRADE narrates about the various available models for tumor
IV. Any kind of people at any age can be affected, and the diagnosis in brain and also from other body parts. Section 3
effect of disease on every person may vary. But, in the describes the working process of the proposed improved
intricate structure of brain, the appropriate detection pro- classification model for brain tumor diagnosis (ICM-BTD).
cess is very complicated. Section 4 provides the results and comparative analysis on
In this developed improved classification model for the accurate detection of proposed model, and comparative
brain tumor diagnosis, gray-level co-occurrence matrix analysis also presented in that section. Finally, conclusion
(GLCM) and discrete wavelet transform (DWT) are used is presented in Sect. 5 along with some paths for further
for segmentation of cancerous tissues in the obtained brain enhancement of the proposed model.
image. Moreover, classification is performed with SVM
classification technique. Based on the reported results,
suitable surgeries or therapies will be suggested by the
medical practitioners. That is, when the tumor detection is

123
ICM-BTD: improved classification model for brain tumor diagnosis using discrete wavelet…

2 Related works minimal redundancy with maximal relevance method has

been used. The input MRI images were divided into two
In Wang et al. (2007), brain tumor detection has been sections as region with normal cells and region with
performed with both magnetic resonance imaging (MRI) abnormal or cancer cells (Coatrieux et al. 2013). Another
and magnetic resonance spectroscopy (MRS). In addition work given in Zanaty (2012) described about the hybrid
to feature extraction, feature selection has also been per- model that integrated FCM, seed growing and Jaccard
formed using concentric circle model for selecting signif- similarity coefficient computation for evaluating the tumor
icant features to be used in classification. The work could image with cancer cells and to segment that appropriately.
be further enhanced with more consideration on spatial The authors of Yao et al. (2009) derived a model based
data about the tumor. In order to classify the brain tumor in on wavelet transformation and SVM for brain tumor
appropriate categories based on the shapes and textures diagnosis and classification. Further, in Kumar and
presented in MRI inputs, pattern-based classification model Vijayakumar (2015), principal component analysis (PCA)
has been discussed in Zacharaki et al. (2009). In the model, has been used for accurate cancer detection and artificial
feature extraction has been done with pixel intensities and neural network-based training and testing model has been
shapes of the MRI scans. SVM classification technique has used for classification. Fuzzy-based clustering for medical
been used for classification. image processing was given in Cui et al. (2013) and
In Rajendran and Madheswaran (2009), a pruned asso- Kavitha et al. (2020). Moreover, the authors used Jaccard
ciative model for tumor diagnosis from medical images is similarity indexing model for segmentation based on the
given. The authors have used computerized tomography variation on white, gray and cerebrospinal fluid. Active
brain images. Statistical association rule mining-based contour method was utilized in Chaddad (2015) for solving
algorithm has been used for tumor diagnosis in Li et al. the issues based on image intensities. Gaussian mixture
(2010). In the model, weight coefficient was calculated for model was applied for brain tumor diagnosis from MRI
each feature, based on that scrupulous classification has input images using PCA (Sachdeva et al. 2013; Sabitha
been done. For earlier brain tumor diagnosis, in Flusser et al. 2016). In a different manner, the work presented in
(2005), perceptron-based neural networks (PNNs) have Bouattane et al. (2019), brain tumor segmentation model
been developed. Moreover, region severance algorithm has has been proposed with respect to the temperature changes
been used for abnormality detection from brain images. on the pathological area. The works presented in Varuna
Further, the authors of El Far et al. (2011) provided a Shree and Kumar (2018) and Kutlu and Avcı (2019) used
comparative analysis between models such as, Close?, discrete wavelet transform-based feature extraction for
Apriori algorithm and association rule mining for deriving tumor diagnosis in brain and liver. By analyzing the liter-
attributes for appropriate medical image detection. ature survey, it can be observed that the accuracy in disease
In Dhanalakshmi and Rajamani (2010), the association diagnosis can be increased with low computational over-
rule mining has been used for kidney disease diagnosis. In head and complexities.
order to reduce the complications in efficient mining, dis-
cretization-based feature selection model has been applied.
In Ion and Udristoiu (2011), semantic association rule 3 Work process of the proposed improved
mining has been used to derive features from visual images classification model for brain tumor
that were in low-level attributes. Further, a combined diagnosis (ICM-BTD)
model of association and classification rule mining has
been explained in Shekhawat and Dhande (2011a, b) for The major motive of this work is to perceive the tumor
effective classification of input images based on disease from MRI brain images to help the clinical practitioners to
presence. Backpropagation neural networks have been used treat patients in better way. The proposed IMC-BTD
for training, and classifying data was presented in Jose comprised the following steps in the process of effective
et al. (2012) for diagnosing kidney images. tumor diagnosis.
Brain image segmentation using K-means clustering 1. Data preprocessing
model based on tumor diagnosis was given in Joseph et al. 2. Skull masking
(2014). Morphological filtering has been used for tumor 3. KMC-based segmentation
detection from MRI brain images. Moreover, support 4. Feature extraction
vector machine (SVM)-based tumor classification was
described in Alfonse and Salem (2016) and Kavitha et al. • Using DWT
(2019). In that work, fast Fourier transform was used for • Using GLCM
feature extraction, and for feature dimensional reduction, 5. SVM-based classification.

123
 A. Gokulalakshmi et al.

3.1 Preprocessing

This process is very significant to enhance the standard of

input images and provided appropriate results that aids in
disease diagnosis in medical image processing. It also aids
in enhancing features of input images that includes
increasing the rate of signal to noise in visual effect of the
input samples. The pixel intensity of each input MRI image
is clearly defined for enhancing the result accuracy.
Moreover, preprocessing process includes unnecessary
noise removal, smoothing inner regions and edge framing.

3.2 Skull masking

Skull detection is the next process that is performed in the

proposed model for properly detecting the exact boundaries
of elements. For appropriate tumor diagnosis, the non-brain
tissues are to be separated from brain tissues, and the
operation is termed as skull masking. Moreover, the
information about the edges is used to determine the region
of interest (ROI), which defines the image and contains the
tumor cells. For this, centroid is computed and a central
lane is marked in the skull center, which can segregate the
skull into two halves. One half is considered as the test
image, and other is taken for reference. For contouring the
cancer presented region boundary, the axial view of each
MRI image is considered, in which it can be stated that the
tumor cells can be presented in any axis symmetry that may Fig. 2 KMC-based segmentation
be left or right. The histogram intensities of both sides are
different, and the intensities outside the boundary of the 3.4 Feature extraction
cancer tissue are alike. In this work, it is to be considered
that the tumor tissues are presented at any one part among The main features of MRI scan images such as color, pixel
the two separations. intensity, texture and shapes are considered for feature
extraction. Here, two methods of feature extraction are
3.3 KMC-based segmentation accomplished, which are explained as follows.

For segmentation, K-means clustering technique is used

here, in which the similar tissues are grouped together. It
3.4.1 Discrete wavelet transformation-based feature
mainly aids in the determination of structure of abnormal
extraction
cells. Moreover, in KMC, the cells are grouped based on
the extracted features, which frames, K number of groups,
In this process, as the name represents, distinct frequency
based on the number of features extracted. Here, the
sets are used at different levels to process the input images.
clustering operation is accomplished with the computation
From the process, wavelet coefficients are derived from the
of minimal Euclidean distance between the data and the
input images. The frequency data of signals are confined by
centroid. Figure 2 shows the process of KMC-based seg-
the computed wavelet coefficients which are significant
mentation of MRI images in tumor diagnosis.
factors for categorization of tumors. By applying the DWT
As in Fig. 2, the input MRI brain image is separated into
model, four sub-bands are framed based on the ROI as
‘K’ number of groups; following, centroids are determined
follows:
for each clusters and the distance between each pixel of
brain image and centroids is evaluated. Then, segmentation i. Low–low (L–L)
process is carried out till the last existing pixel in that ii. Low–high (L–H)
image is completed. iii. High–high (H–H)
iv. High–low (H–L)

123
ICM-BTD: improved classification model for brain tumor diagnosis using discrete wavelet…

Moreover, the disintegration of an MRI image provides angle 90 (at vertical phase) and negative angle at angle
an approximation that denotes the levels of wavelet fre- 135. It is also determined that the frequency of occurrence
quencies in each image. The approximations rates for lower of the pixel intensity ‘x’ happens in accordance with
levels are given as: L–L1, L–L2, whereas the higher levels another pixel ‘y’ at certain distance function D and direc-
are denoted as: L–H1, H–L1, H–H1, H–H2, respectively. tion. Moreover, in this model, the features such as energy,
Those representations are used to represent the horizontal contrast, correlation, entropy, homogeneity and variance
and vertical directions of pixels, based on the wavelet are acquired from the low-level and high-level sub-bands.
levels. Here, the low-level image determinations are used The computation process of the considered features is
to represent the approximation over real data and the high- given as follows:
level approximation is obtained by the decomposition of
i. Energy (EY) Energy is considered here as the
previous representations and image data. This process is
quantity of reoccurring pixel pairs. It is the deriva-
iterated till an appropriate level of pixel resolution is
tion of similarity of pixels in an MRI scan, and the
obtained.
equation is given as:
In this DWT-based feature extraction, the obtained vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
images are divided into spatial frequency elements that are u i1 j1
uX X
derived from lower and the higher sub-bands of wavelets. EY ¼ t f 2 ðx; yÞ: ð2Þ
For efficiently describing the image features, both levels of x¼0 y¼0

sub-bands were utilized. The variant frequency compo-

ii. Contrast (CT) Contrast is determined by the pixel
nents based on its resolution with respect to the frequency
intensities along with the adjacent pixels of an
scale are given as:
 P image, and the derivation is given as:
EAi;j ¼ P f ðsÞhðsÞ  iðs  2ijÞ
FDWT ðsÞ ¼ ð1Þ X
i1 X
j1
EAi;j ¼ f ðsÞlðsÞ  iðs  2ijÞ: CT ¼ ðx  yÞ2 f ðx; yÞ: ð3Þ
x¼0 y¼0
From the above equation, the coefficients ‘EAi;j ’ denote
the element attribute in wavelet transformation function iii. Correlation (CN) It is derived as the computation of
‘FDWT ðsÞ’ for signal ‘s.’ The elements represent the spatial features between the image pixels
approximation rates based on the high level and low level Pi1 Pj1
of functional determinations, ‘hðsÞ’ and ‘lðsÞ,’ respectively. x¼0 y¼0 ðx; yÞf ðx; yÞ  NxNy
CN ¼ : ð4Þ
Further, the factors, ‘i’ and ‘j’ denote the wavelet measure rx ry
and translation parameter of images. iv. Homogeneity (HM) It is observed on the basis of
local regularity in an MRI image. It is computed by
3.4.2 Gray-level co-occurrence matrix-based feature the variations on textured and non-textured features,
extraction which can be stated as inverse variant moment.

Analyzing the texture of input brain images creates a X

i1 X
j1
1
HM ¼ 2
f ðx; yÞ: ð5Þ
greater impact on classifying the normal and abnormal x¼0 y¼0 1 þ ðx  yÞ
brain images easily. This also provides way for effective
machine learning in disease diagnosis. Moreover, it v. Entropy (ET) Entropy is computed by considering
enhances the precision rate of appropriate feature extrac- the designated noisiness of the input image based on
tion for earlier recognition of brain tumor. Initially, in the textures. It is calculated as:
first part of computation, the first-order-based analytical X
i1 X
j1
texture evaluation based on the feature data from the image ET ¼ f ðx; yÞlog2 f ðx; yÞ: ð6Þ
intensities has been derived and grayscale frequencies at an x¼0 y¼0
auxiliary image positions are evaluated. The correlation
Based on the computation of the above features, peak
coefficient or co-occurrence is not considered in the pro-
signal-to-noise ratio (PSNR) and mean square error
cess. In the second part, the second-order-based textural
(MSE) are the factors which are evaluated.
computation is carried out in accordance with the gray-
vi. Peak signal-to-noise ratio (PSNR) It is determined
level occurrence at random distance and pixel intensities.
by the features of reframed image from the obtained
In this work, gray-level co-occurrence matrix (GLCM)
image. The formula is given as:
is incorporated, in which (x, y)th component is the fre-
quency of occurrence of function ‘x’ happening with ‘y.’ It
is a distance function D = 1; at horizontal phase, angle will
be at 0, positive diagonal at angle 45, negative angle at

123
 A. Gokulalakshmi et al.

2m  1 (http://www.dicom.com) that are built by radiologists on

PSNR ¼ 20log10 : ð7Þ
Mean Square Error practice with several modalities of images. Here, for result
analysis, 750 samples are taken into account from 30
vii. Mean square error It is computed by image images of the obtained dataset content. There are some
comparison based on the similarity scores. challenges in processing the dataset images like low con-
1 X X 2 trast, intensity and so on. It is also to be stated that the
MSE ¼ f ði; jÞ  f G ði; jÞ : ð8Þ
XY dataset images are effectively partitioned for training and
testing functions. For providing evidence for the proposed
By these methods of feature extraction, the derived model, the obtained results are compared with the existing
features are given for the SVM classification model for models such as PNN- and FCM-based cancer detection.
training and testing, for efficient tumor detection in MRI
scan images of brain. The overall framework of the pro- 4.2 Result evaluations
posed model is presented in Fig. 3.
In the proposed work, the feature extraction is performed
3.5 SVM-based classification of brain images based on DWT and GLCM. Based on the sub-band rates
obtained from L–L and H–L derivations of wavelet trans-
In this proposed ICM-BTD, the supervised learning tech- formation, the images are divided into different levels.
nique called support vector machine is used for classifi- Following, based on GLCM, the analytical features such
cation. SVM-based classification provides accurate as, energy, contrast, correlation, homogeneity and entropy
classification results by analyzing and processing the large are derived. These features are given for SVM classifier for
dataset of MRI images. Moreover, the classification is tumor image classification from obtained sample of MRI
performed by the formation of decision planes, by which images. Based on the DWT-based image decompositions,
the dissimilar class elements are being separated by for analyzing, brain image (BR) = {BR1, … BR5} is con-
hyperplane. Specifically, linear support vector machine- sidered here with different sub-bands of DWT. As men-
based classification technique is used here for detection of tioned earlier, the extracted features are fed for
tumor presence from the input brain images. Moreover, classification. Moreover, Tables 1 and 2 contain the ana-
Gaussian radial basis function (RBF) is used here to per- lytical features computed from GLCM with different levels
form the binary classification. In this, it is considered that of sub-bands for both the training and testing phases.
the training elements can be linearly divisible. The function Further, the values calculated for MSE, PSNR, area of
is given as: tumor tissue and area of BR in pixel are presented in
FðpÞ ¼ ATp þ qð1Þ ð9Þ Table 3. The brain images presented in Fig. 4 depict the
BR considered for the assessment of the proposed model
where ‘p’ is the training sample; for each sample, the that are acquired from DICOM dataset. In the proposed
function obtains F(p) C 0, in a case, if p = ? 1, and model, after preprocessing, the important operation per-
f(pi) \ 0, when qi = - 1. From Eq. (9), ‘q’ denotes the formed is skull masking. Figure 5 contains the image
invariant factor and ‘A’ is the unit vector. Based on the output, after performing skull masking, for appropriate
provided training dataset, many hyperplanes can be framed disease diagnosis from brain images. Following, the feature
that enhance the dividing margin between the classifica- extraction is performed with DWT and GLCM. According
tions of normal and abnormal images. Further, the derived to that, the classification process of normal and abnormal
support vectors are presented at the boundary line of the brain images is performed. The performance is evaluated
hyperplane between classifications. by the values obtained by the calculation of PSNR and
MSE.
Figure 6 presents the enhanced brain image after the
4 Results and comparative analysis process of feature extraction. From the obtained results
given in Table 3, it indicates that the minimal MSE and
4.1 Dataset description maximal PSNR value denotes better value of signal-to-
noise ratio in the processed input. For evaluating the per-
The evaluation of the proposed model is executed in formance of a classification model and comparative anal-
MATLAB tool, using the benchmark dataset called ysis between models, accuracy-related factors such as
DICOM dataset, which contains MRI brain image samples specificity, sensitivity, precision and accuracy and

123
ICM-BTD: improved classification model for brain tumor diagnosis using discrete wavelet…

Fig. 3 Framework of the proposed model

123
 A. Gokulalakshmi et al.

Table 1 Values obtained for analytical features from GLCM obvious from the graph that the proposed model produces
derivation with variant sub-bands under training phase better results than compared models. And, Fig. 9 compares
Brain images EY CT CN HM ET the processing time taken for providing classification
results. It is obvious from the comparative analysis; the
BR1 0.977 0.0112 0.0206 0.903 0.332
proposed model provides accurate classification results in
BR2 0.992 0.0036 0.0381 0.965 0.339 minimal time, which helps in earlier cancer detection and
BR3 0.966 0.0168 0.0259 0.901 0.337 treats the patients in better ways.
BR4 0.989 0.0054 0.0027 0.766 0.272
BR5 0.974 0.0125 0.0477 0.683 0.337
5 Conclusion

This paper presents a new model called improved classi-

Table 2 Values obtained for analytical features from GLCM
derivation with variant sub-bands under testing phase
fication model for brain tumor diagnosis (ICM-BTD),
which comprises steps such as preprocessing, skull mask-
Brain images EY CT CN HM ET ing, segmentation, feature extraction and classification.
BR1 0.899 0.0073 0.0198 0.870 0.389 Preprocessing is for removing noise from obtained MRI
BR2 0.954 0.0110 0.0295 0.910 0.321 scan images. Skull masking is performed for enhancing the
BR3 0.832 0.0120 0.0243 0.891 0.302 obtained image by deleting the skull tissues, which are not
BR4 0.820 0.0043 0.0034 0.745 0.253 considered for tumor detection in brain. Moreover, KMC-
BR5 0.893 0.0108 0.0450 0.864 0.330 based segmentation is carried out for clustering similar
elements that makes the detection process more efficient.
The significant section of the adduced work is feature
extraction using DWT and GLCM. Based on this, salient
processing time are to be analyzed based on the TRUE features are derived from the smoothened MRI image and
POSITIVE, FALSE POSITIVE, TRUE NEGATIVE and given for SVM-based classification for finding the class of
FALSE NEGATIVE rates in classification results. The the processed MRI, which can be normal or abnormal. The
graph portrayed in Fig. 7 represents the comparison of results are evaluated based on the factors such as accuracy
accuracy rate among models for classifying brain images. It rate, precision and processing time, and it is to be stated
is explicit from the figure that the proposed ICM-BTD that the proposed model provides better results than the
model achieves better results than compared works. The compared works and evidenced the efficacy of the model.
accuracy rate of brain tumor detection between the com- The application of the proposed model can be effective in
pared and the proposed model is evaluated. By the effec- brain tumor diagnosis in clinical practices in earlier stages.
tive incorporation of KMC-based segmentation and feature In future, the work can be enhanced by incorporating
extraction techniques, the proposed model achieved better with some other efficient segmentation and classification
rate of accuracy than others. In average, the model achieves model based on the real-time applications in clinical
94.2% of accuracy in cancer image classification. practice. Another path for enhancement can be considered
Figure 8 presents the factor-based evaluation for eval- the volume analysis of the detected tumor from MRI scans
uating the performance of the proposed model. Based on of brain with some other datasets such as BioGPS and
the classification results with TRUE POSITIVE, FALSE BraTS.
POSITIVE, TRUE NEGATIVE and FALSE NEGATIVE
rates, the sensitivity, specificity, precision and accuracy
rates are computed and the results are presented. It is

Table 3 Performance analysis based on MSE and PSNR

Brain images Peak signal-to-noise ratio Mean square error Area of BR in pixel Area of tumor tissue

BR1 12.82 3.216 66,824 9774

BR2 13.12 8.058 50,608 7323
BR3 13.59 5.54 24,944 4664
BR4 13.72 7.69 50,419 3678
BR5 14.23 6.152 16,284 4397

123
ICM-BTD: improved classification model for brain tumor diagnosis using discrete wavelet…

Fig. 4 Brain images obtained from dataset for evaluation

Fig. 5 Image obtained after the

removal of skull tissues from a
BR

Fig. 6 Enhanced BR after

feature extraction

123
 A. Gokulalakshmi et al.

Fig. 7 Accuracy comparison

among models

Fig. 8 Factor-based evaluation

between compared models

Fig. 9 Processing time

comparison

Funding This research is not supported under any funding. References

Akram MU, Usman A (2011) Computer aided system for brain tumor
Compliance with ethical standards detection and segmentation. In: IEEE
Alfonse M, Salem M (2016) An automatic classification of brain
Conflict of interest The authors declare that they have no conflict of tumors through MRI using support vector machine. Egypt
interest. Comput Sci J 40:11–21
Ananda RS, Thomas T (2012) Automatic segmentation framework
Research involving human participants and/or animal This article for primary tumors from brain MRIs using morphological
does not contain any studies with human participants or animals filtering techniques. In: 5th International conference on biomed-
performed by any of the authors. ical engineering and informatics. IEEE
Bouattane O, Youssfi M, Raihani A (2019) Towards reinforced brain
Informed consent All referred study is highlighted in the Literature tumor segmentation on MRI images based on temperature
Review. changes on pathologic area. Int J Biomed Imaging 2019:1758948
Cha S et al (2006) Review article: Update on brain tumor imaging:
from anatomy to physiology. J Neuroradiol 27:475–487
Chaddad A (2015) Automated feature extraction in brain tumor by
magnetic resonance imaging using Gaussian mixture models. Int
J Biomed Imaging 2015:868031

123
ICM-BTD: improved classification model for brain tumor diagnosis using discrete wavelet…

Coatrieux G, Huang H, Shu H, Luo L, Roux C (2013) A Li W, Lu Z, Feng Q, Chen W (2010) Meticulous classification using
watermarking based medical image integrity control system support vector machine for brain images retrieval. In: 2010
and an image moment signature for tampering characterization. International conference of medical image analysis and clinical
IEEE J Biomed Health Inform 17(6):1057–1067 application (MIACA)
Cui W, Wang Y, Fan Y, Feng Y, Lei T (2013) Localized FCM Mustaqeem A, Javed A, Fatima T (2012) An efficient brain tumor
clustering with spatial information for medical image segmen- detection algorithm using watershed and thresholding based
tation and bias field estimation. Int J Biomed Imaging segmentation. Int J Image Graph Signal Process 4(10):34–39
2013:930301 Rajendran P, Madheswaran M (2009) Pruned associative classifica-
Dhanalakshmi K, Rajamani V (2010) An efficient association rule- tion technique for the medical image diagnosis system. In: 2009
based method for diagnosing ultrasound kidney images. In: 2010 Second international conference on machine vision
IEEE International conference on computational intelligence and Sabitha R, Karthik S, Shanthini J (2016) Breast cancer detection using
computing research (ICCIC) enhanced descriptive approach. J Med Imaging Health Inform
Dubey RB, Hanmandlu M, Vasikarla S (2011) Evaluation of three 6:1887–1892
methods for MRI brain tumor segmentation. In: ITNG. IEEE Sachdeva J, Kumar V, Gupta I, Khandelwal N, Ahuja CK (2013)
Computer Society Segmentation, feature extraction, and multi class brain tumor
El Far M, Moumoun L, Chahhou M, Gadi T, Benslimane R (2011) classification. J Digit Imaging 26(6):1141–1150
Comparing between data mining algorithms: ‘‘Close?, Apriori Salman SD, Bahrani AA (2010) Segmentation of tumor tissue in gray
and CHARM’’ and ‘‘K means classification algorithm’’ and medical images using watershed transformation method. Int J
applying them on 3D object indexing. In: 2011 International Adv Comput Technol 2(4):123–127
conference on multimedia computing and systems (ICMCS), Shekhawat P, Dhande SS (2011a) Building an iris plant data classifier
pp 1–6 using neural network associative classification. Int J Adv
Flusser J (2006) Moment invariants in image analysis. Proc World Technol 2(4):491–506
Acad Sci Eng Technol 2(11):196–201 Shekhawat PB, Dhande SS (2011b) A classification technique using
Ion AL, Udristoiu S (2011) An experimental framework for learning associative classification. Int J Comput Appl 20(5):20–28
the medical image diagnosis. In: Proceedings of information Shen S, Sandham WA, Granat MH (2003) Preprocessing and
technology interfaces segmentation of brain magnetic resonance images. In: IEEE
Jose JS, Sivakami R, Uma Maheswari N, Venkatesh R (2012) An Conference on information technology applications, proceedings
efficient diagnosis of kidney images using association rules. Int J of the 4th annual biomedicine, UK, pp 149–152
Comput Technol Electron Eng (IJCTEE) 2(2):14–20 Telrandhe SR, Pimpalkar A, Kendhe A (2015) Brain tumor detection
Joseph RP, Singh CS, Manikandan M (2014) Brain tumor MRI image using object labeling algorithm and SVM. Int Eng J Res Dev
segmentation and detection in image processing. Int J Res Eng 2:2–8 (Special issue)
Technol 3:1–5 Varuna Shree N, Kumar TNR (2018) Identification and classification
Kalaiselvi T, Somasundaram K (2011) Fuzzy c-means technique with of brain tumor MRI images with feature extraction using DWT
histogram based centroid initialization for brain tissue segmen- and probabilistic neural network. Brain Inform 5:23–30
tation in MRI of head scans. In: Proceedings in IEEE-interna- Wang Q, Liacouras EK, Miranda E, Kanamalla US, Mega-
tional symposium on humanities, science and engineering looikonomou V (2007) Classification of brain tumors using
research, pp 149–154 MRI and MRS. In: Proceedings of SPIE - the international
Kavitha MS, Shanthini J, Sabitha R (2019) ECM-CSD: an efficient society for optical engineering. https://doi.org/10.1117/12.
classification model for cancer stage diagnosis in CT lung 713544
images using FCM and SVM techniques. J Med Syst 43:73. http://www.dicom.com
https://doi.org/10.1007/s10916-019-1190-z Yao J, Chen J, Chow C (2009) Breast tumor analysis in dynamic
Kavitha MS, Shanthini J, Bhavadharini RM (2020) ECIDS-enhanced contrast enhanced MRI using texture features and wavelet
cancer image diagnosis and segmentation using artificial neural transform. IEEE J Sel Top Signal Process 3(1):94–100
networks and active contour modelling. J Med Imaging Health Zacharaki EI, Wang S, Chawla S, Soo Yoo D, Wolf R, Melhem ER,
Inform 10(2):428–434(7). https://doi.org/10.1166/jmihi.2020. Davatzikos C (2009) Classification of brain tumor type and grade
2976 using MRI texture and shape in a machine learning scheme.
Kumar P, Vijayakumar B (2015) Brain tumor MR image segmenta- Magn Reson Med Magn Reson Med 62(6):1609–1618
tion and classification using by PCA and RBF kernel based Zanaty EA (2012) Determination of gray matter (GM) and white
support vector machine. Middle East J Sci Res 23(9):2106–2116 matter (WM) volume in brain magnetic resonance images
Kutlu H, Avcı E (2019) A novel method for classifying liver and brain (MRI). Int J Comput Appl 45:16–22
tumors using convolutional neural networks, discrete wavelet
transform and long short-term memory networks. Sensors Publisher’s Note Springer Nature remains neutral with regard to
19(9):1992 jurisdictional claims in published maps and institutional affiliations.

123