Description:Field of Invention
Now a days research on human brilliance plays a key role. Artificial Neural Networks, Deep learning and machine learning are used extensively in most of the research related to medical simply these are used in aspects of biomedical technology. The reason for using these advanced concepts in biomedical is the solution is not in terms of linear fashion. Generally, there is no particular algorithm for identifying the diseases so Deep neural networks comes into picture and diseases are recognized by using scanning.
One more problem is most of the health organizations have limited number of resources in terms of capital and difficult to update the old infrastructure and technologies. So, improve the performance of disease reorganization and reduction of cost medical industry move towards new approaches like machine learning, artificial intelligence and deep neural networks. From the latest survey most of the health organizations uses these latest technologies for easy identification of brain tumor and liver disorder. By using Artificial and deep neural network most of the diseases like cancer prediction, speech recognition etc.
Background of the Invention
As per the report published by World Health Organization (WHO) in 2018, it is found that the number of cancer cases has increased to 18.1 million fresh instances and 9.6 million fatalities worldwide. The overall death rate due to cancer is increasing rapidly which may create an alarming situation in the near future. Early detection of abnormality is essential for reducing the fatality rate. There are tons of costly, painful, and time-consuming cancer tests and processes that can prolong the detection and subsequently the treatment. The patient usually undergoes a futile biopsy test in the event of the incorrect diagnosis. The new method i.e., artificial neural network and machine learning are being used in this domain resulting in early detection of many kinds of cancer. Automating this analysis has helped the radiologist in enhancing the efficiency of the diagnosis. It can be helpful as the second reader and can reduce the level of errors. Nevertheless, there are not many efficient and accurate strategies. Thus, with the assistance of the recent open-source technologies, we will use a novel, efficient, and viable strategy. It will improve precision and reduce false negativity.
Around the year 2006, the evolution in machine learning, new age research emerged and was called deep structured learning or hierarchical learning or more frequently referred to as deep learning. Deep learning research took momentum at various universities. Many prominent healthcare centers and corporations such as Siemens, Philips, Hitachi, GE, etc., have also come together and joined hands to work towards the optimal solution of large medical imaging (KR1020210067539A). These companies are making significant investments in the field of medical diagnosis using imaging applications.
Feed forward neural networks with many numbers of hidden layers are generally stated as DNN, are more suitable for the models which contain deep architecture. In the early 1980s Back-propagation becomes the most popular algorithm for learning the parameters of these networks. An introduction of a novel and more comprehensive visualization method, which established a top-down generative method in the reverse direction of the classification network and gave understanding into the functioning of intermediate feature layers, are created by Zeiler and Fergus to analyze the characteristics reproduced by deep convolutional networks from image data. The capability of deep networks is demonstrated in their capacity to develop suitable features and to discriminate collectively. The feasible and better way to review ANN's background is to use a "hype cycle," a visual representation of the maturity, social implementation of relevant techniques. It reflects the evolution of technology or application over a period. The five phases are 1. Technology trigger 2. Peak of inflated expectations 3. Trough of disillusionment 4. Slope of enlightenment 5. plateau of production.
Since 1990, convolutional framework with deep models becomes most popular in use of image recognition and computer vision. Simply it is a neutral network which consists of two different layers. One is s-layer acts as an extractor and other one is C-layer acts as structured associations. The hierarchical Artificial Neural Network is used in most of the applications like pattern recognition and handwriting recognition. By using convolution operation with distinct kernel options different activities that are performed on the images are identified. Random matrices are used as convolution operator that recognizes blurring, edge detection and sharpening etc.
The first Convolution Neural network consists of two layers. One is Convolutional Layer and other one is Sub sampling Layer and these two layers are connected for classification with fully connected layer and an RBF layer. Under Convolution Neural network, AlexNet began the era of Image Net classification. The AlexNet is categorized into 2 channels and those are operated on two GPUs to reduce expensive training. It uses augmentation used for horizontal reflection, image translation and patch extraction. In the year 2012, images processing developments improve a lot. In the Image Net LSVRC competition, the task is to train a machine with 1.2 million very high-resolution images and classify images into one of 1000 separate image groups. The deep Convolution Neural network is used for test set that consists of 150k images and it reduces error rates compared to existing models (CN201810300057).
Summary of the Invention
The main invention is to design an efficient and usual brain tumor classification with high performance and high accuracy. The tasks of preprocessing, segmentation, feature extraction and classification are analyzed in proposed system. In preprocessing for removing noise filters are used and for extraction of features such as analysis of textures in MRI brain images GLCM and DWT techniques are used. In DWT the features are extracted by partitioning the signal to get coefficients and approximation into third level using decimation and filtering techniques. In GLCM, geometric methods are used for texture examination using the spatial relations among the pixels of the image. To perform segmentation, k-means clustering technique is used to form four clusters which consists of MRI images of different features of images.
Brief Description of Drawings
Figure 1: Convolution neural network model for the detection of brain tumors in MRI images.
Detailed Description of the Invention
The main objective of the invention is to develop an intelligent brain tumor detection system based on an efficient feature extraction technique using deep neural networks. The real challenge is to detect the abnormal MRI brain images having tumors based on the accuracy i.e. the false positive should be less. The invention categorizes brain tumor based on texture characteristics and shape of the features. The datasets which consist of Magnetic Resonance Imaging of brain are gathered from different medical organizations. For Analysis and preprocessing of images MAT lab is used. After collecting the dataset and storing it in the particular folder different classification techniques are performed to get the accurate result.
The process is in first step the pre-processing stage Median filtering is applied to remove salt and pepper noise. After completion of preprocessing stage Segmentation using K-means clustering is applied and then DWT technique is used to partition the image to the 3 rd Level and then GLCM is used to find different statistical features. Once features are selected then these selected features are provided as input to CNN classifier for classification of tumor. Finally, Performance evaluation is used for testing any image from the test folder to detect the normal or abnormal image.
In the first step an input image is retrieved from the data base by using imread() function. After reading the image, the size is reduced to 256x256 by using imsize() function. After resizing, by using rgb2gray( ) RGB is transformed to grayscale. Database is partitioned into two parts. One part is used for training and other part is used for testing. After train a model take images from the test data and then preprocess the data, resize the image and perform color conversion. In this Preprocessing and Filtering phase, the selected input test image goes through the pre-processing implementation technique of resizing and reshaping. Pre-processing operation is done to improve the image quality using the filtering technique. Filters are mainly used to improve the pixel quality and remove the noise thus reducing distortion of given input image. The median filter is a nonlinear digital filtering technique, often used to remove noise from an image or signal. Such noise reduction is a typical pre-processing step to improve the results for later processing (e.g. edge detection on an image).
The proposed technique uses the Median filter using medfilt2( ) because apart from removing noise which is present in the image it preserves the edges of the image as well. The median filter works by moving through the image pixel by pixel, replacing each value with the median value of neighboring pixels. The pattern of neighbors is called the "window", which slides, pixel by pixel over the entire image. The median is calculated by first sorting all the pixel values from the window into numerical order, and then replacing the pixel being considered with the middle (median) pixel value. After completion of preprocessing feature extraction is performed. Feature extraction is the most important and challenging task to compute the textual characteristics of a digital image. Feature extraction is the first stage of image texture analysis. This activity will help in determining the texture classification and shape determination.
The Discrete Wavelet Transform will decompose the enhanced MRI brain image to obtain the decomposed coefficients. The decomposed coefficients are combined in the wavelet domain based on the fusion rule. The fused image is achieved by taking the inverse DWT on fused coefficients thus reducing the noise to improve efficiency The image in Figure 1 is divided into four regions, a region with high detail. The 1-D wavelet transform can be extended to a two-dimensional (2-D) wavelet transform using separable wavelet features. In a 2D image, the 2D transformation implemented separable filters of 1D transform along all the rows of the input and then repeated on all the columns. After one level of transformation the image can be further decomposed by applying the 2-D sub-band decomposition to the existing LL sub-band. This iterative process results in multiple “transform levels”. These wavelet decompositions in which a pair of wavelet filters including low pass filters and high pass filters are utilized to calculate wavelet coefficients. The image LL undergoes the decomposition using the same filters having always the lowest frequency component located in the upper left corner of the image. Each stage of the analysis produces next 4 sub-images whereby the size is reduced twice compared to the previous scale. Good texture segmentation results can be obtained within 2 to 4 scales of wavelet decomposition.
After performing all these tasks the test image is applied to k-means clustering. K-means uses image pixel as feature point in the location space. The k-means algorithm randomly selects the centers of the cluster. Each data point is assigned to the cluster based on closeness of the point with the mean. The process continues until no convergence. To delete unnecessary data imerode() is used with structuring element on the binary image.
If the training data is large then convolutional neural network becomes the most powerful technique. This invention uses convolutional neural network for MRI brain images. Figure 1 shows the convolutional neural network as classification model and data is divided as two parts training and testing. Consider 10 filters in test phase and dimensions of filter is 8*8 and dimensions of the polling region is 3. The images are categorized based on the labels like tumor and non-tumor images. At last, apply convolution neural network for automatic brain tumor classification. The dataset brain image is taken from the image net. The image net is one of the pre-trained models. The number of pre-trained model-based brain dataset is used for classification steps. So computation time is low meanwhile the performance is high in the proposed automatic brain tumor classification scheme.
The loss function is calculated by using the gradient descent algorithm. The raw image pixel is mapped with class scores by using a score function. The quality of a particular set of parameters is measured by the loss function. It is based on how well the induced scores approved with the ground truth labels in the training data. The loss function calculation is very important to improve the accuracy. The loss function is high when the accuracy is low. Similarly, the accuracy is high, when the loss function is low. The gradient value is calculated for loss function to compute the gradient descent algorithm. Repeatedly evaluate the gradient value to compute the gradient of the loss function. Apply convolution filter in the first layer, the sensitivity of filter is reduced by smoothing the convolution filter (i.e) subsampling, the signal transfers from one layer to another layer and are controlled by the activation layer. Accelerate the training period by using the ReLu. The neurons in the proceeding layer are connected to every neuron in the subsequent layer. During training, the Loss layer is added at the end to give feedback to the neural network.
4 Claims & 1 Figure , Claims:The scope of the invention is defined by the following claims:

Claim:
1. Diagnosis of Cancer using optimized deep learning algorithms comprising the steps of
a) Examines and surveys the present use of deep learning approaches for diagnosis of liver disorder and brain tumor.
b) Adopted a method to enhance the models of Artificial Neural Networks to improve the detection of liver disorder.
c) Presented a simplified model for efficient MRI segmentation and detection of Brain tumor which helps in identification of tissues that causes the tumor.
2. Diagnosis of Cancer using optimized deep learning algorithms as claimed in claim 1, Cascade Correlation algorithm on BUPA Liver disorder is designed to improve the identification of disorders.
3. Diagnosis of Cancer using optimized deep learning algorithms as claimed in claim 1, the abnormality in brain is identified by using convolution neural networks as it is used for clustering and segmenting the images and WCA optimization algorithm is applied for the optimal clustering of images.
4. Diagnosis of Cancer using optimized deep learning algorithms as claimed in claim 1, Presented a method Discrete Wavelet transform that detects the unusual MRI brain images having tumors with high level of accuracy and classifies the brain tumor image based on texture characteristics features and shape.