INTRODUCTION:
It is observed that the capability of information classification in the hybrid system, combination of the fuzzy and neural network makes this automatic diagnostic is an engrossing research platform. Information processing in biomedical images is playing a vital role in the interpretation of disease of the human body. The human brain is a very complex structure which is cannot be analyzed by simple imaging technology. The Magnetic resonance Imaging (MRI) technology promising the highest quality image analysis information of the human brain, which is very useful for clinical and biomedical research platform[1-4].
The crux of this proposal is to classify the brain images into normal and abnormal respectively. The abnormal images further classified on the basis of detecting diseases and size of the tumor. It is very difficult to a radiologist to predict the disease from the set of MR images, as he is adopting the orthodox process for classification of the MR images. The brain MR images provided by the scanner are not providing the detail's information about the patient's brain, the radiologist is not able to diagnose properly. The preprocessing should be done with the help of mathematical tools for picking the maximum and accurate information from the scanned images. These are features based MR images, the fast DWT promising the highest quality multi-resolution feature extraction with various levels, but the wavelet has a limitation of data storage and large computation. To overcome storage issue and computational cost it needs to reduce the feature dimension, which followed by the principal of components analysis. It reduces the dimensions of the features[1-3] [11-18]. It is clearly shown that, the proposed method proving the classifier accuracy around 98 % than others scheme. This indicates the proposed system has great ability to define the brain abnormalities in MR images. Further it can use for medical research.[2][17-18].
Related work:
In a recently published article the various schemes of MR image classification proposed by the researcher. The overall approached applied by the researcher is on the basic preprocessing components, i.e. feature extraction, feature reduction and classification. Y. Zhang and L. Wu presented in [2] classification of the brain. In this select brain diseases which are commonly found in brain illness patient like glioma, meningioma and perform experiment on set of images. Y. Zhang et al. represented hybrid method in [3]. The abnormalities of brain illness is based on forward neural network (FNN) .in this DWT and PCA has been introduce for feature extraction and size reduction respectively. The intention of this proposal is to focus on reducing the computational complexity of image classification. It is based on wavelet, PCA and K fold Neuro-Fuzzy Classifier.
Proposed methodology:
This proposal following the basic three preprocessing techniques, district wavelet transform, principal components analysis and k fold fuzzy c mean. The functionality of this hybrid method depicted in figure 1.
Fig.1 Proposed medical expert NFC system with basic components.
The proposed methodology preprocessing scheme has been illustrated in above figure 1. This scheme has been divided into two phases; primary and secondary respectively. The primary phase consists of a set of brain MR images, wavelet Transform and principal component analysis. A brain MR image database consists of different age group patients with various common brain diseases. To extract the common features from the MR images wavelet tool has been used especially discrete wavelet transformation. It extracts the various features and dimension of the set of images. The processes of extracting the feature through wavelet are quite complex and time costing. It also increases the size requirement of the storage system. To overcome the basic issue we need to reduce the feature and dimension of the images. This can be effectively done through the Principal component analysis[12]. A PCA is playing a greater role in any classifier. To optimize the database it is necessary to reduce the dimension of the images which also result in less memory. The outcome of phase first used for machine training[13-14]. With reference to primary phase, the secondary phase illustrating the testing of the new database of MR images with the primary one. This can be done through Neuro-Fuzzy classifier. The testing of new MR images with a trained set of the data base will perform with the help of k folds Neuro-Fuzzy classifier. The selected features take into account for prediction of normality and abnormality in the brain. The accuracy of the classifier reaches to high rate with this scheme. Here the hybridization of component concept playing a magical role to sort out limitations of others MR image analysis methodology[5-7] [15-16]. De-noising: Better quality of images it is necessary to have a good value of peak signal to noise ratio (PSNR) , before processing on images, it should denoise first. The simple filters were used for filtering the noise such as Gaussian noise, speckle noise and impulse noise.
DWT and PCA:
A transformation is a just remapping of the signal in which the information can be viewed in time as well as frequency domain. The WT is used to perform feature extraction from the MR Images. The problem of continuous wavelet transforms oversampling overcome by DWT. This Cause, increase in a computational cost. Time domain filtering can be used to decompose the signal into different bands of frequency. The detail information of the signal in various resolution levels can be analyzed through DWT. The scaling function and wavelet function are employed by DWT. In wavelet, periodic convolution is often used, in which data set sequentially iterated level by level [5, 11]. While the extraction of features, DWT reduces the features, but computational time gets increased. This can be done through The PCA, in which the computational, as well as the dimensions of the image counter, fitted. The PCA works on the ratio of the total variance of the extracted and reduced features set [12]. The main job of these components is to trim the dimension of features estimated from DWT. These give rise to bring efficient information to the classifier for executing judicious decision [4-5].
Features:
There are various features extracted from the set of images based on texture and intensity-based features. Here we are considering total thirteen features such as Contrast, Correlation, Energy, Homogeneity, Mean, Standard Deviation Entropy, RMS, Variance, Smoothness, Kurtosis, Skewness and Inverse Difference Movement. The features extracted are discussed below with assuming (P→0) (i ,j) is the gray level co-occurrence matrix of the image I(x, y) as intensity pair i and j [19].
· Contrast(C):
it is change in intensity value in images and represented by :
(1)
· Correlation (S):
It is a measure of correlation of a pixel to its neighbor pixel and given by:
(2)
· Homogeneity (H):
It gives the diagonal represents pixels having same intensity in the vicinity. shown by
(3)
· Energy (E):
It is a sum of squared elements in gray level matrix. Shown by
(4)
· Mean (m):
it is the sum of intensity of pixels divided by total pixel in the region.
(5)
· Standard deviation (Std):
It gives the grey level around the mean.
(6)
· Skewness:
It is the measure of symmetry and asymmetry of the gray level around the mean.
(7)
NEURO-FUZZY CLASSIFIER (NFC):
In the above mentioned Neuro-Fuzzy system, k folds fuzzy, c means algorithm has been used for training and testing the MR images. Due to fast convergence time the ANFIS is not suitable for classification. This limitation has been overcome by the Conjugate Gradient Neuro-Fuzzy Classifier (SCGNFC), in which the optimization and k fold fuzzy, c means algorithm performed on a set of features information extracted from the previous preprocessing unit. It provides the higher convergence time, simple and efficient. NFC interference systems are fuzzy ruled based systems, which embrace with database, rules and decision unit. K-fold system is to constitute for segregation of the complete data set. It repeats K times and k-1 fold for training data sets; the remaining K folds data sets used for validation of the final result [9-10], [13-16].
EXPERIMENTAL RESULTS:
This section presents the experimental results of the hybrid NFC scheme. The database consists of various common brain diseases has been analyzed to evaluate the performance and accuracy of the proposed approach. In the extraction of common features, the numerous MR images were taken as a database. In which separate data set of an abnormal and normal brain of human has been examined. The abnormalities in brain examine by marking the above threshold value .It detects an abnormality in the brain by picking the threshold value of the features. Here we are describing the two cases, case-1 and case-2, in both cases the patients suffering from the tumor infection with basic brain disease. Figure 2 illustrated the base input MR image for case-1 and case-2. These images were prepossessed by DFT and PCA to extract the features and reduce the features, which shows in figure 8 and figure 9. The threshold value images and segmented images have been depicted in figure 3 and figure 4 respectively. The selected features from the set of images in both cases were used for training the system, the rule view of case-1 and case-2 represented by figure 6 and figure 7 respectively. The optimization carried out by k-folds fuzzy C-means algorithm. It causes the performance and accuracy of the system achieved the promising level. Figure 5 shows the final result of the NFC in the dialog box. All the analytical, mathematical analysis has been done in MATLAB.
Fig.2 Input image of case-1 and case-2
Fig.3 Process images of case-1 and case-2.
Fig.4 Segmented images of case-1 and case-2.
Fig.5 Result case-1 and case-2.