Classification & Feature extraction of Brain tumor from MRI Images using Modified ANN Approach

░ ABSTRACT-In this research paper, a new modified approach is proposed for brain tumor classification as well as feature extraction from Magnetic Resonance Imaging (MRI) after pre-processing of the images. The discrete wavelet transformation (DWT) technique is used for feature extraction from MRI images and Artificial Neural Network (ANN) is used for the classification of the type of tumor according to extracted features. Mean, Standard deviation, Variance, Entropy, Skewness, Homogeneity, Contrast, Correlation are the main features used to classify the type of tumor. The proposed model can give a better result in comparison with other available techniques in less computational time as well as a high degree of accuracy. The training and testing accuracies of the proposed model are 100% and 98.20% with a 98.70 % degree of precision respectively.


░ 1. INTRODUCTION
In such a modern day's there is no expectation to live a good life without healthy body organs.According to World Health Organization (W.H.O) brain tumors are the most widely recognized disease and worldwide more than 4 lakhs of human beings are diagnosed with the brain tumor per year.Radiologists and doctors required a large amount of time to diagnose the tumorous cells in MRI images and exact labeling of tumorous cells is also a more typical and time-consuming task.Cell generation and dead cell replacement is a process that is controlled by the human body, if this process is failed due to some malfunctioning of the human body then a large number of cells are generated.If these extra generated cells in the place of one cell gain some mass, then called a tumor.So tumor is the uncontrolled growth of cells due to malfunctioning of the human body.There are mainly two types of tumor; Benign (Non-Cancerous) and malignant (cancerous) tumors.Benign tumors are noncancerous means they cannot spread to another part of the body.The extra generated cells are covered by a membrane and easily removed by a small surgery.If these cells are diagnosed in an early stage, then chances to suffer by the cancer of human being is reduced.The malignant tumor is cancerous and they are easily traveled to the other part of the body through the bloodstream.In the last decades, lots of techniques are developed to diagnose brain tumor cells but still, there is a chance to improve the accuracy and precision.In this approach, the error is removed by using 2d filtering then main features are extracted using DWT, and further the ANN model is designed to classify the type of tumor according to provided features.This proposed model takes less time to predict the type of tumor with a higher order of accuracy and precision.

MRI Dataset and Image Pre-processing
The MRI dataset is collected from open source, https://www.kaggle.com/navoneel/brain-mri-images-for-braintumor-detection/data, in grayscale form i.e. intensity range from 0 to 255.A set of eight images are selected randomly for experimental purpose.The sample of input MRI images is shown in figure 2.

Figure 2: Sample data set of Brain MRI Images
The main purpose of pre-processing step in tumor diagnosis from MRI image to enhance the quality of the image in terms of black and white pixels.The noise of the MRI image is reduced using a median filter then the corresponding histogram is constructed to visualize the frequency distribution.A sample of eight image histograms after preprocessing is shown in figure 3.

Confusion Matrix
The performance measures of the proposed ANN model are measured with the help of confusion matrix parameters.Figure 8 (a) & (b) shows a confusion matrix in which, the x-axis of shows the predicted labels (model output) and the y-axis represents the true labels and corresponding heat map for predicted values.Some ANN model performance parameters are calculated is calculated by using given equations ( 4), ( 5), ( 6), ( 7) as under: ( The accuracy of different performance parameters like 'Precision', 'Sensitivity', 'Specificity', and 'Accuracy' of the proposed ANN model are shown below in Table 2.

Receiver Operating Characteristic (ROC) and Area under the ROC Curve (AUC)
ROC is used to plot the performance of the classification of the proposed model by plotting True Positive Rate (TPR) and False Positive Rate (FPR).The probability model performance measure is represented by the Area covered under the receiver operating characteristic curve.Figure 9 shows a ROC curve of the proposed ANN model with AUC 0.9863 which means the proposed model can predict correctly up to 98.63 %.

░ 2 .
PROPOSED METHODOLOGY The proposed model is a hybrid model for brain tumor classification and segmentation that consists mainly; MRI Image Pre-processing, Discrete Wavelet Transformation (DWT), Principal Component Analysis (PCA), Artificial Neural Network (ANN).The proposed model classifies the New MRI image into 4 stages only are; Noise removal, feature extraction, feature reduction, and classification.The Proposed model has been illustrated in figure 1.

Figure 1 :
Figure 1: Block diagram representing different stages of the proposed methodology

Fig. 3
Fig.3 Histogram of a Pre-processed set of Brain MRI Images

░ 3 .
ARTIFICIAL NEURAL NETWORKAn artificial Neural Network is a layered network consist mainly of three layers; the Input layer, Hidden Layer and Output Layer.Every layered has some pre-processing unit, called a node.The number of a node in the input layer is equal to the number of features and the number of nodes in the output layer is equal to the number of labels or classes.This model is used to classify the input data in one of the predefined labeled classes after training the model with hues amount of the same type of classes labeled data.The neural network mainly deals with human brain learning and it consists of a neuron that is responsible to create the layers in the ANN model.These neurons are also called tuned parameters.The structure of a single neuron is shown below in figure5.

Fig. 7 Feature
Fig. 7 Training accuracy and loss of proposed ANN Model

Table 1
shows the sample of extracted features using 2D DWT for input MRI image numbers 3 & 4.

Table 1 .
Extracted features from MRI Image

Table 2 :
Accuracy metrics of performance parametersThese performance parameters are extracted from the 'Confusion Matrix'.The highest performance of 'precision', 'sensitivity', 'specificity', and 'accuracy' for the benign and malignant tumor are shown in the table.The overall accuracy of the proposed ANN model for tumor classification is 98.70 %.