Automatic Segmentation and Edge Detection in MRI Scan for Brain Tumor Classification and Evaluation

I. INTRODUCTION

The brain is a delicate, fragile, non-repliable mass of tissue. Considered as a kernel component of the body, the brain has a quite complicated architecture. Apart from cerebrospinal fluid (CSF) consisting of enzymes, glucose, salts, and white blood cells, the brain consists of two types of tissues: gray matter (GM) and white matter (WM). There are three main divisions of the brain: frontbrain, midbrain, rear brain. Many diverse, specialized individual cells make up the body. Most of the body's cells grow and split to generate fresh cells of the same type required for the human body to run as designed. These cells produce a lot of unwanted tissue when they veer off course and proliferate quickly, which forms a tumor [3].

A brain tumor results from a gathering of aberrant cells either inside or outside of the brain. Apart from directly destroying healthy brain cells, tumors induce inflammation, brain swelling, and pressure inside the skull [1]. Originally Latin, the word "tumor" describes swelling. The degree and location of the tumor will determine the symptoms; headaches, nausea (typically in the morning), personality changes, irritability, drowsiness, sadness, declining cardiac and respiratory functions, and finally coma if treated [7].

Three forms define brain tumors: benign, pre-malignant, and malignant. Benign tumors neither fast expansion nor influence other healthy cerebral tissues. Pre-cancerous or premalignant tumors, however, may cause malignancies [6] if improperly treated. More severe and fast spreading than benign tumors, malignant ones sometimes cause patient death. Among the most threatening malignant tumors are gliomas, particularly glioblastoma multiforme (GBM) [2]. GBM is well-known for its aggressiveness and bad outlook [4].

The American Brain Tumor Association (ABTA) projects 62,930 new primary brain tumor cases in 2010; predictions for 2030 range to 26 million new cases and a death toll exceeds 1.8 million persons [1]. In 2005, the American Cancer Society estimated 12,600 fatalities from brain cancer as well as 18,500 fresh brain tumor diagnosis. According to projections made by the National Cancer Institute (NCI), brain and central nervous system (CNS) tumors will account for 22,070 newly identified cases [1]. According to the World Health Organization (WHO), brain tumors develop in around 120 different forms [8].

Usually shown on CT or MRI scans as distinctively colored masses, neoplasms are Radiologists can physically check a patient using MRI and CT scans; MRI is more useful for its lack of radiation and capacity to create images in several planes [6]. MRI pictures of brain architecture, tumor location, and size enable radiologists identify tumors and guide surgical excision [5].

Current traditional diagnostic techniques run a great risk of erroneous brain tumor identification and detection since they mostly depend on human knowledge in MRI image interpretation. Faster and more accurate tumor identification is provided by digital image processing incorporating segmentation methods. Segmentation separates an image into areas homogeneous in space concerning a predefined criterion [3]. Particularly for the diagnosis of brain tumors, this approach has lately attracted interest because of its efficiency in extracting information from difficult medical images [2][4].

Because their efficiency in locating objects and borders inside pictures, the segmentation of brain tumors in MR images has attracted a lot of interest in automated medical diagnosis. Depending on their features, brain tumors could show as hyperintense, isointense, or hypointense [7]. Brain segmentation uses two-fold threshold-based extraction and contour refining among other image processing techniques. Histogram equalization and skull boundary elimination are among pre-processing methods used to raise image quality and increase tumor detection [6][8].

II. RELATED WORK

Image Segmentation and Tumour Detection

A fundamental area of medical image processing research is the segmentation of brain tumors from MRI images. Over the years, several strategies have been suggested with different degrees of computing complexity and accuracy.

1. Threshold-Based Methods

Early approaches for brain tumor segmentation frequently drew on straightforward thresholding methods. [15] approach, for instance, was extensively applied to maximize the threshold value and thereby isolate tumor areas from the background. Although useful in some situations, these techniques usually failed with noise and fluctuation in image intensity.

2. Region-Based Methods

From seed sites to segment tumor areas, region-growing algorithms [9] extend areas. These techniques aggregate pixels with similar characteristics to improve segmentation, although they can be susceptible to image noise and initial seed location.

3. Model-Based Methods

Brain tumor segmentation has benefited from active contour models, sometimes known as snakes [12]. These models evolve a contour around the tumor boundary by means of energy minimizing strategies. They are computationally demanding and require careful parameter calibration even if they offer exact border definition.

4. Machine Learning Approaches

Support Vector machines (SVM) and Random Forests among other machine learning methods have lately been used to segment brain tumors [13]. These techniques categorize areas as tumor or non-tumor using aspects gleaned from the images. Although they show potential, for training they usually need a lot of labelled data.

5. Deep Learning Approaches

Deep learning's arrival has transformed segmentation of medical images. By learning hierarchical features from big datasets, CNNs especially U-Net [16] have shown notable advances in brain tumor segmentation. CNNs have been extensively embraced for their exceptional performance and automation powers since they shine in capturing intricate patterns.

6. Hybrid Methods

Additionally, investigated is combining conventional approaches with machine learning or deep learning methodologies. For example, hybrid techniques combining CNNs with thresholding [17] can use the advantages of both techniques, therefore offering improved segmentation accuracy.

Edge Detection Techniques

Clearly defining tumor boundaries depends on edge detection. Established and proven for best edge detecting capability is the Canny edge detector (Canny, 1986). Recent developments in edge detection have concentrated on enhancing the Canny approach by means of adaptive thresholding and noise reduction strategies [11].

1. Canny Edge Detector

Because the Canny approach effectively detects edges while reducing noise, it is still the accepted benchmark for edge detection. It runs throughout several phases: Gaussian smoothing, gradient computation, non-maximum suppression, and hysteresis thresholding.

2. Improved Edge Detection

Adaptive Gaussian filters [18] among other improvements to the Canny detector have been suggested to solve edge localization and noise management constraints. These enhancements seek to lower false positives and increase edge detection quality.

3. Edge Detection in Medical Imaging

Edge detection techniques have been developed in the framework of medical imaging to address the special difficulties of MRI scans, including picture artifacts [14] and different intensity levels. Improved accuracy in identifying tumor boundaries has been shown by enhanced edge detecting systems.

III. PROPOSED APPROACH

A. IMAGE ACQUISITION

Fig 1 shows the Architecture of the proposed system. First in my proposed method we took into account that the MRI scan images of a certain patient are either color, Gray-scale, or intensity images herein displayed with a default size of 220×220. If it is a color image, a Gray-scale converted image is described by using a big matrix whose entries are numerical values between 0 and 255, where 0 corresponds to black and 255 to white for instance. The identification of a brain tumor for a particular patient then consists in two basic phases: image segmentation and edge detection.

Fig 1: Architecture of the Proposed System

B. IMAGE SEGMENTATION

Pixel clustering’s in the image segmentation aim to create noticeable image regions. Gray level picture segmentation is applied to offer anatomical structure and identify the Region of Interest, so guiding tumor, lesion, and other anomaly detection. Based on the knowledge of anatomical structure of the healthy sections, the suggested method compares them with the contaminated sections. It then finds the aberrant sections in the sick patient brain scan image by means of reference image information.

1. Smoothing

Depending on the noise type and characteristics—that is, Gaussian noise and impulsive noise—different approaches meet different kinds of noise. blurring, another term for smoothing. One can find several explanations for smoothness. Here noise is lowered by means of smoothing. We will filter our image to execute a smoothing process. Linear filters are the most often used kind of filters; they find the value of an output pixel by weighted sum of input pixel values i.e., Called the kernel; it is nothing more than the filter's coefficients. It makes a filter seem as a window of coefficients sliding across the picture clear. Based on Gaussian filters and Smoothing filters used to either eliminate or minimise Gaussian noise from the MRI image, the proposed noise enhancing method sharpening filters grounded in the usage of first and second order derivatives for highlighting edges in an image.

2. Smoothing using Gaussian Filter

Usually the most practical filter is a Gaussian one. Convolution of every point in the input array with a Gaussian kernel then sums all of the resulting output array. If a picture is 1D, you will find that the middle pixel has the most weight. As the spatial distance between its neighbours and the center pixel rises, their weight falls. One can obtain a 2D Gaussian via Shows the variance (for each of the variables x and y) and indicates the mean, sometimes known as the peak. But compared to the linear filters, this kind of filters improved the noise reducing degree.

C. EDGE DETECTION

Calculated from the image function behaviour in a neighborhood of a pixel, an edge is a feature associated to a single pixel. Edge detection often serves to drastically cut the data count in an image while also maintaining structural integrity. In this work, various tumor kinds are suggested to be identified apart from ROI filtering. It also introduced to improve the processing time by running the features processing method in the found areas rather than the complete image frame. In this work, we first used a vector subtraction technique; subsequently, the ROI is found by identifying the relevant nearby areas in the output image from the vector subtraction. Each connected adjacent section's area is calculated; the irrelevant sections are eliminated to get the intended tumor area. We effectively used Canny's mathematical ideas to improve the proposed edge detection method's performance. Though somewhat old, it is now one of the accepted edge detection techniques and is still applied in research.

D. CANNY EDGE DETECTION

An edge detection operator using a multi-stage approach, the Canny edge detector finds a broad spectrum of edges in images. Canny also devised a computational theory of edge detection. Canny edge detection method can be dissected into five distinct phases:

1. Apply a Gaussian filter to blur the image therefore eliminating the noise.

2. Discover the image's intensity gradients here.

3. Eliminate false response to edge detection by using non-maximum suppression

4. Use twofold threshold to ascertain possible margins.

5. Track edge by hysteresis: Eliminate all the remaining weak and non-connected edges thereby completing the edge detection process.

Although conventional canny edge detection offers really straightforward solutions. The difficult edge detection job cannot be handled by the conventional method.

E. IMPROVEMENT ON CANNY EDGE DETECTION

1. Smoothing

Filtering image noise is crucial since it quickly influences all edge detection findings and helps to reduce false detection brought on by it. Blurring of the image refers to the noise removal. One convolves the image with a Gaussian filter to smooth it. Our proposed approach weights neighborhood using an original image and a 5x5 Gaussian template. This phase will gently smooth the image to minimize the impact of evident noise on the edge detector. A Gaussian filter kernel of size (2k+1) × (2k+1) has the equation:

It is crucial to realize that the performance of the detector will change with the choice of the Gaussian kernel size.

The sensitivity of the detector to noise decreases with increasing size. Furthermore, the Gaussian filter kernel size will somewhat affect the localization error to identify the edge. For most scenarios, a 5x 5 is a reasonable size; yet, this will also change depending on particular circumstances.

2. Finding gradients

Where image gradients have significant magnitudes, the edges should be noted. Better magnitude and direction value is obtained by employing a 3*3 neighborhood window instead of a 2*2 neighborhood window to estimate the gradient magnitude values and directions.

The equations are displayed as

Where g_x and g_y are the gradients in the x and y-directions respectively and respectively shows the results of the original image filtered along rows and lines. Θ is the direction of gradient.

Suggested system makes advantage of Sobel operator. Two 3x 3 convolution kernels make up the operator.

Edge orientation's angle of view

3. Non-Maximum Suppression

Edge thinning is non-maximum suppression. Applied to "thin" the edge is The edge derived from the gradient value is still somewhat blurry after gradient computation. Therefore, except from the local maximum, non-maximum suppression can aid to suppress all the gradient values to 0 by means of

a) comparison between the edge strength of the current pixel with those of the pixels in the positive and negative gradient directions.

b) The value will be kept if the edge strength of the current pixel is the highest among the other pixels in the mask pointing the same direction. The value will be inhibited otherwise.

4. Double Thresholding

Thresholding helps define possible edges. Stronger than the high threshold is indicated as edge pixels; weaker than the low threshold are suppressed and edge pixels between the two thresholds are marked as weak.

5. Edge Tracking by Hysteresis

Reducing all edges that are not related to a quite definite strong edge helps to define final edges. Included are weak edges solely in cases of connection to strong edges. Of course, noise and other minute fluctuations are unlikely to produce a strong edge. Strong edges so will only result from actual edges in the original image. Either actual edges or noise/color variances can explain the weak edges. Weak edges resulting from real edges are far more likely to be directly connected to strong edges. This work uses adaptive approach based on edge detection to find the threshold value for several photos. It split all the pixel values in the image into two groups' c0 and c1 based on an unknown threshold value T. Noted is the appropriate pixel count of intensity level i as follows. thus the probability is defined as

Where n is the total number of the pixel points in the image.

6. Source of the Tumor

Two techniques— Euclidian and lumino—allow one to determine the tumor's origin. Every pixel in the expanded area finds the closest edge pixel using the Euclidean distance norm. Euclidean distance is the most known point of reference for measuring distances. Direct and straightforward is the classifier built on this distance criterion. Using mean class values as class centers, pixel-center distances for application under the Euclidean distance rule are computed. This technique is better for main level classification of a homogeneous area. Its favourable character results from the shortest time needed to cluster or aggregate brightness values using Distance Measures. Common approach to determine proximity in space is Euclidean distance between points. Given by the Pythagorean formula, Euclidean metric is the "ordinary" distance between two points one might measure with a ruler.

Although Euclidean classifier requires much less time than other classifiers, the accuracy obtained with this approach is decent. Given that the categorization of a data sample depends on less evaluation of the decision function for every class under consideration, the findings unequivocally show that this classifier is quite fast. Since every data point in this size of the data set is classified independently, the process is unaffected by it.

A photometric measurement of the luminous intensity per unit area of light flowing in a specified direction is luminance. It shows the light's passage through, emission or reflection from a certain area, and fall within a specified solid angle. Many times, luminance is employed to describe emission or reflection from flat, diffuse surfaces. The brightness tells an eye staring at the surface from a given angle of view how much light power will be sensed. Thus, luminance serves as a guide for surface appearance brightness.

7. Age of the Tumor

The three forms of benign, pre-malignant, malignant tumors help one to identify the tumor. Benign tumors are those unable of sudden expansion influencing the other healthy brain tissues. A pre-cancerous stage, premalignant tumors could cause malignancies if not treated correctly. Many times, people assume it to be a disease. Malignant tumors spread quickly with time, and finally they cause patient death. The medical word for a severe development of a disease is malignant. One can find the tumor's age by means of the density of the tumor area.

IV. RESULT

Several slices of brain MRI scans were used for testing the suggested technique. For every slice, the area of the found tumor has been computed.

1.THE TUMOR REGION AND THE SPREAD OF TUMOR

Fig 2 shows the sequence of image processing actions followed on the tumor area of MRI images, therefore highlighting the efficiency of the suggested segmentation technique. The six pictures in the snapshot show the change of the original input image at several phases of processing.

Input Image

The first MRI scan of the brain revealed the area of the tumor before any processing. With varied brightness corresponding to different tissue types, the image offers a raw view of the tumor and adjacent brain structures.

Gaussian Filtered Image

The image following input Gaussian filter application. This stage helps to eliminate artifacts by smoothing the image and lowering high-frequency noise, therefore preparing the image for edge detection. The Gaussian filter blurs the image; this is seen in the tumor boundaries' smoother look and lower noise.

Non-Max Suppressed Image

Image non-maximum suppression is used following Gaussian filtering. This stage thinning the edges and preserves just the most noticeable ones, therefore improving the accuracy of edge recognition. The picture so emphasizes the edges more clearly, which helps to identify the tumor margins.

Strong Image

The picture displaying the outcome of a high threshold applied in the edge detection mechanism. This stage finds the image's strong edges, which most certainly represent tumor real edges. Clearly shown are the firm edges, defining the tumor's major limits.

Weak Image:

Applying a low threshold results in an image highlighting the less distinct but perhaps tumor-related weaker edges than in strong edges. This stage catches more border information necessary for a complete segmentation of the tumor.

Final Canny Image:

By means of hysteresis thresholding, the Canny edge detection technique produces a final output comprising both strong and weak edges. This picture precisely shows the form and spread of the tumor, therefore presenting its polished and whole edge map. The tumor boundaries are precisely and clearly presented by the Canny edge detector, therefore enabling additional study and diagnosis.

From first raw MRI data to a final, well-defined tumor border map, this sequence of images shows the increasing perfection of the tumor detection and segmentation procedure. Effective medical diagnosis and treatment planning depend on the accuracy and clarity of tumor identification, which depends on each stage being absolutely important.

2. THE SOURCE OF THE TUMOR AND THE AGE BY EUCLIDIAN

Fig 3 shows the result produced by Euclidian method.

3. THE SOURCE OF THE TUMOR AND THE AGE BY LUMINO

Fig 4 shows the result produced by Lumino method.

V. CONCLUSION

In this work, we report an innovative method for brain tumor detection and segmentation using enhanced automatic image segmentation approaches applied to MRI images. By means of a thorough series of image processing operations comprising Gaussian filtering, non-maximum suppression, and improved Canny edge recognition, the suggested method shows notable advances in both the accuracy and efficiency of tumor diagnosis. Our results show how well the suggested method precisely defines the tumor limits. Incorporating iterative thresholding helps the method not only detects the presence of the tumor but also evaluates its properties like age and spread. The segmentation procedure is further refined by the capacity to distinguish between strong and weak edges, therefore offering a clear and comprehensive view of the extent of the tumor. From the raw MRI image to the last Canny edge detection result, the snapshot of the processing steps shows the increasing improvement in picture quality and tumor definition. This thorough depiction emphasizes how useful the suggested approach is in separating tumor areas from surrounding tissues, therefore enabling more precise diagnosis and efficient treatment planning. Moreover, the segmentation procedure is complemented by the application of Euclidean and luminance-based approaches for tumor origin detection, therefore providing more understanding of the traits of the tumor. Apart from improving tumor diagnosis accuracy, the suggested method offers important data for evaluating tumor development and directing therapy options.

MRI-based diagnostics' dependability is much enhanced overall by the inclusion of these cutting-edge technologies into the brain tumor detecting process. This work advances medical imaging technology by providing a strong framework for automatic picture segmentation and edge detection, therefore allowing appropriate and prompt diagnosis and hence improved patient outcomes. Future research will concentrate on improving the segmentation strategies and investigating the integration of deep learning approaches to so increase the performance and adaptability of the suggested methodology over many clinical settings.

AUTHOR’S PROFILE

Karthik S is a Research Scholar in CSE department at Bharath Institute of Higher Education and Research, Chennai. He received his B.E. Degree in Computer Science and Engineering from Anna University, Chennai in the year 2005, M.Tech Degree in Computer Science and Engineering from Dr.M.G.R. Educational and Research Institute, Chennai in the year 2007. He is having more than 10 years of experience in Teaching as an Associate Professor & HOD in the Department of Computer Science and Engineering in various Engineering Colleges. His research interest includes Image Processing, Image Mining, AI and Cloud computing. He has published enormous journals and also he presented papers in many National & International Conferences. He attended many Seminars and Technical workshops. He is life Member of Indian Society of Technical Education of India (ISTE). He attended many Faculty Development Programmes.

Dr. M. Sriram is an Associate Professor in the Department of Information Technology in Bharath Institute of Higher Education and Research, Chennai. He is having More than Ten Years of Experience in Teaching. His research focuses are Data Mining, Image Processing, Data Science, Cloud Computing, AI and Big Data. He has published enormous International journals and also attended many National & International conferences, seminars and technical workshops etc.

REFERENCES

[1] American Brain Tumor Association. (2010). Facts and statistics,2010. Retrieve from http://www.abta.org/sitefiles/pdflibrary/ABTAFactsandStatist ics2010.pdf

[2] C. A. Parra, K. Iftekharuddin and R. Kozma, ―Automated Brain Tumor segmentation and pattern Recognition using ANN,‖ Computational Intelligence Robotics and Autonomous Systems, 2003

[3] V. J. Nagalkar and S. S. Asole, ―Brain tumor detection using digital image processing based on soft computing,‖ Journal of Signal and Image Processing, Vol. 3, No. 3, pp.- 102-105, 2012

[4] S. M. Bhandarkar and P. Nammalwar, ―Segmentation of Multispectral MR images Using Hierarchical Self-Organizing Map,‖ Proceedings of Computer-Based medical system CBMS 2001.

[5] C. A. Parra, K. Iftekharuddin and R. Kozma,

―Automated Brain Tumor segmentation and pattern Recognition using ANN,‖ Computational Intelligence Robotics and Autonomous Systems, 2003.

[6] S. Datta and M Chakraborty, ―Brain Tumor Detection from Pre-Processed MR Images using Segmentation Techniques,‖ CCSN, 2011.

[7] S. S. Kumar, R. S. Moniand J. Rajeesh, "Liver tumor diagnosis by gray level and contourlet coefficients texture analysis," ICCEET, Pp.557-562, 21-22 March 2012.

[8] A. Kothari ―Detection and classification of brain cancer using artificial neural network in MRI images‖, World Journal of Science and Technology, Vol. 2, No. 5, pp. 2231 – 2587, 2012.

[9] Adams, R. P., & Bischof, L. (1994). Seeded region growing. IEEE Transactions on Pattern Analysis and Machine Intelligence, 16(6), 641-647.

[10] Canny, J. (1986). A computational approach to edge detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 8(6), 679-698.

[11] Hsu, L., Chen, Y., & Lin, C. (2011). An improved Canny edge detection algorithm based on fuzzy logic. Journal of Computing, 3(7), 56-62.

[12] Kass, M., Witkin, A., & Terzopoulos, D. (1988). Snakes: Active contour models. International Journal of Computer Vision, 1(4), 321-331.

[13] Liu, X., Zhang, H., & Zheng, Y. (2016). Random forest-based segmentation of brain tumor in MRI images. Computational and Mathematical Methods in Medicine, 2016, Article ID 5726187.

[14] Matuszewski, B., Dabrowski, W., & Wawrzyniak, T. (2019). Edge detection in medical imaging: A review. Computers in Biology and Medicine, 111, 103350.

[15] Otsu, N. (1979). A threshold selection method from gray-level histograms. IEEE Transactions on Systems, Man, and Cybernetics, 9(1), 62-66.

[16] Ronneberger, O., Fischer, P., & Brox, T. (2015). U-Net: Convolutional networks for biomedical image segmentation. In Medical Image Computing and Computer-Assisted Intervention (MICCAI), 234-241.

[17] Suk, H. I., Lee, S. W., & Shen, D. (2017). Hierarchical feature learning for brain tumor segmentation. Neurocomputing, 264, 179-188.

[18] Zhang, H., Wang, H., & Zhang, X. (2018). An improved Canny edge detector using adaptive thresholding. Journal of Visual Communication and Image Representation, 55, 47-56.