Intelligent Segmentation of Breast MR Images
Master of Engineering
Department of EEE
PSG College of Technology Coimbatore, India
Assistant Professor (Sr.Gr)
Department of EEE
PSG College of Technology Coimbatore, India
Abstract – Breast cancer is a major cause of mortality in young women in the developing countries. Early diagnosis is the key to improve survival rate in cancer patients. Cancer is one of the most leading causes of deaths among the women in the world. Among the cancer diseases, breast cancer is especially a concern in women. MRI is one of the methods to find tumors in the breast, which is helpful for the doctor to detect the cancer. Doctor can miss the abnormality due to inexperience’s in the field of cancer detection. Segmentation is very valuable for doctor to analyze the data in the MR Images. Accuracy rate of breast cancer in MR Images depends on the image segmentation. In this project Darwinian Particle Swarm Optimization (DPSO) method is used for clustering, Grey-level co-occurrence matrix (GLCM) technique for feature extraction and neural network technique for classification is being used. In order to improve the efficiency of the searching process clustering techniques are recommended.
Keywords— Magnetic Resonance Images (MRI); Image segmentation; clustering; Darwinian Particle Swarm Optimization (DPSO); feature extraction; Grey-level co-occurrence matrix (GLCM); classification; neural network technique.
Breast cancer is the most common type of cancer in women in the world. Early detection, intervention, and postoperative treatment, decreases the breast cancer mortality. MR Images is the ideal screening assessment for breast cancer. It is extensively available, well-tolerated and inexpensive. Processing MR Images accounts for the greatest contribution to early detection and decrease in breast cancer mortality, although its use has resulted in a minor increase in the number of in-situ cancers detected. Breast tumor detection requires high-resolution breast MR Images. The image and resolution produced by MRI is quite detailed and can detect tiny changes of structures within the body. The obtained MRI is pre-processed at the first stage. In the pre-processing stage the image is converted to grey scale image, resized, enhanced and filtered for commonly encountered noise. Weiner filter and Gaussian filters are used to effectively remove the noise component. Followed by pre-processing clustering is performed on the pre-processed image to obtain the segmented image. The consequence of image segmentation is an assortment of segments which combine to form the complete image 1. In this paper Darwinian Particle Swarm Optimization (DPSO) method is used for clustering. The segmented image is used to extract the features utilizing the Grey-level co-occurrence matrix (GLCM) technique. The final stage of classification is performed by applying neural network technique.
Figure 1: Flow Chart of the Entire Procedure
II. IMAGE ACQUISITION
Detection of breast tumors requires high-resolution breast MRI (Magnetic Resonance Image). Most Medical Imaging Studies, analysis and detections are conducted using MRI, Positron Emission Tomography (PET) and Computer Tomography (CT) Scan. The MR Image has a multidimensional nature of data provided from different sequential pulses. The MRI scan is a radiology technique that utilizes magnetism, radio waves, and a computer to generate multiple images of any body structure 1. The MRI scanner is a tube enclosed by a giant circular magnet. The patient is positioned on a moveable bed that is inserted into the magnet. The magnet generates a strong magnetic field that aligns the protons of hydrogen atoms, which are then exposed to a beam of radio waves. This spins the various protons of the body, and they produce a faint signal that is detected by the receiver portion of the MRI scanner. The receiver information is processed by a computer, and an image is produced. The image produced by MRI is quite detailed with high resolution and can detect tiny changes of structures within the body. For some procedures, contrast agents, such as gadolinium are used to increase the accuracy of the images. MRI scanners can produce 1500 images per second. Images of a patient obtained by MRI scanning are displayed as an array of pixels (a two dimensional unit based on the matrix size and the field of view) and stored in memory. The 0.5T intra-operative Magnetic Resonance Scanner is used to acquire 256*256*58(0.86mm, 0.86mm, 2.5 mm) T1 weighted images with the fast spin echo protocol (TR=400, TE=16 ms, FOV=220*220 mm) in 3 minutes and 40 seconds.
Pre-processing of an image is the process of preparing the input image to commence it to an algorithm for particular task. The aim of pre-processing is to develop the image data that removes unwanted distortions or holds back the image features which are important for further processing. Pre-processing techniques also known as radiometric or geometric corrections includes those operations that are normally required prior to the main data analysis and extraction of information. It comprises of correcting the data for sensor irregularities and unwanted sensor or atmospheric noise, removal of non-breast tissues. Image noise is most apparent in image regions with low signal level such as shadow regions or under exposed images. Medical images in no doubt will contain some Film artifacts like unwanted or critical parts, labels and marks which are spoted and removed using Tracking Algorithm. Here, initiated from the first row and first column of the input image, the intensity value of the pixels is measured and the threshold value of the film artifacts is found. The pixel with intensity value greater than that of the threshold value is removed from the image.
Step 1: The MRI image is read and stored in form of a two dimensional matrix.
Step 2: The peak threshold value is selected.
Step 3: The flag value is set to 255.
Step 4: The pixels whose intensity value is equal to 255 is selected.
Step 5: If the intensity value is 255, the flag value is set to zero. As a consequence the labels are removed from MRI image.
Step 6: if not pass over to the next pixel
IV. RESIZING THE IMAGE
Image interpolation occurs when the image is resized or distorted from one pixel grid to another. Image resizing is necessary when the total number of pixels is to be increased or decreased, whereas remapping can occur when correcting the lens distortion or rotating an image. Zooming refers to increase the quantity of pixels, so that when an image is zoomed, more details can be seen. Interpolation works by using known data to estimate values at unknown points. Image interpolation works in two directions, and tries to achieve a best approximation of a pixel’s intensity based on the values at surrounding pixels.
Normalization is a process that alters the range of pixel intensity values. Normalization is also known as contrast stretching or histogram stretching. In more universal fields of data processing, such as digital signal processing, it is referred to as dynamic range expansion. The purpose of dynamic range expansion in the various applications is usually to bring the image, or other type of signal, into a range that is more common or normal to the senses.
VI. WIENER FILTERING
The inverse filtering is a restoration technique for deconvolution, i.e., when the image is blurred by a known low-pass filter, it is possible to recover the image by inverse filtering or generalized inverse filtering. However, inverse filtering is very sensitive to additive noise. The approach of reducing one degradation at a time allows us to develop a restoration algorithm for each type of degradation and simply combine them. The Wiener filtering executes an optimal tradeoff between inverse filtering and noise smoothing. It removes the additive noise and inverts the blurring simultaneously.
The Wiener filtering is optimal in terms of the mean square error. In other words, it minimizes the overall mean square error in the process of inverse filtering and noise smoothing. The Wiener filtering is a linear estimation of the original image. The approach is based on a stochastic framework. The orthogonality principle implies that the Wiener filter in Fourier domain can be expressed as follows:
Where, Sxxf1,f2, S??(f1,f2) are respectively power spectra of the original image and the additive noise, and H(f1,f2) is the blurring filter. It is easy to see that the Wiener filter has two separate parts, an inverse filtering part and a noise smoothing part. It not only performs the deconvolution by inverse filtering (highpass filtering) but also removes the noise with a compression operation (lowpass filtering).
To implement the Wiener filter in practice we have to estimate the power spectra of the original image and the additive noise. For white additive noise the power spectrum is equal to the variance of the noise. To estimate the power spectrum of the original image many methods can be used. A direct estimate is the periodogram estimate of the power spectrum computed from the observation:
Where, Y(k,l) is the DFT of the observation. The advantage of the estimate is that it can be implemented very easily without worrying about the singularity of the inverse filtering. Another estimate which leads to a cascade implementation of the inverse filtering and the noise smoothing is,
eqn(3), is a straightforward result of the fact: Sxx|H|2+S??=SyyThe power spectrum can be estimated directly from the observation using the periodogram estimate. This estimate results in a cascade implementation of inverse filtering and noise smoothing,
VI. GAUSSIAN FILTERING
In image processing, Gaussian smoothing is the result of blurring an image by a Gaussian function. It is a widely used effect in graphics software, typically to reduce image noise and reduce detail. The visual effect of this blurring technique is a smooth blur resembling that of viewing the image through a translucent screen, distinctly different from the bokeh effect produced by an out-of-focus lens or the shadow of an object under usual illumination. Gaussian smoothing is also used as a pre-processing stage in computer vision algorithms in order to enhance image structures at different scales.
Mathematically, applying a Gaussian blur to an image is the same as convolving the image with a Gaussian function. This is also known as a two-dimensional Weierstrass transform. By contrast, convolving by a circle (i.e., a circular box blur) would more accurately reproduce the bokeh effect. Since the Fourier transform of a Gaussian is another Gaussian, applying a Gaussian blur has the effect of reducing the image’s high-frequency components; a Gaussian blur is thus a low pass filter. The Gaussian blur is a type of image-blurring filter that uses a Gaussian function (which also expresses the normal distribution in statistics) for calculating the transformation to apply to each pixel in the image. The equation of a Gaussian function in one dimension is,
In two dimensions, it is the product of two such Gaussians, one in each dimension,
Where x is the distance from the origin in the horizontal axis, y is the distance from the origin in the vertical axis, and ? is the standard deviation of the Gaussian distribution. When applied in two dimensions, this formula produces a surface whose contours are concentric circles with a Gaussian distribution from the center point. Values from this distribution are used to build a convolution matrix which is applied to the original image. This convolution process is illustrated visually in the figure on the right. Each pixel’s new value is set to a weighted average of that pixel’s neighborhood. The original pixel’s value receives the heaviest weight (having the highest Gaussian value) and neighboring pixels receive smaller weights as their distance to the original pixel increases. This results in a blur that preserves boundaries and edges better than other, more uniform blurring filters; see also scale space implementation. In theory, the Gaussian function at every point on the image will be non-zero, meaning that the entire image would need to be included in the calculations for each pixel. In practice, when computing a discrete approximation of the Gaussian function, pixels at a distance of more than 3? are small enough to be considered effectively zero. Thus contributions from pixels outside that range can be ignored. Typically, an image processing program need only calculate a matrix with dimensions to ensure a result sufficiently close to that obtained by the entire Gaussian distribution.
6? x 6? (7)
VII, CLUSTERING BY DARWINIAN PARTICLE SWARM OPTIMIZATION (DPSO)
The main purpose of clustering is to divide a set of objects into significant Groups. The clustering of objects is based on measuring of correspondence between the pair of objects using distance function. Thus, result of clustering is a set of clusters, where object within one cluster is further similar to each other, than to object in another cluster. The Cluster analysis has been broadly used in numerous applications, including segmentation of medical images, pattern recognition, data analysis, and image processing. Clustering is also called data segmentation in some applications because clustering partitions huge data sets into groups according to their resemblance.
A particle swarm is a population of particles, in which each particle is a moving object which can move through the search space and can be attracted to the better positions. PSO must have a fitness evaluation function to decide the better and best positions, the function can take the particle’s position and assigns it a fitness value. Then the objective is to optimize the fitness function. In general, the fitness function is pre-defined and is depend on the problem.
Particle Swarm Optimization (PSO), an evolutionary algorithm for optimization is extended to determine if natural selection, or survival-of-the fittest, can enhance the ability of the PSO algorithm to escape from local optima. To simulate selection, many simultaneous, parallel PSO algorithms, each one a swarm, operate on a test problem. Simple rules are developed to implement selection. The ability of this so-called Darwinian PSO to escape local optima is evaluated by comparing a single swarm and a similar set of swarms, differing primarily in the absence of the selection mechanism, operating on the same test problem. The selection process is shown to be capable of evolving the best type of particle velocity control, which is a problem specific design choice of the PSO algorithm.
Figure 2: Flowchart of the Darwinian Particle Swarm Optimization Algorithm.
In a typical implementation of PSO, a single swarm of test solutions is utilized. To implement natural selection with a single swarm, the algorithm must detect when stagnation has occurred. Since a single swarm is unable to differentiate between a global optimum and a local optimum it cannot simply be extended to model natural selection. One could “time-out” the optimization and restart the algorithm4 or delete information about the current global optimum in hopes that the swarm will not return to it. At the end of each swarm update, the current fitnesses of the particles are used to order the particles. The top half of the particles are then duplicated and replace the positions and velocities of the bottom half of the particles. The personal bests of the particles are not changed. Each swarm individually performs just like an ordinary PSO algorithm with some rules governing the collection of swarms that are designed to simulate natural selection. The selection process implemented is a selection of swarms within a constantly changing collection of swarms.
VII. FEATURE EXTRACTION BY GRAY-LEVEL CO-OCCURRENCE MATRIX (GLCM)
The segmented breast image is used for feature extraction. A feature is significant piece of information extracted from an image which provides the more detailed understanding of an image. In this paper Gray Level Co-occurrence Matrix is used for feature extraction. A statistical method of examining texture that considers the spatial relationship of pixels is the gray-level co-occurrence matrix (GLCM), also known as the gray-level spatial dependence matrix. The GLCM functions characterize the texture of an image by calculating how often pairs of pixel with specific values and in a specified spatial relationship occur in an image, creating a GLCM, and then extracting statistical measures from this matrix. (The texture filter functions, described in Texture Analysis cannot provide information about shape, that is, the spatial relationships of pixels in an image.)With the help of GLCM we take following features from the image:
IX. CLASSIFICATION BY NEURAL NETWORKS
Classification is one of the prominent step in detection of breast tumors. The extracted features are considered as input to the classifier to classify the detected suspicious areas into normal, benign or malignant. Classifier such as neural network (NN) has performed well. The classification of breast cancer detection is divided into the training phase and the testing phase. During training, the features are extracted from the segmented images are input to NN whose weights are optimized by particle swarm optimization in which the diagnosis is known. Whenever an image is taken as input to the algorithm, it is simulated with the trained networks and goes for testing the image.
Neural Network (NN) is a technique stimulated from functioning of the biological nervous systems to process information. An NN consists of a collection of processing elements that are highly interconnected and transform a set of inputs to a set of desired outputs in which each connection has a weight associated with it. The advantage of NN is their capability of self-learning, and often suitable to solve the problems that are too complex to use the conventional techniques, or hard to find algorithmic solutions. The neural network trained by adjusting the weights so as to be able to predict the correct class. It gives the output as, the given input image is normal or abnormal.
Automated classifiers could substantially upgrade the diagnosis process, in terms of both accuracy and time requirement by distinguishing benign and malignant patterns automatically. Neural network (NN) serves as an automated classifier. Hybrid NN is the most efficient technique in breast cancer detection. In medical image processing, NNs have been applied to a variety of data-classification and pattern recognition tasks and become a promising classification tool in breast cancer. NN applications in ultrasound, and MRI and IR imaging for early detection of breast cancer .Image features can be distinguished in many aspects, such as texture, colour, shape, and spatial relations. They can reflect the subtle variance in many degrees.
The neural network resembles the function of the biological neuron, and it is composed of neurons with different layers and these neurons are interconnected by numeric weights; these weights can be changed due to the learning behavior of the network to approach the optimum result. Usually in image processing applications, the number of the neurons is directly related to the number of pixels in the input image, and the number of layers depends on the processing steps. For cancer detection and classification, image segmentation has been widely used. Many image segmentation methods, based on histogram features, edge detection, region growing, or pixel classification, have been trained using NNs. Many neural networks models are utilized to aid MRI for enhancing the detection and the classification of the breast tumors, which can be trained with previous cases that are diagnosed by the clinicians correctly, or can manipulate the signal intensity or the mass characteristics (margins, shape, size, and granularity), Multistate cellular Neural Networks have been used in MR image segmentation to estimate the density of the breast regions for evaluation of the fat contents.
Dimensions of Neural Network in the proposed work:
Figure 3: Pre-Processed Output of the Abnormal Image
Figure 4: Clustered Output of the Abnormal Image
Figure 5: Segmented Output of the Abnormal Image
Figure 6: Classified Result
Figure 7: Imput of the Normal Image
Figure 8: Resized Image
Figure 9: Normalised Image
Figure 10: Filtered Image
Figure 11: Clustered Output of the Normal Image
Figure 12: Segmented Output of the Normal Image
Figure 13: Classifier Result
In this project image processing techniques were used on the MR images. The proposed approach detects breast cancer of medical image MRI. Then segmentation technique applied is based on DPSO swarm optimization. The feature values are extracted with gray level co-occurrence matrix. The normal and cancerous image classified based on neural network. The proposed algorithm has been tested on standard images
C.Harris and M.Stephens, “A Combined Corner and Edge Detection,” Proc.Fourth Alvey Vision Conf., pp.147-151, 1988.
Shirakawa, S., and Nagao, T., “Evolutionary Image Segmentation Based on Multiobjective Clustering”. Congress on Evolutionary Computation (CEC ’09), Trondheim, Norway, 2466-2473, 2009.
K. S. Chuang., H. L. Tzeng., S. Chen., J. Wu., and T. J. Chen., “Fuzzy C-Means Clustering with Spatial Information for image Segmentation,” Comput. Med. Imaging Graph, vol. 30, no. 1, pp. 9–15, Jan. 2006.
R. Szeliski., D. Tonnesen., and D. Terzopoulos., “Modeling Surfaces of Arbitrary Topology with Dynamic Particles”, In: Proceedings of CVPR, pp. 82– 87, 1999.
Puzicha, J., Hofmann, T. and Buhmann, J. M.,”Histogram Clustering for Unsupervised Image Segmentation”., Computer Vision and Pattern Recognition , Vol.2.IEEEpress,602-608, 2000.
Ahmedi, J., Rebecca, S., Taylor, M., Jiaquan, X., Ward, E., and Thun, M.J., “Global cancer statistics,” Cancer Journal for Clinicians, 2011, pp.69-90
Hong, S., Xiangfei, C., and Feifei, S., “Breast Tissue 3D Segmentation and Visualization on MRI”. International Journal of Biomedical Imaging, 2013, pp 8.
Ali, Q., Umi, K., Nor A., and Ibrahim, L., “Breast MRI Tumour Segmentation using Modified Automatic Seeded Region Growing Based on Particle Swarm Optimization Image Clustering”. Adfa, 2011, pp 1-11.
K Krajsek, R Mester, “The edge preserving wiener filter for scalar and tensor valued image”, DAGM, pp. 91-100,2006.
Alexandra Flowers MD, “Brain Tumors in t he Older Person “,CancerControl, Volume 7, No.6, pages 523- 538,November/December2000.
André Collignon,Dirk Vandermeulen,Paul Suetens, Guy Marchal,” 3D multi-modality medical image registration using feature space clustering “,SpringerLink, Volume 905/1995, pages 193-204 Berlin 1995.
Alexis Roche, Gregoire Malandain, Nicholas Ayache, Sylvain prima, “Towards a better comprehension Medical Image Registration”, Medical Image Computing and Computer-Assisted Intervention-MICCAI’99,Volume 1679,pages 555-566, 1999.
A Buades, B Coll, JM Morel, “A non-local algorithm for image denoising”, CVPR, vol. 2, pp. 60-65, 2005.
J.Jaya , K.Thanushkodi , M.Karnan, “Tracking Algorithm for De-Noising of MR Brain Images” IJCSNS International Journal of Computer Science and Network Security, VOL.9 No.11, pp.262-267, 2009.
K. H. Ting, P. C. W. Fung, C. Q. Chang, F. H. Y. Chan, “Automatic correction of artifact from single-trial event-related potentials by blind source separation using second order statistics only”, Med. Eng. Phys., vol. 28, no. 8, pp. 780-794, 2006.
Raman Maini, Himanshu Aggarwal, “A Comprehensive Review of Image Enhancement Techniques”, JOURNAL OF COMPUTING, VOLUME 2, ISSUE 3,pp. 8-14, 2010.
R.W.Jr. Weeks,” Fundamental of Electronic Image Processing ” Bellingham: SPIE Press, 1996.