This present invention lead to develop a Computer Aided Diagnosis system(CADs) for classification of brain tumor from Magnetic Resonance Images (MRI) using Soft Computing.

Prior Work In Brain Tumor Classification:
Traditionally, brain tumor detection and classification are done by manual investigation but proper diagnosis varied from person to person and is also a very time-consuming process. Well experienced doctors can do precise diagnoses but they are far away from the reach of all the patients and these doctors are also overburdened. So, to reduce their workload and to provide accurate diagnosis in less time, there is a requirement to automate this tumor segmentation and classification part. Many new methods have been proposed by researchers in the last few years to fulfill this task. But most of the methods only segment and classify the tumor region if it exists, they did not classify into different types of a brain tumor which are most frequently occurs so there was research scope available to classify these brain tumor automatically with accuracy in less time.
It has been observed from the literature that soft computing plays significant role in the field of medical image processing. Soft computing has been extensively used in feature reduction and classification of many CAD systems. Out of many techniques of soft computing, bio-inspired optimization algorithms are good enough for feature reduction. Therefore, in this invention Grey Wolf optimization (GWO) has been used for feature selection and Multiclass Support Vector Machine (MSVM) has been used for tumor classification.

Objective of the Invention:
The main objective of the invention is to classify most commonly occurring brain tumor types using Soft Computing. In this invention, Grey Wolf optimization (GWO) has been used for feature selection and Multiclass Support Vector Machine (MSVM) has been used for tumor classification by taking pre-processed MRI images. Following are the objectives which are attained in this invention:
Classify the brain tumor from MRI images using soft computing technique.
Improve the efficiency of brain tumor classifier using bio-inspired optimization algorithms.
Performance evaluation of optimization algorithms with different classifiers for detecting brain tumor.
Summary of the Invention:
Accordingly following invention will provide easy way to classify different type of brain tumor with accuracy of about 99%. This invention will help in reducing the workload of neurologist and also provide good alternate for the patients of remote area.

Detailed Description of the Invention:
In the present invention, a framework is developed for the classification of brain tumor using a soft computing-based classifier. Three nature-inspired optimization algorithms named Grey Wolf Optimization (GWO), Particle Swarm Optimization (PSO) and Firefly Algorithm (FA) have been used for features selection task. This framework contains mainly 3 steps as (i) Input, (ii) Processing and (iii) Output as described below in Fig. 1.

Fig 1. Optimized-SVM Framework for Brain Tumor Classification
Step-1:
Input: An MRI image having brain tumor is provided as the input to this developed framework.
Step-2:
Processing: The processing of the input MRI image is done in 4 sub-steps as Pre-processing, Tumor segmentation, feature extraction and selection using Grey Wolf Optimization (GWO) which is Bio-inspired optimization algorithm. Later the results have been compared with other optimization algorithms like Particle swarm Optimization (PSO) and Firefly Algorithm (FA).
Step-3:
Output: Tumor with its class will be the output from the framework.

To increase the accuracy of any medical disease classifier, input images need to be pre-processed. In the invention, MRI images are pre-processed to remove any kind of noise in the images by using biorth 3.9 wavelet filter. In another step of pre-processing, skull are which is at outer circle around the brain which may sometime create illusion of having brain tumor so it have to be removed. Morphological opening and closing operations have been used for removing skull area.
For detecting tumor Fuzzy C Means (FCM) algorithm has been used for segmenting the tumor area from the brain. After segmenting tumor , total 14 features have three categories named shape, intensity and texture have been extracted using Grey Level Co-occurrence Matrix (GLCM) and Grey Level Difference Matrix (GLDM) methods. Now these features can be used for training the classifier so that it can predict the class of tumor for any input MRI image.
From the prior work in this field it has been observed that not all the features which are extracted will participate in increasing the accuracy of classifier therefore in this invention feature selection algorithm has been used to optimize the features. Optimized Multiclass Support Vector Machine (MSVM) has been used after reducing features using GWO algorithms which is based on the hunting behaviour of wolves.

Grey Wolf Optimization (GWO) Algorithm
Grey wolf hunting behaviour is the inspiration for the GWO algorithm[138]. The grey wolves hunt into a group and encircle the prey. Wolves follow the hierarchical structure in the group to hunt down the prey. As illustrated in Fig 2, the wolf in the closest position to prey will be at the top of the hierarchy, which is given the name alpha (a). This wolf gives command beta (ß), delta (d), omega (?)and others wolves in the group.

Fig 2 Hierarchy of grey wolves (dominance decreases from top down).
The wolves encircle the prey, which may be represented mathematically as:
D=|(C.) ?(X_p ) ?(t)-X ? (t)| (1)
X ?(t+1)= (X_p ) ?(t)-A ? .D ? (2)
where, the current iteration is indicated by t , A ? and C ? are coefficient vectors, (X_p ) ?and X ? are used for position vectors of prey and the grey wolf and D is the distance between prey and wolf.
(A ) ?and C ? which are position vectors can be determined as:-
A ?=2 a ? .(r_1 ) ?-a ? (3)
C ?=2.(r_2 ) ? (4)
Here components of 'a' are linearly decreased from 2 to 0 of iterations and r1, r2 both are random vectors in [0, 1].
In the hunting process the beta and delta follow the position of alpha to reach at optimal position by means of the prey. The algorithm considers that alpha, beta and delta wolves are at optimal positions than others, so keeping these three's positions in consideration, other wolves are updated as:
(D_? ) ?=|(C_(1 ) ) ?.(X_? ) ?-X ?|,(D_ß ) ?=|(C_(2 ) ) ?.(X_ß ) ?-X ?|,(D_d ) ?=|(C_3 ) ?.(X_d ) ?-X ?| (5)
(X_1 ) ?= (X_? ) ?-(A_1 ) ?.((D_? ) ?),(X_2 ) ?=(X_ß ) ?-(A_2.) ?((D_ß ) ? ),(X_3 ) ?=(X_? ) ?-(A_3 ) ?.((D_d ) ? ) (6)
(X ) ?(t+1)=(? X?_1 ) ?+(X_2 ) ?+(X_3 ) ? (7)
When prey is surrounded by eq.3 and 4 then the (a.) ?A ? is reduced from [-2a to 2a] to get near to prey where (a.) ? reduces from 1 to 0 as shown in Fig. 3. Eq. 5. calculates distance between prey and the alpha (a), beta (ß), and delta (d) wolves and by using Eq. 6 and Eq. 7 based on these wolves update their positions.

Fig.3 Position vectors with possible next locations in Grey Wolf Optimization
Results and Discussion
This section describes the step-by-step process of classification of brain tumor using SVM classifier along-with GWO, PSO and FA optimization algorithms. Later tumor classification is also done with two kNN and Naive bayes classifiers also. There performance is compared with all the three optimization algorithms.
(a) Data Set
In this proposed work, a publically available data set from "figshare.com" has been used. This brain tumor data set consists of 3064 T1-weighted contrast - enhanced MR Images which have three types of tumor. After the pre-processing of these images, tumor is segmented properly as proposed. Tumor images have been assigned three labels each for the tumor class. Label "Type 1" represents the Meningioma tumor which has 708 slices, label "Type 2" represents the Glioma tumor which has 1426 slices and label "Type 3" represents the pituitary tumor which has the 930 slices after segmentation. In these images how these three category tumors are presented in detail is given in the Table 1 and shown in Fig 4.
Table 1. Labels for Tumor Segmentation
Name of tumor Meningioma tumor Glioma tumor Pituitary tumor
Labels 1 2 3
No. of Slices 708 1426 930

(a) (b) (c)
Fig. 4 MR Images (a) Glioma tumor (b) Meningioma and (c) Pituitary tumor
(b) Feature Extraction and optimization
Total 14 features are extracted from tumor segmented MR Images. These features are "Area, Perimeter, Circularity, Mean, Intensity gradient, Standard Deviation, Skewness, Kurtosis, Contrast, Correlation, Energy, Entropy, Homogeneity and mean of probability density function". Feature selection is done with the help of optimization algorithms. GWO which is proposed algorithm works on the principle of wolf hunting behavior as shown in previous section. The objective function for this will be the accuracy of the classifier which needs to be maximized with different features subsets for tumor detection. The proposed method is based on machine learning (ML) model that understands the language of numeric only so the feature set has been converted into the form of strings and many attributes have a maximum number of zeros which don't contribute to classification and bias the network training. So the module removes them from the features set. Data normalization has also been performed because some features have high numeric values and some of them have comparatively small values.
Grey wolf optimization and data classification are two isolated algorithms but these work in a closed loop scenario. Both modules work in equilibrium, GWO gives the input as a binary matrix to the ML module and gets the accuracy in its input from the ML module as shown in Fig 5. This binary matrix is the set of features which must be included. The '1' in the matrix represents the feature selected and '0' represents that this feature is not selected.

Fig. 5 Relation between feature selection and GWO optimization

Fig 6 GWO-SVM framework for classification

(c) Classification:
The GWO-SVM is proposed framework for brain tumor classification is shown in Fig. 6 that takes accuracy as an objective function that have to be maximize by selecting only those important features. In the same way as GWO has been used for feature selection, PSO and Firefly algorithms could be applied for feature selection.
Radial Basis Function(RBF) Kernel based Multiclass-SVM has been used for classification at the initial phase. To evaluate the performance of classifiers data has been divided in 80:20 ratio and 80% images is used as training data and remaining 20% is used as testing data. So, in this case out of 3064 total images 2451 images have been used for training and 613 images have been used for testing. In second phase, two more classifiers were also compared to evaluate their performance with optimization algorithms. Table 2 shows types of tumor classification with input image, segmented tumor image and classification image.

Fig. 7 (a) illustrates the accuracy of multiclass SVM classifier for a given MRI which is shown in Table 2. GWO algorithm takes 9 features instead of total 14 features as shown in Fig. 7 (b), and produced accuracy of 99.3% which is almost 23% improvement than classification without optimization that was 76.12%. Other two optimization algorithms PSO and FA also improved the accuracy by 20% and 15% respectively. PSO shows 95.56%accuracy while FA gives 90.04 % accuracy by considering less numbers of features instead of all. It can be observed that accuracy of classifier with all optimization algorithms has increased by significant amount.
Table 2 Tumor Classification after Segmentation
Input Image Segmented Tumor Image Classified Image Tumor Type
Type-1
Type-2
Type-3

(a) (b)
Fig. 7 Accuracy of optimized features with GWO and SVM classifier

Achievement of Invention:
Apart from achieving several objectives, very important contributions has been done for the medical industry for classifying brain tumor automatically. The main achievement of invention is its high accuracy as compared to some of the research proposed in past some years as shown in Table 3.
Table 3 Comparision of different techniques
Researchers Method Accuracy(%)
Cheng et al., 2015
Bow-SVM 91.28
Ismael, 2018
DWT-NN 91.90
Pashaei et al., 2018
CNN-ELM 93.68
Abiwinanda et al., 2019
CNN 84.19
Afshar et al., 2019
CapsNet 90.89
Deepak & Ameer, 2019
CNN 97.1
Method used by us GWO-SVM 99.3

This Invention is also contributing for future researchers who are working on the brain tumor segmentation and classification problems. There is many more application of this research work and following are some of them which can also represented as the scope of the work:
Noise removal in medical images.
Automated Brain tumor segmentation.
Brain tumor classification using soft computing.
Feature selection using bio-inspired algorithm.
Performance evaluation of different classifiers in brain tumor classification.
Increasing the accuracy of brain tumor classifiers by using optimized features.

While the invention has been described and illustrated with reference to certain particular embodiments thereof, those skilled in the art will appreciate that various adaptations, changes, modifications, substitutions, deletions, or additions of procedures and protocols may be made without departing from the spirit and scope of the invention.

Claims:

We claim that, this invention will do brain tumor classification with high accuracy using soft computing.
We claim that we can improve the performance of classifier by using bio-inspired optimization algorithm.
We claim that Grey Wolf optimization algorithm increases the performance of different classifiers more than Particle Swarm Optimization and Firefly Algorithm.
We claim that, this invention will provide help in the proper treatment of patients suffering from brain tumor in remote area where good neurologist are not available.

Abstract
Traditionally, brain tumor detection and classification are done by manual investigation but proper diagnosis varied from person to person and is also a very time-consuming process. Well experienced doctors can do precise diagnoses but they are far away from the reach of all the patients and these doctors are also overburdened. So, to reduce their workload and to provide accurate diagnosis in less time, there is a requirement to automate this tumor segmentation and classification part. Many new methods have been proposed by researchers in the last few years to fulfill this task. But most of the methods only segment and classify the tumor region if it exists, they did not classify into different types of a brain tumor which are most frequently occurs so there was research scope available to classify these brain tumor automatically with accuracy in less time. Therefore, in this work, a more accurate and reliable brain tumor classification method, for designing CAD systems has been proposed by using soft computing.
Multiclass Support Vector Machine(MSVM) with optimized features has been used for classifying the type of brain tumor. Classifier is trained with features which are extracted from tumor image after optimizing with Grey Wolf Optimization (GWO) which is based on soft computing. Trained classifier will predict about the type of brain tumor in the input query MRI image. Accuracy of this Optimized MSVM classifier is 99.3 %.

Claims:
1. We claim that, this invention will do brain tumor classification with high accuracy using
soft computing.
2. We claim that we can improve the performance of classifier by using bio-inspired
optimization algorithm.
3. We claim that Grey Wolf optimization algorithm increases the performance of different
classifiers more than Particle Swarm Optimization and Firefly Algorithm.
4. We claim that, this invention will provide help in the proper treatment of patients
suffering from brain tumor in remote area where good neurologist are not available.