Detection of Brain Tumor using Deep Learning and Combination of Spatial Information

Download Full-Text PDF Cite this Publication

Text Only Version

Detection of Brain Tumor using Deep Learning and Combination of Spatial Information

1Sridhar S.R

Asst. Prof/CSE Muthyammal Engineering College,

Rasipuram, Tamilnadu

2Aravinth S,

IV Year/CSE Muthyammal Engineering College,

Rasipuram, Tamilnadu

3Dineshraj B

IV Year/CSE Muthyammal Engineering College

Rasipuram, Tamilnadu.

4Hariharan P

IV Year/CSE Muthyammal Engineering College,

Rasipuram, Tamilnadu.

Abstract: The brain tumor classification of an infected tumour area from EEG signals by the method of the extraction are a significant concern and time- consuming task performed by medical specialists by experience only this accuracy depends consequently it's essential to beat there by computer- aided technology (CAD). This paper deals with the method to extend the performance and reduce the issue includes within the interface method, which has considered Berkeley Wavelet Transformation (BWT) based segmentation with k-means clustering. The proposed method first employs wavelet transforms to extract features from signals is followed by applying Principle Component Analysis (PCA) to scale back the size of features. The reduced features were submitted to a Kernel Support Vector Machine (KSVM). The signal-to noise ratio is improved to eliminate the effect of unwanted noise and combined technique is employed to detect and classify tumour from brain EEG signals. When compared with the manual detection performed by radiologists or clinical experts, the experimental results performed on the various signals it's clear that the study of the tumor detection is fast and accurate. The proposed approach shows within the experimental results it can aid in accurate and detection of the tumor in exact location at time. Moreover, to extend the accuracy and quality rate, this approach extracts the statistical features using Berkeley Wavelet Transform and also the result's then given to Kernel Support Vector Machine for tumour classification as Benign or Malignant. The techniques are less complex, efficient and provides higher accuracy.

Keywords: BWT, EEG signals, KSVM, Wavelet Transformation, K-Mean Clustering.

1. INTRODUCTION

Brain is that the management centrewithin the build. it's responsible to execute all activities throughout an outsized number of connections and an enormous number of neurons. brain tumor is one in all the foremost serious diseases, occurred thanks to an abnormal growth of cells within the brain, affecting the functions of the systema nervosum. There are differing types of brain tumours, which might be either malignant or benign. the first stage of tumour detection depends on the physician's knowledge and knowledge, making the patients have an

opportunity to recover his life and survival. an automatic organization of brain tumours is a good tool for supporting the physicians to follow a successful treatment option. Such system uses the photographs captured by resonance (MR) imaging devices, which are widely employed by the radiologists of brain diagnosis. In recent years, several studies are proposed and different automated systems are developed for detecting and classifying brain tumours using MR images. for example, a technique for brain tumor segmentation supported a hybrid of fuzzy cmeans algorithm and cellular automata. during this method, the seed-growing problem of segmentation methods is solved by employing a new similarity function with a gray level co- occurrence matrix (GLCM) and evaluated on BraTS2013 dataset. an automatic method for detecting brain tumor supported image segmentation and tumour extraction. The authors utilized the circularity feature and also the area to extract the tumour from segmented brain images. The authors validated their methods by comparing their segmented images with the bottom truth images and achieved a mean of 0.729 (i.e., 72.9%) similarity, developed a semi- automatic 2 technique for MR brain image segmentation supported human interaction to come up with a feature map from MR images and used it to initialize the active contour model for segmenting the Region of Interest (ROI) area. Overlap index parameter and Jaccard coefficient are accustomed compare the results with ground truth ROI images, which are manually segmented from the initial images. Multistage approach which detects brain tumor from MR images employing a set of steps, including image preprocessing through cropping, noise reduction, scaling, and histogram equalization; feature extraction using histogram and GLCM techniques and classification using random forest (RF) classifier.

A dataset of 120 patients are utilized to test this approach and they achieved 87.62% of classification accuracy. A wavelet-based method to extract features from MR images. An automatic detection method to detect brain tumour from 3D images. After that, the RF method was used for brain tumour detection. Classification of brain tumours from computed tomography (CT) images

using deep learning methods such as multiple convolutional neural networks (CNNs) with discrimination method and single CNN method have also been proposed. Recently, the deep learning method to classify brain tumour either to malignant or benign using extreme learning machine local receptive fields (ELM- LRF). This method was evaluated on a dataset that consists of images collected from sixteen patients.

Even though deep learning methods improve the 3 classification of brain tumours, they need a large amount of training samples and a high cost of computation, and they take a long time for training. Defining whether a tumour is benign or malignant according to the image is also a subject that the medical community has not stopped. The feature extraction of the image is to first segment the image, and then extract the texture features in the image after segmentation. There are a lot of textures on trees, cloths, and clothes. These things that can be seen visually are taken from their textures, of course medically, for tumours, it is not as intuitive as trees or clothes. It can only be sliced into tumours, relying on imaging equipment to make its texture features into medical images such as CT or MRI. Texture is a basic and very useful information feature in an image. It is an important parameter for describing image content, and its academic research has gradually become a new topic.

  1. LITERATURE SURVEY

    BRAIN TUMOUR SEGMENTATION METHODS [1]

    Author: Demirhan

    Proposed a novel tissue segmentation method that has segmented brain MR images into tumour, WM, GM, and CSF and edema. The recognition of the healthy tissues of brain has been executed concurrently with the diseased tissues since change was examined that is happened by the increase of tumour on healthy tissues of brain can be most significant for planning of treatment process. In this work T1, T2, and Fluid-attenuated inversion recovery (FLAIR) MR images of 20 subjects were utilized that were suffered from glial tumour. Before the segmentation process, an algorithm has been enhanced for striped the skull. The image segmentation was done by applying the self-organizing map (SOM) that has been trained with fine-tuned with learning vector quantization (LVQ) technique and the unsupervised learning algorithm. In this method, an algorithm was enhanced to cluster the SOM replacement of an additional network. Input feature vector has been measured with the features that have been acquired from the coefficients of stationary wavelet transform (SWT).

      1. EANS AND FUZZY C-MEANS CLUSTERING METHODS [2]

        Author: Andac Hamamci

        The brain structure is explored by applying MRI image technology. Brains MR scanned image was focused for brain tumour segmentation. The MRI image

        filter can be more pleasant compared to a number of other outputs for examination. It cannot be control the human body as it does not sharpen some radiation waves. The brain tumour segmentation in digitization MRI scanned image can be difficulties and therefore, it will be decisive to medical diagnosis. Consequently, the segmentation has required being precise, robust, and useful to prevent the collisions happened by different huge and compound biases added to MRI images. For the brain tumour segmentation, the clustering methods were applied mostly. Kmeans clustering and Fuzzy C-means clustering methods to find the tumour in MRI image and to extract it from the given image. Comparative analysis has been performed with help of Relative area, Peak Signal to Noise Ratio (PSNR), Segmented area, Mean Squared Error (MSE) was executed between K-means clustering and FCM clustering methods. The experimental results of this method have been demonstrated that the FCM algorithm efficiency was selected over the K-means algorithm.

        NOVEL FUZZY METHOD FOR THE AUTOMATIC SEGMENTATION OF STANDARD AND VOLUMETRIC DATASETS OF PATHOLOGICAL BRAIN MRI [3]

        Author: El-Melegy, M.T, and Mokhtar, H.M This method has reformulated the

        well-liked FCM algorithm to get into some obtainable data about the class centre of image. In this data, the uncertainty has also been modelled. This kind of data has worked to normalize the clusters that were made by the FCM algorithm therefore boosting its presentation under unpredicted data acquisition and noisy situation. Furthermore, the convergence process of this method is enhanced. Experimental of this method was simulated and actual, both standard and pathological, MRI volumes of the human brain has demonstrated that this presented method has significant improved segmentation accurateness, strength against noise situation, and quicker reply than the different well-known fuzzy and non-fuzzy method.

        EFFORTLESS CASES LIKE FITTING A CIRCLE, LINE, CUBIC SPLINE CONTOUR OR ELLIPSE BY UTILIZING THIS SEGMENTATION METHOD [4]

        Authors: Wang, Q and Boyer, K.L

        In phase-contrast magnetic resonance (PC- MR) images, segmentation method has been utilized to identify subarachnoid spaces the cross-section shaving the cerebrospinal fluid (CSF) wherever the object of interest was defined by applying the distorted ellipse. In this method, the discovery results were employed by an st graph cut to produce a CSF structures segmentation. This method has illustrated that specified a correctly configured arithmetical contour method and force area. This method can be robust to remove the noise and faults in the MRI image. This method does not based on huge training datasets by exploiting an arithmetical contour method and physical labelling of the training images does

        not needed as can be required while applying the point sharing methods.

        AUTOMATIC TUMOR SEGMENTATION TECHNIQUE

        Author: Meiyan Huang

        For MRI brain images, this kind of technique has treated tumor segmentation very efficiently. In addition, the local independent projection-based classification (LIPC) technique has been employed to categorize every voxel into several classes. This classification structure was derived by using the LIP into the traditional classification approach. For LIPC, the locality was significant in the computation of local independent projections. Locality has also been measured in demonstrating whether local fix embedding is more appropriate in resolving the linear projection weights than the other coding techniques. Furthermore, LIPC has measured the data sharing of several classes by examining a SoftMax regression method that has developed the performance of classification.

  2. SYSTEM REQUIREMENT Anaconda is that the installation program utilized

    by Fedora, Red Hat Enterprise Linux and a few other distributions. During installation, a target computers hardware is identified and configured and also the appropriate file systems for the systems architecture are created. Finally, anaconda allows the user to put in the software software on the target computer. Anaconda also can upgrade existing installations of earlier versions of the identical distribution. After the installation is complete, you'll beable to reboot into your installed system and continue doing customization using the initial setup program. Anaconda could be a fairly sophisticated installer. It supports installation from local and remote sources like CDs and DVDs, images stored on a tough drive, NFS, HTTP, and FTP. Installation may be scripted with starting motor to produce a totally unattended installation that may be duplicated on various machines. It also can be run over VNC on headless machines. a spread of advanced storage devices including LVM, RAID, iSCSI, and multipath are supported from the partitioning program. Anaconda provides advanced debugging features like remote logging, access to the python interactive debugger, and remote saving of exception dumps.

    TensorFlow is a multipurpose open source software library for numerical computation using data flow graphs. It has been designed with deep learning in mind but it is applicable to a much wider range of problems. TensorFlow can be used from many programming languages.

    TensorBoard may be a utility to visualise different aspects of machine learning. the subsequent guides explain a way to use TensorBoard. TensorBoard Visualizing Learning, which introduces TensorBoard. TensorBoard Histogram Dashboard which demonstrates a way to use TensorBoard's histogram dashboard.

    Performance is a very important consideration when training machine learning models. Performance hurries up and scales research while also providing end users with near instant predictions. Performance overview contains a group of best practices for optimizing your TensorFlow code.

  3. WORK FLOW

    Fig. 1. Sensing Brain Tumor

  4. MODULE DESCRIPTION

        1. Deep Learning

          Deep learning is that the role of machine learning. it's supported artificial neural networks. this can be because the neural network can replicate the functions of the human brain. In deep learning, there's no must program all the data explicitly. Deep learning architectures like deep neural networks, deep belief networks, recurrent neural networks and convolutional neural networks are applied to fields including computer vision, speech recognition, linguistic communication processing, audio recognition, social network filtering, AI, bioinformatics, drug design, medical image analysis, material inspection and parlour game programs, where they need produced results corresponding to and in some cases surpassing human expert performance.

        2. Wavelet Transformation

          When the frequencies of signals vary, a mathematical calculation for analyzing the signals are termed wavelet transformation. Wavelet analysis provides more accurate information about signals and pictures, compared to other signal analysis techniques. A wavelet could be a wave-like oscillation with an amplitude that begins at zero, increases, so decreases back to zero. It can typically be visualized as a "brief oscillation" like one recorded by a seismograph or cardiac monitor. Generally, wavelets are intentionally crafted to own specific properties that make them useful for signal processing. employing a "reverse, shift, multiply and integrate" technique called convolution, wavelets is combined with known portions of a damaged signal to extract information from the unknownportions.

        3. EGG Signal

          The EEG (Electroencephalogram) signal could be a test. And it's wont to evaluate the electrical activity within the brain. Cells within the brain interact with one another through electrical impulses. The EEG signal is employed to detect a related problem during thi process. The EEG signal pathways and records the wave patterns within

          the brain. Electroencephalography (EEG) is an electrophysiological monitoring method to record electrical activity of the brain. it's typically noninvasively, with the electrodes placed along the scalp, although invasive electrodes are sometimes used, as in electrocorticography. Clinically, EEG refers to the recording of the brain's spontaneous electrical activity over a period of your time, as recorded from multiple electrodes placed on the scalp. Diagnostic applications generally focus either on event-related potentials or on the spectral content of EEG. The latter analyses the kind of neural oscillations (popularly called "brain waves") that may be observed in EEG signals within the frequency domain.

          Fig. 2. EGG Signal

        4. Berkeley Wavelet Transformation

          Berkeley wavelet transformation uses a two- dimensional triangular bandwidth. And in terms of complete orthogonal, this can be the simplest method to spot the affected tumour area from the brain image. So, during this paper, we used the Berkeley wavelet transformation method for brain tumor division. it's accustomed analyze or process image or signals. The BWT shares many characteristics with the receptive fields of neurons in mammalian primary cortical region (V1). Like these receptive fields, BWT wavelets are localized in space, tuned in spatial frequency and orientation, and form a collection that's approximately scale invariant. The wavelets even have spatial frequency and orientation bandwidths that are comparable biological values. Although the classical Gabor wavelet model may be a more accurate description of the receptive fields of individual V1 neurons, the BWT has some interesting advantages. it's a whole, orthonormal basis and is therefore inexpensive to compute, manipulate, and invert. These properties make the BWT useful in situations where computational power or experimental data are limited, like estimation of the spatiotemporal receptive fields of neurons.

        5. K-Means Clustering

          K-means clustering could be a method of vector quantization, originally from signal processing, that's popular for cluster analysis in data processing. k-means clustering aims to partition n observations into k clusters during which each observation belongs to the cluster with the closest mean, serving as a prototype of the cluster. This ends up in a partitioning of the info space into Voronoi cells. k-Means minimizes within-cluster variances (squared Euclidean distances), but not regular

          Euclidean distances, which might be the harder Weber problem: the mean optimizes squared errors, whereas only the geometric median minimizes Euclidean distances. Better Euclidean solutions can for instance be found using k-medians and k-medoids. The problem is computationally difficult (NP-hard) however, efficient heuristic algorithms converge quickly to a neighbourhood optimum. These are usually kind of like the expectation- maximization algorithm for mixtures of Gaussian distributions via an iterative refinement approach employed by both k-means and Gaussian mixture modelling. They both use cluster centres to model the data; however, k- means clustering tends to search out clusters of comparable spatial extent, while the expectation-maximization mechanism allows clusters to possess different shapes. The algorithm includes a loose relationship to the k- nearest neighbour classifier, a well- liked machine learning technique for classification that's often confused with k- means because of the name. Applying the 1-nearest neighbour classifier to the cluster centres obtained by k- means classifies new data into the present clusters. this is often called nearest centroid classifier or Rocchio algorithm.

        6. Kernal Support Vector Machine

    Support Vector Machines (SVMs, also Support Vector Network) are supervised learning models with associated learning algorithms that analyze data used for classification and multivariate analysis. Given a group of coaching examples, each marked as belonging to at least one or the opposite of two categories, an SVM training algorithm builds a model that assigns new examples to at least one category or the opposite, making it a non- probabilistic binary linear classifier (although methods like Platt scaling exist to use SVM in an exceedingly probabilistic classification setting). An SVM model may be a representation of the examples as points in space, mapped in order that the samples of the separate categories are divided by a transparent gap that's as wide as possible. New examples are then mapped into that very same space and predicted to belong to a category supported the side of the gap on which they fall.

    Fig. 3. The comparison between Level and Normal filters

    Fig. 4. Comparision of Normal wavesignal and EEG wavesignal

  5. CONCLUSION

Local binary model algorithm and convolutional neural network algorithm are used to image and extract features of tumour CT images in medical field. The local binary mode is an image recognition method based on image-based translational rotation invariance. The paper also introduces two models of convolutional neural networks. For the same data samples and sample sizes, the convolutional neural network model has a recognition rate of 99.7% for medical images, which is provided for the follow-up development expert diagnosis system and self- checking system, which is Strong technical support. The algorithm and model described in this paper, as an algorithm for continuous optimization, use the extended data set to train a new tumour image-sensitive deep convolutional neural network, and strive to add other different types of cost factors to the classification process to continue thealgorithm.

REFERENCES

  1. Demirhan (2015), Novel tissue segmentation method that has segmented brain MR images into tumor, WM, GM, and CSF and edema.

  2. D. L. Schwartz., FDG-PET prediction of head and neck squamous cell cancer outcomes,'' Arch. Otolaryngol. Head Neck Surg., vol. 130, no. 12, pp. 1361_1367, Dec. 2004.

  3. B. Ganeshan, K. Skogen, I. Pressney, D. Coutroubis, and K. Miles, Tumour heterogeneity in oesophageal cancer assessed by CT texture analysis: Preliminary evidence of an association with tumour metabolism, stage, and survival,'' Clin. Radiol., vol. 67, no. 2, pp. 157_164, Feb. 2011.

  4. Y. Sun, X. Wang, and X. Tang, Deep learning face representation from predicting 10,000 classes,'' in Proc. IEEE Conf. Comput. Vis. Pattern Recognit., Sep. 2014, pp. 1891_1898.

  5. L. Chen, S. Wang, W. Fan, J. Sun, andS. Naoi, Beyond human recognition: A CNN-based framework for handwritten character recognition,'' in Proc. IEEE Asian Conf. Pattern Recognit. (IAPR), Nov. 2015, pp. 695_699.

  6. Y. Y. Liu, M. Chen, H. Ishikawa, G. Wollstein, J. S. Schuman, and

    J. M. Rehg, Automated macularpathology diagnosis in retinal OCT images using multi-scale spatial pyramid and local binary patterns in texture and shape encoding,'' Med. Image Anal., vol. 15, no. 5, pp. 748_759, Oct. 2011.

  7. D. Sidibé ., An anomaly detection approach for the identi_cation of DME patients using spectral domain optical coherence tomography images,'' Comput. Methods Programs Biomed., vol. 139, pp. 109_117, Feb. 2017.

  8. C. S. Lee, A. J. Tyring, Y. Wu, A. Rokem, A. Y. Lee, and N. P. Deruyter, Deep-learning based, automated segmentation of macular edema in optical coherence tomography,'' Biomed. Opt. Express, vol. 8, no. 7, pp. 3440_3448, Jun. 2017.

  9. D. S.Kermany ., Identifying medical diagnoses and treatable diseases by image-based deep learning,'' Cell, vol. 172, no. 5, pp. 1122_1131, Jun. 2018.

  10. Z. Ji and L. H. Nie, Texture image classi_cation with noise- tolerant local binary pattern,'' J. Comput. Res. Develop., vol. 53, no. 5, pp 1128_1135, May 2016.

  11. K. Zhou, L. M. Yang, and H. Song, Face recognition method based on dynamic threshold local binary pattern,'' Comput. Eng., vol. 35, no. 17, pp. 167_169, Sep. 2009.

  12. A. Yang, S. Li, H. Lin, and D. Jin, Edge extraction of mineralogical phase based on fractal theory,'' Chaos, Solitons Fractals, vol. 117, pp. 215_221, Nov. 2018.

  13. H.-C. Shin ., Deep convolutional neural networks for computer- aided detection: CNN architectures, dataset characteristics and transfer learning,'' IEEE Trans. Med. Imag., vol. 35, no. 5, pp. 1285_1298, May 2016.

  14. K. Zhang, W. Zuo, Y. Chen, D. Meng, and L. Zhang, Beyond a Gaussian Denoiser: Residual learning of deep CNN for image denoising,'' IEEE Trans. Image Process., vol. 26, no. 7, pp. 3142_3155, Jul. 2017.

  15. Y. Wu, S.-Y. Wang, and Y.-B. Hou, A study on spectral super- resolution of hyperspectral imagery based on redundant dictionary,'' Inf. Eng., vol. 3, no. 5, pp. 67_71, Oct. 2014.

  16. S. Li, R. Fan, G. Yue, C. Hou, and G. Lei, A two-channel convolutional neural network for image super- resolution,'' Neurocomputing, vol. 275, no. 31, pp. 267_277, Jan. 2018.

  17. T.-Y. Lin, A. R. Chowdhury, and S. Maji, Bilinear CNN models for _ne- grained visual recognition,'' in Proc. IEEE Int. Conf. Comput. Vis., Apr. 2015, pp. 1449_1457.

  18. M. Simon and E. Rodner, Neural activation constellations: Unsupervised part model discovery with convolutional networks,'' in Proc. 15th IEEE Int. Conf. Comput. Vis. (ICCV). vol. 3, Apr. 2015, pp. 1143_1151.

  19. J. Liu, M. Li, J.Wang, F.Wu, T. Liu, and Y. Pan, A survey of MRI-based brain tumour segmentation methods,'' Tsinghua Sci. Technol., vol. 19, no. 6, pp. 578_595, Dec. 2014.

  20. Y. J. Huang and Q. J. Feng, Segmentation of brain tumour on magnetic resonance. Images using 3D full-convolutional densely connect,''J. Southern Med. Univ., vol. 38, no. 6, pp. 661_668, Jun. 2018.

Leave a Reply

Your email address will not be published. Required fields are marked *