This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.
TUMOR AND ATROPHY DISEASE DETECTION IN BRAIN USING MRI
4.1 MATERIAL & METHODS FOR TUMOR AND ATROPHY
The data presented in this methodology was taken the Brain Web dataset . In this research methodology, Brain Web dataset was utilized with different BMRI images. Based on standard tissue segmentation mask, Brain Web datasets had given the MRI brain images with unreliable image quality and the datasets were too based on an anatomical structure of a normal brain, which had resulted from the tasks of registering and preprocessing of 27 scans from the same individual with segmentation. Different kinds of tissues were well identified in this dataset, both the types of tissue memberships “fuzzy” and “crisp” were assigned to each voxel. The sample brain MRI images from the Brain Web data set were specified in the figure 4.1 given below.
Fig 4.1 Sample images from Brain Web data set 
In this study, the material was depended on the Brain Web data . Brain images were taken for this study; it was examined and evaluated to detect the Brain tumor and Atrophy disease in the brain using MRI. The methods adopted in the research were Artificial Intelligence and Neuro Fuzzy Classifier for BMRI image segmentation for brain tumor detection and atrophy diseases; it was analyzed in this chapter 4. The brain images utilized in the data set, which was used to evaluate the various images and its atrophy level and detected the highest atrophy level the images in the Brain Web dataset. Thus the performance analyzes of the brain images was used to estimate the following parameters such as False Positive Rate, False Negative Rate, Sensitivity, Specificity and Accuracy, the overall performance of the system was determined. The basic count values such as True Positive (TP), True Negative (TN), False Positive (FP) and False Negative (FN) were used in this performance measures. Hence, the Brain Web images was included for the both the normal and abnormal images .
4.2 ARTIFICIAL INTELLIGENCE BASED BMRI IMAGE SEGMENTATION
Artificial Intelligence had been developed for range of applications such as function approximation, optimization, extraction, pattern detection and categorization. Especially, they had been developed for brain images; Features extraction, images segmentation, pattern and categorization. Among the above applications, brain image segmentation was very essential as it plays a vital role for the high standard of processing for brain tumor detection in the medical images. Artificial Intelligence was also used for various methods such as Back Propagation Networks, Radial Basis Function, Multi-Layer Perception, Self Organized Map (SOM) and Adaptive Resonance Theory . These neural networks had been utilized for brain image segmentation. These techniques had been classified into feed forward (associative) and feedback (auto associative) artificial neural networks.
Artificial neural networks had been demonstrated themselves as a technical categorizer and were especially well fitted for brain tumor categorization . The back propagation network was the important tool in this study. The input image was given to the network, which was not feasible because the weight matrix was large . Another method was features extraction from the given input images, which was used to train the network. The back propagation network  was a supervised learning standard for a feed forward, multi layer neural network that utilized tan sigmoid with predefined step size to accomplish training or supervised learning methods by using the error rectification process and weight updating in this study.
BACK PROPAGATION NETWORKS
Backward propagation network was originated from the supervised learning method. It was a nonlinear abstraction of the squared error gradient descent supervised learning approach for updating the level and weights of the neurons in a single-layer perception, extrapolated to feed forward networks. Back Propagation needed the activation function, which was utilized by the artificial intelligence nodes or neurons, which was distinguishable from its own derivative to a simple function . The Back Propagation Networks had permitted to equate the gradient of the relative error function to the hidden division and therefore to calculate the weights of the hidden division for determining its value by gradient descent in the same way as it to be estimated for the outcome sections. Back propagated errors were theoretically obtained by utilizing the classical calculus chain principle . In on-line supervising learning approach had estimated the weights of the artificial neural network were upgraded at each and every input information point presentation. It had expended the resilient back propagation version, which had employed only the derivative function to execute the weight upgrading.
Fig 4.2 Back Propagation Network 
The artificial neural network was aimed for training neither discrete value nor continuous valued features to evaluate the performance of the neural network for changing inputs to recognize the brain tumor detection. This methodology had required initialization of weights, errors of back propagation and updating the weights. There was pattern or pre-structured learning and batch or pre-epoch supervised learning. In pattern learning methodology, provided input images was patternized and then it was moved to propagated forward, the error was estimated and back propagated and weights were also updated. In batch learning methodology, the whole set of training networks weights were updated at the end and hence it had been demonstrated to the neural networks. Thus the update of the weights was calculated only after every epoch.
The neuron outcomes were equated in the output layer with the expected or direct outcomes . The propagation error had been calculated and then it was propagated through the neural network and this determined value was utilized for the weight updating and correction. The overall error was reduced at each and every state by using the neuron weight adjustment . Generally the supervised learning rate had fixed what quantity of the measured error gradients had been utilized for the weight correction . The accurate value of supervised learning rate was based on the error values. Some denotation had been provided by error calculation and noticed the previous weight corrections . To improve the accuracy in this methodology, accelerate the neurons training for the neural networks.
Artificial Neural Network training approach was estimated by a group of conflicting necessity such as flexibility, efficiency and simplicity . Simplicity was determined as the amount of the effort needed to utilize the computational complexity. The pattern of Back Propagation Network (BPN) was displayed in figure 4.2. Flexibility was associated to the extendibility of the approach to train various methodology architectures and efficiency which referred the computational necessity for training and the success of the training stage . Thus the weight matrix problem solution was viewed as a byproduct of the training stages it was executed effectively.
In this proposed methodology, all the brain MRI images were utilized in the experiment were in DICOM format and were developed from the original brain images. The size of the brain images utilized as 512 512 and were gathered from various patients. Hence, from these affected patients, 33 brain images were studied for this study. The 20 dataset was trained and examined .
In pattern detection network, which was referred as feed-backward network with tan-sigmoid transfer functions in both the hidden units and the output units, was utilized. The neural network had only one output neuron, because it had 20 input parameters. The hidden units of neurons were 100 and the supervised learning level was 0.1. The momentum component was 0.7 and overall epochs were 400. The error was reduced by 0.0012 and the performance of the categorizer was analyzed by computing accuracy of the images.
Fig 4.3 Detection of brain tumors 
The shape and intensity features of 33 brain images with tumors were calculated. The brain images were first classified into sub images and the representing features were received utilizing Gabor method. The experimental outcomes were displayed in the figure 4.3. The generalized features of the brain tumors area for each MRI slice were fixed and were fed to Back Propagation Networks. Utilizing the trained features, brain tumor segmentation outcome for the same selected combinations of the extracted features in the single mode of slice was found. As the outcomes, the brain tumor region was identified within a pixel size and precision level was also increased using trained artificial neutral networks. In this methodology, brain tumor detection was applying preprocessing step and then Gabor feature extraction and finally Back Propagation Networks categorization was introduced. The stimulation result had presented the categorization accuracy of 88 %. The system had been examined with various brain images. It was very important to utilize for huge number of patients information, which would also further enhance the precision level of the system .
4.3 PHASES OF NFBS (NEURO FUZZY BASED BMRI IMAGE SEGMENTATION)
Brain tissue segmentation from the MRI images was significance study in the medical research field. The accurate segmentation of the normal as well as the abnormal tissues was the complex assignment in this process. Because of the inconsistency and difficulty of abnormal tissues, MRI Brain Image Segmentation turned into more hard procedure. In this proposed methodology, a technique was proposed for segmenting the abnormalities such as Tumor and Atrophy in the MRI Brain images. By using three stages such as feature extraction, classification and segmentation were presented in this proposed methodology. At first, the features extraction was estimated by the following factors such as energy, entropy, homogeneity, contrast and correlation from MRI brain images were extracted. Later, by utilizing Neuro-Fuzzy classifier, the classification process was carried out and for this stages, the feature set was specified as the input. From the outcome of classification, the brain images were categorized into normal as well as abnormal. The further procedure segmentation was performed according to this outcome only. The abnormal MRI images were segmented into abnormal tissues like Tumor and Atrophy using Region Growing method. Utilizing MATLAB platform the implementation of the proposed technique was made. The experimentation was carried out on the MRI brain images by BrainWeb data sets. The performance of proposed technique was assessed with the help of the metrics namely FPR, FNR, specificity, sensitivity and accuracy. Therefore, using proposed technique with enhanced classification, the abnormal tissues of MRI Brain images were segmented accurately .
Initially, the BMRI images were given as input to the proposed work and the feature sets were extracted from these input images. From these feature sets, the images were classified into two kinds of tissues – normal and abnormal using the Neuro-Fuzzy classifier. Then the abnormal tissues Tumor and Atrophy were segmented using Region Growing Method. The proposed work was illustrated in Fig. 4.4. Thus the stages of the proposed methodology were demonstrated below such as given below:
- Feature set Extraction
- Neuro-Fuzzy classifier based Classification
- Classified tissue’s Segmentation
4.3.1 FEATURE SET EXTRACTION:
In brain tumor detection in the brain MRI images; feature extraction was a peculiar form of dimensionality simplification. When the brain input information was given to an algorithm, which was very huge to be processed and it was mistrusted to be notoriously surplus and then given input parameters information had been translated into a minimized presented group of features. Feature Extraction was useful in detecting brain tumor where, it was precisely occurred and assistants in identifying further stages. Translation of the input information into the group of features was called feature extraction .
In order to classify the given Brain MRI images, the features from these MRI images were initially extracted. In this proposed work, the statistical features such as Energy, Entropy, Homogeneity, Contrast and Correlation are extracted from these input BMRI images.
Fig 4.4 Proposed Neuro-Fuzzy based segmentation block diagram
4.3.2 CLASSIFICATION USING NEURO- FUZZY CLASSIFIER
The BMRI images were classified using the Neuro-Fuzzy classifier. The extracted features were given as the input to the Neuro-Fuzzy Classifier for classifying all the given BMRI images into 2 classes such as Normal BMRI images and Abnormal BMRI images. The Neuro-fuzzy system had a three-layered architectural design; the following diagram fig.4.5 showed the basic structure of the neuro-fuzzy classifier system. Neuro-Fuzzy classifier was a fuzzy based system that was trained by a learning algorithm derived from Neural Networks. The learning algorithm only performs on the local information and provided the local modifications in the fuzzy system. In general, a neuro-fuzzy system had generated very powerful solutions instead of using the system components individually.
Fig 4.5 Architecture diagram for Neural Networks
Fig. 4.5 had displayed that proposed methodology for segmentation of normal tissues using improved machine learning approach. In this proposed work, it had used BrainWeb datasets for experimentation which included both normal and abnormal images. Energy, Entropy, Homogeneity, Contrast, and Correlation were the features extracted from the brain MRI images . These features were then applied to a Neuro-fuzzy classifier that classifies the images into normal and abnormal. Proposed improved machine learning approach was a Neuro-fuzzy system that had a three-layered architectural design. This classifier was a fuzzy based system that was trained by a learning algorithm derived from neural networks. The learning algorithm performs only on the local information and provided the local modifications in the fuzzy system. In general, Neuro-fuzzy system had generated very powerful solutions when compared to the use of system components individually.
4.3.3 SEGMENTATION OF CLASSIFIED TISSUES
In medical image processing, segmentation plays crucial role in implementing in brain images. Segmentation had been useful in different applications such as identification of tumor in the brain, calculating tumor size and its volume, identification of the coronary cells in the angiograms, categorization of the blood cells , brain images and then finally in extraction of heart images etc . In generally, it had been useful for classify the images into anatomical area in few diligence for example muscles, blood cells, bones, and blood vessels, mean while few had been in the pathological area for example multiple sclerosis, cancer cells, and tissues deformities. To separate images into area, which were similar to few characteristics and it had basic objective in the segmentation methods . The main goal of the segmentation was to detach the entire images accurately into sub areas considering white matter, Cerebrospinal fluid and gray matter of the brain MRI . Particularly, while including the quantity of the neurological diseases such as Alzheimer’s disorder and Multiple sclerosis, the size and its volumes modified according to the total brain, gray matter and white matter was offered as main region in axonal loss and neuronal .
The Brain MRI images were categorized according to the neuro-fuzzy classifier, later the brain mages were made available in this two different images only such as pathological images or abnormal and normal. In general normal images used to have the normal tissues such as cerebrospinal fluid, gray matter and white matter, which were utilized for segmentation. The main step in segmenting the brain images, at first the images as to be preprocessed and then it was performed only in the normal brain images, it was applicable for abnormal brain images. Finally, it was unproblematic to settle the cerebrospinal fluid tissue of the normal brain images in the region that covered the cortex by the usage of the pre-processing approach. To segment the brain tissues, it had gone through the preprocess stage and then it was shifted to the classification of the images effectively.