Soft Computing Based Medical Image Analysis

By Nilanjan Dey, Amira S. Ashour, Fuquian Shi and Valentina Emilia Balas

Soft Computing Based Medical Image Analysis presents the foremost techniques of soft computing in medical image analysis and processing. It includes image enhancement, segmentation, classification-based soft computing, and their application in diagnostic imaging, as well as an extensive background for the development of intelligent systems based on soft computing used in medical image analysis and processing. The book introduces the theory and concepts of digital image analysis and processing based on soft computing with real-world medical imaging applications. Comparative studies for soft computing based medical imaging techniques and traditional approaches in medicine are addressed, providing flexible and sophisticated application-oriented solutions.

• Covers numerous soft computing approaches, including fuzzy logic, neural networks, evolutionary computing, rough sets and Swarm intelligence
• Presents transverse research in soft computing formation from various engineering and industrial sectors in the medical domain
• Highlights challenges and the future scope for soft computing based medical analysis and processing techniques

• Language: English
• Publisher: Academic Press
• Release date: Jan 18, 2018
• ISBN: 9780128131749

Nilanjan Dey

Nilanjan Dey is an Associate Professor in the Department of Computer Science and Engineering, Techno International New Town, Kolkata, India. He is a visiting fellow of the University of Reading, UK. He also holds a position of Adjunct Professor at Ton Duc Thang University, Ho Chi Minh City, Vietnam. Previously, he held an honorary position of Visiting Scientist at Global Biomedical Technologies Inc., CA, USA (2012–2015). He was awarded his PhD from Jadavpur University in 2015. He is the Editor-in-Chief of the International Journal of Ambient Computing and Intelligence , IGI Global, USA. He is the Series Co-Editor of Springer Tracts in Nature-Inspired Computing (SpringerNature), Data-Intensive Research(SpringerNature), Advances in Ubiquitous Sensing Applications for Healthcare (Elsevier). He was an associate editor of IET Image Processing and editorial board member of Complex & Intelligent Systems, Springer Nature. He is an editorial board member of Applied Soft Computing, Elsevier. He is having 35 authored books and over 300 publications in the area of medical imaging, machine learning, computer aided diagnosis, data mining, etc. He is the Fellow of IETE and Senior member of IEEE.

Book preview

Soft Computing Based Medical Image Analysis

First Edition

Nilanjan Dey

Amira S. Ashour

Fuqian Shi

Valentina E. Balas

Table of Contents

Cover image

Title page

Copyright

Contributors

Preface

Acknowledgments

Section A: Medical Image Analysis and Processing

Chapter 1: Computing in Medical Image Analysis

Abstract

1 Introduction

2 Medical Image Segmentation Techniques

3 Metaheuristics

4 Segmentation Algorithms for Medical Images

5 Conclusion

Chapter 2: Automated Pathology Image Analysis

Abstract

1 Introduction

2 Need for Quantitative Image Analysis in Pathology

3 Histology-Imaging Technologies

4 Automated Pathology Image Analysis

5 Pathology Image Data Sources

6 Discussion

7 Future Trends and Open Issues

Chapter 3: Multiple Kernel-Learning Approach for Medical Image Analysis

Abstract

Acknowledgments

1 Introduction

2 Related Literature

3 Nature and Characteristics of Biomedical Images

4 Feature Extraction and Image Descriptors

5 Computer-Aided Diagnosis

6 Kernel-Based Machine Learning

7 Multiple Kernel-Learning Model

8 Multiple Kernel Learning for Biomedical Image Analysis

9 Discussion

10 Conclusion

Section B: Medical Image Enhancement

Chapter 4: Efficient Medical Image Enhancement Technique Using Transform HSV Space and Adaptive Histogram Equalization

Abstract

1 Introduction

2 Proposed Method

3 Experimental Results and Discussions

4 Comparative Study

5 Conclusion

Chapter 5: Enhancement and Despeckling of Echocardiographic Images

Abstract

1 Introduction

2 Methodology for Contrast Enhancement

3 Despeckling of Echocardiographic Images

4 Results

5 Discussions

6 Conclusions

Section C: Detection and Prediction in Medical Imaging

Chapter 6: Automated Detection of Early Oral Cancer Trends in Habitual Smokers

Abstract

1 Introduction

2 Literature Review

3 Screening Methodology

4 CAD-Based Analysis of Multimodal Images

5 Methodology

6 Results and Discussion

7 Conclusion and Future Scope

Chapter 7: ScPSO-Based Multithresholding Modalities for Suspicious Region Detection on Mammograms

Abstract

Acknowledgment

1 Introduction

2 Methods

3 Experiments and Discussion

4 Discussions

5 Conclusions

Chapter 8: A Set of Texture-Based Methods for Breast Cancer Response Prediction in Neoadjuvant Chemotherapy Treatment

Abstract

1 Introduction

2 Related Works

3 Materials and Methods

SFTA Extraction Algorithm

4 Results and Discussion

5 Conclusion

Chapter 9: Dempster-Shafer Fusion for Effective Retinal Vessels’ Diameter Measurement

Abstract

1 Introduction

2 Materials and Methods

3 Experimental Results

4 Discussion

5 Conclusions

Section D: Machine Learning in Medical Image Segmentation and Classification

Chapter 10: State-of-the-Art of Level-Set Methods in Segmentation and Registration of Spectral Domain Optical Coherence Tomographic Retinal Images

Abstract

Acknowledgments

1 Introduction

2 Segmentation

3 Thresholding Based

4 Level-Set Methods

5 Snakes: Active Contour Approach

6 Variational Level-Set Models

7 Statistical Region-Based Level-Set Method

8 Segmentation Using Graph-Cut Method

9 Segmentation Based on Diffusion Maps

10 Segmentation Based on Machine Learning

11 Kernel Regression-Based Classifier Training

12 Registration

13 Registration of Fundus and SDOCT Retinal Images for Identification of Geographic Atrophy

14 Registration of SDOCT and HDOCT Retinal Images

15 Future Trends and Summary

Chapter 11: Deep Learning for Automated Brain Tumor Segmentation in MRI Images

Abstract

1 Introduction

2 Conventional Methods of Brain Tumor Segmentation

3 Deep Learning

4 CNN Architectures for Brain Tumor Segmentation

5 Tools for CNN

6 Discussions

7 Summary, Challenges, and Future Direction

Acknowledgments

Chapter 12: Machine-Learning Approach for Ribonucleic Acid Primary and Secondary Structure Prediction from Images

Abstract

1 Introduction

2 System Architecture

3 Result Analysis and Discussion

4 Conclusion

Chapter 13: Classification of Breast Density Patterns Using PNN, NFC, and SVM Classifiers

Abstract

1 Introduction

2 Methodology Adopted

3 Experiments and Results

4 Conclusion

Chapter 14: Classification of Sonoelastography Images of Prostate Cancer Using Transformation-Based Feature Extraction Techniques

Abstract

1 Introduction

2 Related Works

3 Theory of Sonoelastography

4 Methodology

Algorithm for Proposed Methodology

5 Results and Discussions

6 Conclusion and Future Direction

Index

Copyright

Academic Press is an imprint of Elsevier

125 London Wall, London EC2Y 5AS, United Kingdom

525 B Street, Suite 1800, San Diego, CA 92101-4495, United States

50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States

The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom

© 2018 Elsevier Inc. All rights reserved.

No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Details on how to seek permission, further information about the Publisher’s permissions policies and our arrangements with organizations such as the Copyright Clearance Center and the Copyright Licensing Agency, can be found at our website: www.elsevier.com/permissions.

This book and the individual contributions contained in it are protected under copyright by the Publisher (other than as may be noted herein).

Notices

Knowledge and best practice in this field are constantly changing. As new research and experience broaden our understanding, changes in research methods, professional practices, or medical treatment may become necessary.

Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using any information, methods, compounds, or experiments described herein. In using such information or methods they should be mindful of their own safety and the safety of others, including parties for whom they have a professional responsibility.

To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume any liability for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions, or ideas contained in the material herein.

Library of Congress Cataloging-in-Publication Data

A catalog record for this book is available from the Library of Congress

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library

ISBN 978-0-12-813087-2

For information on all Academic Press publications visit our website at https://www.elsevier.com/books-and-journals

Publisher: Mara Conner

Acquisition Editor: Chris Katsaropoulos

Editorial Project Manager: Thomas Van Der Ploeg

Production Project Manager: Sruthi Satheesh

Cover Designer: Mark Rogers

Typeset by SPi Global, India

Contributors

Rupal R. Agravat     Ahmedabad University, Ahmedabad, India

Mohammed Ammar

University of Tlemcen, Tlemcen

University of Boumerdes, Boumerdes, Algeria

Amira S. Ashour     Tanta University, Tanta, Egypt

Ananya Barui

IIEST, Shibpur

Indian Institute of Engineering Science and Technology, Howrah, India

Harvendra S. Bhadauria     G.B. Pant Engineering College, Pauri, Garhwal, India

Satish S. Bhairannawar     Sri Dharmastala Manjunatheshwara College of Engineering and Technology (SDMCET), Dharwad, India

Rahime Ceylan     Selcuk University, Konya, Turkey

Kalpana Chauhan     SIRDA Group of Institutions, Sunder Nagar, Sundernagar, India

Rajeev K. Chauhan     Indian Institute of Technology Mandi, Mandi, India

Linkon Chowdhury     Chittagong University of Engineering and Technology, Chittagong, Bangladesh

Susmita Dey     B.P. Poddar Institute of Management & Technology, Kolkata, India

Nilanjan Dey     Techno India College of Technology, Kolkata, India

Hasan Koyuncu     Selcuk University, Konya, Turkey

Kriti     Thapar University, Patiala, India

Indrajeet Kumar     G.B. Pant Engineering College, Pauri, Garhwal, India

Koushik Layek     Indian Institute of Engineering Science and Technology, Howrah, India

Saïd Mahmoudi     University of Mons, Mons, Belgium

Santi P. Maity     IIEST, Shibpur, Howrah, India

Santi Prasad Maity     Indian Institute of Engineering Science and Technology, Howrah, India

Luminita Moraru     Dunarea de Jos University of Galati, Galati, Romania

Saeeda Naz

Hazara University, Mansehra

Govt. Girls Postgraduate No.1 College, Abbottabad, Pakistan

Cristian D. Obreja     Dunarea de Jos University of Galati, Galati, Romania

Natarajan Padmasini     Rajalakshmi Engineering College, Chennai, India

Manoj K. Panda     G.B. Pant Engineering College, Pauri, Garhwal, India

Mehul S. Raval     Ahmedabad University, Ahmedabad, India

Khalid Raza     Jamia Millia Islamia, New Delhi, India

Muhammad Imran Razzak     King Saud bin Abdulaziz University for Health Sciences, Riyadh, Saudi Arabia

Shamim Ripon     East West University Bangladesh, Dhaka, Bangladesh

Anup Sadhu     Kolkata Medical College, Kolkata, India

Anju Saini     VIT Bhopal University, Bhopal, India

Sourav Samanta     University Institute of Technology, The University of Burdwan, Burdwan, India

Syed Hamad Shirazi     Hazara University, Mansehra, Pakistan

Mohamed Yacin Sikkandar     CAMS, Majmaah University, Riyadh, Saudi Arabia

Drisis Stylianos     Jules Bordet Institute, Brussels, Belgium

Dewaki N. Tibarewala     Jadavpur University, Kolkata, India

Rengasamy Umamaheswari     Velammal Engineering College, Chennai, India

Arif Iqbal Umar     Hazara University, Mansehra, Pakistan

Jitendra Virmani     CSIR-CSIO, Chandigarh, India

Nisar Wani

Jamia Millia Islamia, New Delhi

Govt. Degree College Baramulla, Baramulla

University of Kashmir, Srinagar, India

Ahmad Zaib     Women Medical College, Women and Children DHQ Hospital, Abbottabad, Pakistan

Preface

Advancement in medical imaging modalities enables acquiring thin-sliced, sectional, high-dimensional, huge number of images in a short time. The massive data volume becomes a great challenge for physicians’ interpretation of such acquired medical images. Consequently, image analysis and processing becomes indispensable to explore these complex data. Several researchers have been developing various techniques based on soft computing to support medical image analysis. Soft computing handles inherent uncertainties in the acquired images and data efficiently. Thus, it is used in several medical image analyses and processing applications. In the past decade, machine learning, artificial intelligence, and fuzzy techniques have been used for accurate and efficient image enhancement, detection/prediction, segmentation, and classification. Typically, soft computing methodologies include neural networks, chaos theory, probabilistic methods, evolutional computation, fuzzy logic, and support vector machines. These methodologies are often exploited to realize accurate diagnosis based on image processing for prompt detection of life-threatening diseases.

This book comprises 13 chapters, containing four sections, where Section 1 includes two chapters that deal with medical image analysis and processing. These chapters include a comprehensive study on machine-aided automatic analysis of microscopic histology images as an impact tool of diagnosis. Furthermore, different machine-learning techniques, including kernel methods and support vector machine, have been expansively applied to medical image analysis as multiple kernel learning is considered a state-of-art computational framework. Furthermore, Section 2, including two chapters, proposes several enhancement techniques including contrast enhancement using the difference in visual properties, edge enhancement due to compression, and image fusion to enhance the image quality. Enhancement and despeckling techniques for echocardiographic images are discussed. Moreover, detection and prediction are vital aspects in the medical domain. Thus, Section 3, including three chapters, proposes and discusses these processes including the analysis of a computer-aided diagnosis system in early cancer monitoring, where the support vector machine is employed to detect the precancer risk in habitual smokers through classification using the basis of selected features with morphological features. However, the LBP method for texture feature extraction proved its efficiency in this work. Another study handled the design and stochastic multithresholding modalities are proposed using an effective optimization algorithm, namely, the Scout Particle Swarm Optimization (ScPSO), while texture-based feature techniques, such as co-occurrence matrix, fractal analysis, and local binary pattern features extracted from magnetic resonance imaging, are carried out to quantify response of tumor treatment. Diameter of the retinal vessels is also measured, where Laplacian of Gaussian, Canny, and Gabor filter edge detection algorithms with a combined approach using the Dempster-Shafer fusion algorithm in an edge detection framework is proposed. Finally, in Section 4, five chapters deal with an elaborate and illustrative discussion about various machine-learning techniques in medical image segmentation and classification. Segmentation plays a significant role in the quantitative assessment of diabetic maculopathy from Spectral Domain optical coherence tomography (SDOCT) images during the analysis of intraretinal fluid-filled regions. A chapter offering a comprehensive review of the soft computing techniques applied to SDOCT retinal image analysis, particularly for image segmentation and registration techniques, is included. Then, a state-of-the-art review chapter that covers automated brain tumor segmentation as well as supervised learning, including the conventional methods, is introduced. Another chapter proposes Ribonucleic acid (RNA) prediction approach from RNA images based on hidden Markov model and Chapman Kolmogrov equation with filtering process. Yet another one introduces a computer-assisted framework to classify the breast density using digitized screen film mammograms by extracting texture features for classification. Finally, a computer-aided diagnosis system is presented for sonoelastography images to determine cancers through analyzing tissue elasticity.

We, the editors, have immense gratitude for the authors’ high-quality contributions as well as to the respected referees for their accurate, detailed, and timeless review comments. Special thanks go to our publisher, Elsevier. This book is aimed at a wide readership—from undergraduate students to postgraduate students as well as for professionals, researchers, and engineers. We hope this book will stimulate further research in medical imaging applications based on soft computing techniques and algorithms. We wish also that this book brings promising and outstanding research results that can support further development in medical imaging and soft computing techniques.

Volume Editors

Nilanjan Dey, Techno India College of Technology, Kolkata, India

Amira S. Ashour, Tanta University, Tanta, Egypt

Fuqian Shi, Wenzhou Medical University, Wenzhou, PR China

Valentina E. Balas, University Aurel Vlaicu Arad, Arad, Romania

Acknowledgments

Programming in machine code is like eating with a toothpick.

Charles Petzold

We would like to thank our parents and families for their endless support through the compilation of this book. Our appreciation is also directed to all who supported, shared, read, wrote, and offered comments through the book journey, including the authors for their valuable contributions and the reviewers for their accurate, detailed, and timeless review process.

Special thanks to the Elsevier publishing team, who showed us the ropes to start and continue as well as our readers, who gave us their thrust and hope our work inspired and guided them.

Last but not the least, we would like to thank our readers reading our book and hoping they will find it a useful resource in its domain.

Section A

Medical Image Analysis and Processing

Chapter 1

Computing in Medical Image Analysis

Nilanjan Dey⁎; Amira S. Ashour†    ⁎ Techno India College of Technology, Kolkata, India

† Tanta University, Tanta, Egypt

Abstract

Artificial intelligence is the outlet of computer science that deals with creating computers that perform as humans. It compromises expert systems, playing games, natural language, and robotics. However, soft computing (SC) varies from the hard (conventional) computing in its tolerance of partial truth, uncertainty, imprecision, and approximation; thus, it models the human mind. The most common SC techniques include neural networks, fuzzy systems, machine learning, and the metaheuristics stochastic algorithms (e.g., Cellular automata, ant colony optimization, Memetic algorithms, particle swarms, Tabu search, evolutionary computation, and simulated annealing). Due to the required accurate diseases analysis, magnetic resonance imaging, computed tomography images and images of other modalities segmentation remain a challenging problem. Over the past years, soft computing approaches have attracted attention of several researchers for problems solving in medical data applications. Image segmentation is the process that partitioned an image into some groups based on similarity measures. This process is employed for abnormalities volumetric analysis in medical images to identify the disease nature. Recently, metaheuristics algorithms are conducted to support the segmentation techniques. In the current chapter, different segmentation procedures are addressed. Several metaheuristics approaches are reported with highlights on their procedures. Finally, several medical applications using metaheuristics based-approaches for segmentation are discussed.

Keywords

Soft computing; Medical image analysis; Image segmentation; Magnetic resonance imaging; Computed tomography; Metaheuristics algorithms; Particle swarm optimization; Genetic algorithm

1 Introduction

Image analysis, pattern recognition, image disciplines are the foremost domains of computer engineering and computer science in several domains, such as medical, military, astronomy, and real-world applications. In the medical applications, image-guided therapy is one of the vital methods for accurate diagnosis. Medical image computing has a rising prominence for medical diagnosis. Image analysis systems, incorporated with innovative image computing procedures, were carried out to extract quantifiable parameters from the medical image in order to support the diagnosis and treatment. In order to achieve accurate clinical routine, automated and robust medical image computing techniques has become an active research area. Model-based image analysis as well as image-based modeling methods have become significant tools for accurate assessable analysis of the objects in the medical image. These methods require earlier information about the medical images’ structures, including bones, tumors, tissue, vessels, and organs. Afterward, the image-based modeling approaches are applied to extract the significant features automatically. For complex visualization and quantitative measures, processing of digital indicative imaging data was carried out to support disease progression monitoring, diagnosis, and preoperative planning. Nevertheless, successful image analysis requires optimized and complex processing systems, which is a challenging aspect. Currently, research in medical image analysis is pursued by an ongoing stream of successful new clinical applications to achieve robust solutions based on computing techniques [1–6].

Medical images’ segmentation and classification processed considered the main image-processing approaches. High soft-tissue contrast of segmentation of magnetic resonance (MR) images has a major role for evaluating the brain tumors’ therapy. Manual segmentation by physicians is still the segmentation gold standard of atypical brain images; however, it is disposed to human bias/error as well as its tedious process. An endeavor for reliable computerization of medical image segmentation is consequently extremely desirable. This leads to the necessity to use clustering algorithms to label the pixels of the medical image into prearranged clusters. Furthermore, medical image classification is extensively used to discriminate the abnormal and normal images. Automated classification process is highly desired to support the physicians in the analysis monotonous task.

Artificial intelligence and soft computing play a significant role in several applications [7–12]. Computing algorithms have an imperative role in medical image analysis and processing to solve the inherent uncertainties in the captured medical image data. Recently, researchers developed fuzzy methods for image segmentation, fuzzy clustering approaches, and fuzzy methods for object delineation and recognition [13]. Generally, computing methods comprise support vector machines, neural networks, fuzzy logic, probabilistic methods, chaos theory, and evolutional computation. Medical image computing improves mathematical and computational approaches to solve problems related to medical images [14]. Medical image computing focuses on the images computational analysis. Generally, medical image computing methods can be categorized into image segmentation-, image classification-, and image registration-based computational techniques. Typically, medical image computing works on sampled data with consistent spatial space. The data under concern are based on the used modality, namely, Computed Radiography (CT), positron emission tomography (PET), MRI, X-ray, optical coherence tomography (OCT), PET-MRI, and PET-CT. These imaging modalities capture images of the different human organs [15]. Recently, numerous meta-heuristics optimization algorithms have been developed and involved in several medical and real applications, namely, cuckoo search (CS), Particle swarm optimization (PSO), genetic algorithms (GA), and Firefly algorithm (FA) [16–19].

2 Medical Image Segmentation Techniques

Automation of image-processing and analysis techniques is compulsory to assist physicians in treatment planning and clinical diagnosis. Consistent algorithms are mandatory for the delineation of regions of interest (ROI) and the anatomical structures. Subsequently, computer-aided diagnosis (CAD) was developed to acquire accurate medical images analysis process, achieve fast results using high-speed computers, and support information technology for faster communication with patients at remote areas. Medical images segmentation techniques are specific to the imaging modality used, the application under concern, and type of the considered body part. However, there is no universal segmentation algorithm that can be used efficiently with all medical image types, where each imaging modality has its own definite limits.

Segmentation is defined as the process of allotting an image into several regions having similar properties, including color, gray level, contrast brightness, and texture. Its basic role in the medical domain is to identify the ROI, such as lesion, tumor, and any abnormalities, study the anatomical structures, measure the tissue volume, and assist treatment planning. Automated medical images segmentation is complex; in addition, the inhomogeneous intensity, low contrast, partial volume effect, artifacts, and the close gray-level values of the different soft tissues affect the segmented images. Consequently, several segmentation techniques have been employed in the medical applications, which can be categorized into (i) Shape-based segmentation, (ii) interactive segmentation, (iii) atlas-based segmentation and (iv) shape-based segmentation. Another classification of the segmentation techniques is as follows, (i) histogram-based segmentation, region-based segmentation, and edge-based segmentation, which is based on the gray-level features, and (ii) texture-features-based segmentation [20].

2.1 Histogram-Based Segmentation

Histogram-based segmentation depends mainly on a thresholding value of the histogram. This technique is used with uniform brightness regions within the image under consideration, where the threshold is employed to segment the ROI and background. The threshold can have single value or multiple values based on the number of ROI within the image [21]. Several computing algorithms are used to perform thresholding segmentation effectively. Commonly, the whole image is scanned pixel by pixel to label the pixels into object or the background based on the gray-level value compared to the thresholding function (T). The main steps of the global thresholding are as follows.

Algorithm: Global Thresholding Algorithm

Start

Chose an initial threshold value T

Apply T to segment the image leading to two groups of pixels

      R1 are pixels with gray level > T

      R2 are pixels with gray level <= T

Measure the average gray level values mean1 and mean2 for the pixels in R1 and R2.

Calculate the new threshold value Tnew = (mean1 + mean2)/2

Repeat the previous steps till the difference in T is smaller than a pre-defined threshold T0 in the successive iterations

End

2.2 Region-Based Segmentation

Region-based segmentation partitions the image into dissimilar regions to determine the images’ disjoint regions directly. Region split-and-merge segmentation as well as region growing segmentation are the major region-based segmentation techniques [22]. The steps of the region-based segmentation are as follows.

Algorithm: Region-Based Segmentation Algorithm (Region Growing)

Start

Merge iteratively an initial small area set based on similarity

Select an arbitrary pixel

Compare the arbitrary pixel with the neighboring pixel

Add similar neighboring pixels that increase the region size from the seed pixel

If one region growth stops

Select another seed pixel that does not belong to any region

Repeat

Endif

Repeat the whole process until all pixels fit some region

End

2.3 Split-and-Merge Segmentation

Split-and-merge-based segmentation depends on the quad quadrant tree data depiction, where the image segment is divided into four quadrants providing the nonuniform original segment. Afterward, the four neighboring squares are fused based on the segments uniformity [23]. This split/merge process is iterative and repeated until all possible split/merge occurs using the following steps.

Algorithm: Split-and-Merge Segmentation Algorithm

Start

Delineate homogeneity criterion.

Divide the image into 4 square quadrants

If inhomogeneous square result

Split it further into 4 quadrants

Endif

Merge the two/more neighboring regions that satisfy the homogeneity condition at each level

Carry on the split/merge till no additional split/merge of regions is possible

End

2.4 Edge-Based Segmentation

Edge-based segmentation is considered one of the vital segmentation methods, where edges embrace much information about the image. Edges represent the boundaries between any two dissimilar regions that provide information about the objects location, their size/shape, and their texture [24]. Thus, the gradient can be calculated to determine the pixel values differences between the regions at which the image intensity changes from high value to low value or vice versa. Hough transform-based, border detection, edge relaxation are different methods of edge-based segmentation.

Algorithm: Edge-Based Segmentation Algorithm

Start

Use the derivative of the image for detecting the edges

Measure the gradient amplitude to calculate the edges strength

Preserve all edge with greater magnitude than a threshold T

Determine the crack edges

Repeat the previous two steps with different threshold values to realize the closed boundaries

End

3 Metaheuristics

Global optimization (GO) algorithm is applied to determine the global optimum of a fitness function in the search space. The GO algorithms have two sets, namely, evolutionary and deterministic. The deterministic approaches locate the local minimums of the fitness function, while the evolutionary procedures activate over the candidate solutions population, thus, localizing the global optimum faster than the deterministic ones. Recently, metaheuristics (MHs) algorithms are applied for optimization; however, such approaches are still deliberated as an open research problem due to several complexities, including the overcoming of the local optimum and the premature convergence [25].

Metaheuristics are defined as a stochastic optimization procedure that use brute force or random search to find the optimal solutions of the problems under consideration. The foremost aim of the MHs’ learning/optimization algorithms is to realize a trade-off between diversification and intensification, where the exploration (diversification) denotes creating different solutions in order to discover the search space on a global scale, whereas the exploitation (intensification) entails directing the search onto a local region at which good solutions are established. These optimization algorithms have common characteristics, such as (i) being reliable and robust, (ii) implied parallelism, (iii) easy implementation, (iv) approximate and nondeterministic, (v) explore efficiently search spaces, and (vi) global search ability. MHs techniques can be categorized into Memetic algorithms, Population-based methods, and Trajectory methods [26].

Memetic algorithms are hybrid local/global search approaches, where a local enhancement procedure is integrated into a population-based procedure [27]. The basic concept is to emulate the social interaction/learning effect of the individuals by a local enhancement procedure attached to the established solutions by the global search operators. Thus, memetic algorithms contain several virtual potential hybridizations of prevailing approaches. Population-based approaches handle a solution’s population in every iteration of the procedure. Such algorithms include genetic algorithms, evolutionary programming, evolution algorithms, and swarm intelligence techniques (e.g., particle swarm optimization). Generally, swarm intelligence approaches imitate the collective activities of dispersed, self-organized artificial systems at which global search is an evolving action of the agents’ population. Furthermore, the evolutionary algorithms depend on a computational model that imitates appliances inspired by the biological evolution, namely, reproduction, mutation, recombination, and selection, in order to solve the optimization problems.

The most popular procedure for continuous optimization is the differential evolution that inherits its characteristics from both the swarm intelligence algorithms and the evolutionary algorithms [28]. Moreover, the trajectory procedures are considered as the evolution algorithms in the discrete time of a distinct dynamical system. In such algorithms, the search process designates a trajectory in the search space [29]. Generally, each MHs category includes a huge number of methods. Generally, the most prevalent swarm-inspired algorithms are Bee colony optimization, bacteria optimization algorithm, firefly algorithm, and cuckoo search algorithms.

3.1 Genetic Algorithm

The GA is a search procedure for optimizing common combinational problems. It is one of the most prevalent evolutionary algorithms that is inspired by Darwin’s evolution theory (i.e., survival of the fitness) to solve difficult optimization problems. A solution in the GA is called individual, while the term populations is used to refer to set of individuals [30]. The GA operators are the selection, crossover, and mutation, where the selection associates to the fitness survival, the crossover signifies the mating between individuals, and the mutation presents random modifications. The genetic algorithm steps are as follows.

Algorithm: Genetic Algorithm

Start

Produce random population of n appropriate solutions (chromosomes)

Assess the fitness f(x) of each solution in the population

Generate a new population by reciting following phases until completing the new population

Choose from the population, two parent chromosomes based on their fitness (Selection)

Cross over to form a new offspring (Crossover)

Mutate new offspring with a mutation probability at each position in the chromosome (Mutation)

Place in a new population, new offspring

Use new produced population

Test once satisfying the end condition

Stop/return the best solution

Repeat previous steps

End

3.2 Particle Swarm Optimization

The PSO is one of the contemporary meta-heuristic population-based stochastic optimizations that is carried out on noncontinuous and nonlinear optimization problems. It delivers an evolutionary-based search to discover the near optimal/optimal solutions. The PSO behavior is envisioned from the searching strategy for optimal food sources by the bird swarms. The bird movement direction is influenced by its present movement to find the best food source. Typically, the birds are motivated by their personal knowledge, inertia, and the swarm knowledge. Thus, in the PSO algorithm, the particle movement is affected by its personal-/global-best position, and its inertia [31]. The PSO algorithm consists of multiple particles, where each has its position, velocity, and current objective value. It preserves the global best value, which is related to the best objective value, and the global best position at which the global best value achieved. The PSO algorithm consists of the following repeated steps until the predetermined stopping condition is achieved.

Algorithm: Particle Swarm Optimization

Start

Initialize the position and velocity of the particles

Determine stopping criteria

Assess the each particle’s fitness using the objective function

Update the individual/global best positions and finesses

Update position/velocity of each particle

Stop

3.3 Ant Colony Optimization

The ACO metaheuristic algorithm is a modern population-based approach enthused by the real ants colony collective foraging behavior. In ACO, a pheromone trail is known as the ants’ movement in a straight line that joins the food source to their nests to obtain optimum value from a population. Problem solutions are constructed in a stochastic iterative process, where each individual ant creates a part of the solution using a pheromone [32].

3.4 Bat Optimization Algorithm

The bats’ echolocation behavior inspired the implementation of the BAT algorithm. The ability of microbats echolocation is attractive. This bats’ behavior inspired the BAT algorithm based on the microbats’ echolocation characteristics. The rules used in the BAT algorithm are, namely (i) to sense distance, background barriers, and dissimilarity between prey/food, bats use echolocation, (ii) bats fly randomly with certain velocity at specific position with a fixed frequency but varying wavelengths to search for prey, and (iii) the loudness changes from a large to a minimum constant value [33]. The inclusive pseudocode of the BAT algorithm is as follows.

Algorithm: Particle Swarm Optimization

Start

Delineate the initial population and velocities vector

Chose a pulse frequency, rates and loudness

Do until get to number of iterations

Compute new solutions via frequency

Update velocity/location for each particle (bat)

Produce a random value and compare to

If a solution is the best solutions, a new local solution is produced

Use a random bat’s fly to generate a new solution

Re-evaluate all particles to determine the current new best

End

4 Segmentation Algorithms for Medical Images

Several researches are interested in different organs segmentation for extracting the suspicious regions from the medical images [13]. Prevalent procedures using supervised algorithms include active appearance models (AAM), supervised support vector machine (SVMs), and artificial neural network (ANNs) and are considered for medical image processing, which require training set. The ANNs and SVMs are nonlinear statistical data exhibiting methods for modeling complex associations between inputs and outputs. The classifiers’ weights are chosen by optimizing the energy function distinct by the features of organs, structures, cells, etc. These weights are reorganized through handling each sample in the training set. Thus, metaheuristics can be involved for optimal weights selection. From the training set, the extracted information offers essential cues of the structures, including shape, position, and intensity that can be valued corresponding information for the test images segmentation. However, the AAM are statistical models of the structures’ shape, where the training samples are employed to extract the shape parameters’ ranges, mean appearance, and mean shape. In order to ensure the similarity between the segmentation result and the training samples, restrictions on shape parameters are required, where the segmentation technique is to find the superior locations of the shape points based on the appearance information. Consequently, in medical images, the algorithms established on classifiers can be widely applied to segment organs, such as the brain and cardiac images.

4.1 Metaheuristics-Based Segmentation of Magnetic Resonance Images

Brain tumor occurs when infrequent cells shape appear inside the cerebrum, which has two main types, namely, benign tumors and malignant tumors. Malignant tumors can be categorized into basic tumors, and secondary tumors that spread elsewhere. Currently, the MRI is one of the unsurpassed technologies for brain tumor diagnosis. In addition, segmentation has a significant role to extract suspicious areas from complex brain medical images. Automated brain tumor detection through MRI can offer valued outlook and earlier accurate detection of the brain tumor. Gopal and Karnan [34] designed an intelligent system for brain tumor diagnosis through MRI exhausting image-processing clustering procedures, namely, Fuzzy-C Means along with optimization intelligent algorithms, including PSO and GA. The tumor detection is performed in several stages, namely, enhancement, segmentation, and classification.

Karnan and Logheshwari [35] employed a population-based approach, namely, the Ant Colony Optimization (ACO) metaheuristic, which is stimulated real ants colony and their collective foraging performance. The authors proposed hybrid technique of the ACO with Fuzzy segmentation. Initially, the MRI brain image has been segmented using the proposed approach to extract the suspicious region. Afterward, the pixel similarity and tumor position of the proposed segmented process and the radiologist report were compared. Hamdaoui et al. [36] compared the performance of two metaheuristics swarm intelligence methods, namely, the PSO and Shuffled Frog Leaping Algorithm (SFLA) for MR brain medical images segmentation.

Ladgham et al. [37] designed a new metaheuristic algorithm, called the modified SFLA or MSFLA for fast MR brain image segmentation. The MSFLA allow the segmentation process without using denoising filter. A new fitness function has been used to quickly evaluate the particle frogs to arrange them in descendent order. The results included a comparative study with other metaheuristics for segmentation, namely, Genetic Algorithm (GA), and 3D-Otsu thresholding with SFLA. It has been established that the proposed MSFLA is capable to realize superior segmentation quality with less execution time compared to the use of GA or the SFLA.

Si et al. [38] implemented an MRI segmentation technique of brain tumor images based on entropy maximization using Grammatical Swarm (GS) algorithm. Quantitative assessment of MRI lesion load of patients with multiple sclerosis is vital for a better understanding of the pathology history as well as for natural/modified therapies. Numerous methods have been conducted for the MS lesions segmentation in MR images. Zangeneh and Yazdi [39] employed a constrained Gaussian Mixture Model (GMM) and GA to define the optimal highly nonlinear model’s parameters for the segmentation process. A preprocessing step has been executed for artifacts suppression and unwanted skull portions removal from the brain MRI. The proposed technique has been evaluated on real MR images proving the efficiency of the proposed method for segmenting the MS lesions in the MR images.

4.2 Metaheuristics Based Segmentation of Computed Tomography Images

For CT medical images segmentation, Bruyninckx et al. [40] proposed an algorithm for segmenting the liver portal veins from an arterial stage. Using minimal mechanical energy, the physiological model stated that the vasculature pattern is arranged such that the entire organ is perfused. In the image, the proposed method has been locally detecting the possible candidate vessel segments. The segments subset that produces the most plausible vessel tree is based on the physiological model and the image is subsequently sought by a global optimization technique. From CT images, the proposed technique has been applied for segmenting the lung vessel trees. In addition, an SVM has been used to cope with the low contrast to locally detect vessels. This proposed model can be applied to the liver, lungs, and kidney.

Bong et al. [41] proposed a multiobjective clustering ensemble technique, to segment lung CT images for candidate nodule detection. Fuzzy clustering has been used with optimization of three objective functions, namely, symmetry distance-based cluster validity index, global fuzzy clusters compactness, and fuzzy separation. The optimal solution has been determined using the metaclustering procedure. The results established that the proposed algorithm achieved positive predictive rate of 90%. Active Contour Models (ACMs) have superior performance compared to the traditional low-level method to segment ill-defined medical images. However, it is sensitive to the contour initial position and the setup in the local minima. Sahoo and Chandra [42] considered the ACM-based segmentation as an optimization problem find a minimal energy contour. A nature-inspired metaheuristic procedure, namely, the L'evy flight firefly algorithm (LFA) has been employed effectively to solve the global optimization problems. Thus, a hybrid technique based on integrating the ACM with LFA has been designed to improve its segmentation capability of real abdomen CT images.

Ramakrishnan and Sankaragomathi [43] proposed a technique for classifying CT images into tumor and the nontumor images followed by the tumor region segmentation in CT images. The classification process has been carried out using SVM with different kernel functions and optimization procedures. The Sequential Minimal Optimization (SMO)-based SVM classifier has a significant role. Modified Region Growing (MRG) based on threshold optimization has been applied for the segmentation process after the classification. Gray Wolf Optimization (GWO), Evolutionary Programming (EP), and Harmony Search (HS) have been used for threshold optimization. The experimental results reported 99.05% accuracy of the segmentation process using the GWO algorithm. It has been established that the proposed MRG-GWO achieved high accuracy with superior tumor detection compared to the HS and EP.

Liver segmentation is a challenging initial stage of liver diagnosis due to its likeness with other structures in terms of the intensity values. For liver image segmentation of the abdomen CT images, Mostafa et al. [44] proposed a gray wolf optimization-based approach. This approach carried out the gray wolf optimization, simple region growing, statistical image of liver, and Mean shift clustering method. Gray Wolf (GW) optimization algorithm has been applied on the preprocessed image to calculate the centroids of a predefined number of clusters. In the image, according to the intensity value of each pixel, the number of the nearest cluster was labeled on the pixel. In order to extract the probable area of the liver, a binary liver statistical image has been used. Lastly, the mean shift clustering procedure has been used for extracting the ROI in the liver.

5 Conclusion

In the medical domain, MRI, CT, and other modalities are conducted to distinguish pathological tissues from normal ones, and to acquire images of the different body parts for further analysis and processing. Image segmentation is the furthermost significant task in several computer-aided medical imaging applications. From MRI data, Tumor segmentation is considered an imperative process, while it is time consuming if accomplished manually. Thus, automated image analysis becomes essential to facilitate image-based diagnosis. Several techniques that have been employed in several applications can be used for medical image analysis [7,11,12,45–52]. In computer-aided systems, the analyzed computer-based images are used to support the radiologists and physicians in diagnosis in a faster mode. The current chapter reported different metaheuristic approaches and their uses in the MRI and CT images segmentation, including GA, PSO, ACO, and ABCO. These optimization algorithms are carried out to obtain the optimal parameters required during the segmentation process, where different segmentation methodologies are explained.

References

[1] Dey N., Ashour A.S. Classification and Clustering in Biomedical Signal Processing. IGI Global; 2016.

[2] Kotyk T., Dey N., Ashour A.S., Balas-Timar D., Chakraborty S., Ashour A.S., Tavares J.M.R. Measurement of glomerulus diameter and Bowman's space width of renal albino rats. Comput. Methods Prog. Biomed. 2016;126:143–153.

[3] Saba L., Dey N., Ashour A.S., Samanta S., Nath S.S., Chakraborty S., Suri J.S. Automated stratification of liver disease in ultrasound: an online accurate feature classification paradigm. Comput. Methods Prog. Biomed. 2016;130:118–134.

[4] Ahmed S.S., Dey N., Ashour A.S., Sifaki-Pistolla D., Bălas-Timar D., Balas V.E., Tavares J.M.R. Effect of fuzzy partitioning in Crohn’s disease classification: a neuro-fuzzy-based approach. Med. Biol. Eng. Comput. 2017;55(1):101–115.

[5] Ashour A.S., Beagum S., Dey N., Ashour A.S., Pistolla D.S., Nguyen G.N., Le D.-N., Shi F. Light microscopy image de-noising using optimized LPA-ICI filter. Neural Comput. & Applic. 2016;1–17.

[6] Dey N., Ashour A.S., Althoupety A.S. Thermal imaging in medical science. In: Recent Advances in Applied Thermal Imaging for Industrial Applications. IGI Global; 2017:87–117.

[7] Acharjya D., Anitha A. A comparative study of statistical and rough computing models in predictive data analysis. Int. J. Ambient Comput. Intell. 2017;8(2):32–51.

[8] Boulmaiz A., Messadeg D., Doghmane N., Taleb-Ahmed A. Design and implementation of a robust acoustic recognition system for Waterbird Species using TMS320C6713 DSK. Int. J. Ambient Comput. Intell. 2017;8(1):98–118.

[9] Bureš V., Tučník P., Mikulecký P., Mls K., Blecha P. Application of ambient intelligence in educational institutions: visions and architectures. Int. J. Ambient Comput. Intell. 2016;7(1):94–120.

[10] Mhetre N.A., Deshpande A.V., Mahalle P.N. Trust Management Model based