Advances in Intelligent Systems and Computing 1112

Jyotsna Kumar MandalSomnath Mukhopadhyay   Editors

Proceedings of the Global AI Congress 2019

Advances in Intelligent Systems and Computing

Volume 1112

Series Editor

Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences,Warsaw, Poland

Advisory Editors

Nikhil R. Pal, Indian Statistical Institute, Kolkata, IndiaRafael Bello Perez, Faculty of Mathematics, Physics and Computing,Universidad Central de Las Villas, Santa Clara, CubaEmilio S. Corchado, University of Salamanca, Salamanca, SpainHani Hagras, School of Computer Science and Electronic Engineering,University of Essex, Colchester, UKLászló T. Kóczy, Department of Automation, Széchenyi István University,Gyor, HungaryVladik Kreinovich, Department of Computer Science, University of Texasat El Paso, El Paso, TX, USAChin-Teng Lin, Department of Electrical Engineering, National ChiaoTung University, Hsinchu, TaiwanJie Lu, Faculty of Engineering and Information Technology,University of Technology Sydney, Sydney, NSW, AustraliaPatricia Melin, Graduate Program of Computer Science, Tijuana Instituteof Technology, Tijuana, MexicoNadia Nedjah, Department of Electronics Engineering, University of Rio de Janeiro,Rio de Janeiro, BrazilNgoc Thanh Nguyen , Faculty of Computer Science and Management,Wrocław University of Technology, Wrocław, PolandJun Wang, Department of Mechanical and Automation Engineering,The Chinese University of Hong Kong, Shatin, Hong Kong

The series “Advances in Intelligent Systems and Computing” contains publicationson theory, applications, and design methods of Intelligent Systems and IntelligentComputing. Virtually all disciplines such as engineering, natural sciences, computerand information science, ICT, economics, business, e-commerce, environment,healthcare, life science are covered. The list of topics spans all the areas of modernintelligent systems and computing such as: computational intelligence, soft comput-ing including neural networks, fuzzy systems, evolutionary computing and the fusionof these paradigms, social intelligence, ambient intelligence, computational neuro-science, artificial life, virtual worlds and society, cognitive science and systems,Perception and Vision, DNA and immune based systems, self-organizing andadaptive systems, e-Learning and teaching, human-centered and human-centriccomputing, recommender systems, intelligent control, robotics and mechatronicsincluding human-machine teaming, knowledge-based paradigms, learning para-digms, machine ethics, intelligent data analysis, knowledge management, intelligentagents, intelligent decision making and support, intelligent network security, trustmanagement, interactive entertainment, Web intelligence and multimedia.

The publications within “Advances in Intelligent Systems and Computing” areprimarily proceedings of important conferences, symposia and congresses. Theycover significant recent developments in the field, both of a foundational andapplicable character. An important characteristic feature of the series is the shortpublication time and world-wide distribution. This permits a rapid and broaddissemination of research results.

** Indexing: The books of this series are submitted to ISI Proceedings,EI-Compendex, DBLP, SCOPUS, Google Scholar and Springerlink **

More information about this series at http://www.springer.com/series/11156

Jyotsna Kumar Mandal •

Somnath MukhopadhyayEditors

Proceedings of the Global AICongress 2019

123

EditorsJyotsna Kumar MandalDepartment of Computer Scienceand EngineeringKalyani UniversityNadia, West Bengal, India

Somnath MukhopadhyayDepartment of Computer Scienceand EngineeringAssam UniversitySilchar, Assam, India

ISSN 2194-5357 ISSN 2194-5365 (electronic)Advances in Intelligent Systems and ComputingISBN 978-981-15-2187-4 ISBN 978-981-15-2188-1 (eBook)https://doi.org/10.1007/978-981-15-2188-1

© Springer Nature Singapore Pte Ltd. 2020This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or partof the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmissionor information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilarmethodology now known or hereafter developed.The use of general descriptive names, registered names, trademarks, service marks, etc. in thispublication does not imply, even in the absence of a specific statement, that such names are exempt fromthe relevant protective laws and regulations and therefore free for general use.The publisher, the authors and the editors are safe to assume that the advice and information in thisbook are believed to be true and accurate at the date of publication. Neither the publisher nor theauthors or the editors give a warranty, expressed or implied, with respect to the material containedherein or for any errors or omissions that may have been made. The publisher remains neutral with regardto jurisdictional claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,Singapore

Preface

Institute of Engineering and Management Kolkata organized the First InternationalConference on “Global AI Congress” (GAIC 2019), during 12–14 September 2019at IEM Gurukul Campus. This mega event covered all aspects of artificial intelli-gence, data science, computer vision, and internet of things, where the scope wasnot only limited to various engineering disciplines such as computer science,electronics, and biomedical engineering researchers but also included researchersfrom allied communities like data analytics and management science, etc. Thesubtracks of the conference were signal processing, intelligent forensics (privacyand security), and intelligent devices and networks.

The volume is a collection of high-quality peer-reviewed research papersreceived from all over the world. GAIC 2019 could attract a good number ofsubmissions from the different areas spanning over four tracks in variouscutting-edge technologies of specialized focus which were organized and chairedby eminent professors. Based on rigorous peer-review process by the technicalprogram committee members along with external experts as reviewers (inland aswell as abroad), best quality papers were identified for presentation and publication.The review process was extremely stringent with minimum three reviews for eachsubmission and occasionally up to six reviews. Checking of similarities andoverlaps are also done based on the international norms and standards. Out of thesubmission pool of received papers, only 30% were able to have got berth in thisfinal proceedings.

The organizing committee of GAIC 2019 was constituted with a strong inter-national academic and industrial luminaries and the technical program committeecomprising more than two hundred domain experts. The proceedings of the con-ference were published in Advances in Intelligent Systems and Computing (AISC),Springer. We, in the capacity of the volume editors, convey our sincere gratitude toSpringer Nature for providing the opportunity to publish the proceedings of GAIC2019 in AISC series.

v

The conference included distinguished speakers such as Dr. Mike Hinchey,President IFIP, Professor, University Of Limerick; Dr. L M Patnaik, Professor,Indian Institute of Science Bangalore; Prof. Goutam Chakraborty, Head, IntelligentInformatics Lab; Dr. Basabi Chakraborty, Iwate Prefectural University, Japan;Dr. Niloy Ganguly, Professor, IIT Kharagpur; Dr. K.K. Shukla, Professor,Department Of Computer Science And Engineering, IIT BHU, India; Dr. AmlanChakraborty, Dean Of Engg., University of Calcutta, (ACM DistinguishedSpeaker); Dr. Debasish De, Professor and Director, Department Of ComputerScience And Engineering, MAKAUT (WBUT); Dr. Arpan Pal, Principal Scientistand Head of Embedded Systems And Robotics, TCS; Mr. Joy Mustafi, Director andPrincipal Researcher, Salesforce; and Dr. Koyel Mukherjee, Researcher, IBMResearch Lab.

Sincerest thanks are due to Dr. Aninda Bose, Senior Publishing Editor withSpringer Nature and Prof. P.K. Roy, APIIT, India, for their valuable suggestionsregarding enhancing editorial review process. The editors also thank the othermembers of the award committee of GAIC 2019 for taking all the trouble to make avery contentious assignment of selecting the best papers from a pool of so manyformidable acceptances of the conference.

Special mention of words of appreciation is due to Prof. Debika Bhattacharyyaand Prof. Nilanjan Dutta Roy of IEM, Kolkata, for coming forward to host to theconference which incidentally was the second in the series. It was indeed hearteningto note the enthusiasm of all faculty, staff, and students of IEM to organize theconference in a professional manner. The involvement of faculty coordinators andstudent volunteers are particularly praiseworthy in this regard. The editors leave nostone unturned to thank the technical partners and sponsors for providing all thesupport and financial assistance.

It is needless to mention the role of the contributors. But for their active supportand participation, the question of organizing a conference is bound fall through. Theeditors take this privilege to thank authors of all the papers submitted as a resultof their hard work, more so because all of them considered the conference as aviable platform to ventilate some of their latest findings, not to speak of theiradherence to the deadlines and patience with the tedious review process. Thequality of a refereed volume primarily depends on the expertise and dedicationof their viewers who volunteer the associated trouble with smiling face, thanklessalbeit. The editors are further indebted to the technical program committee membersand external reviewers who not only produced excellent reviews but also did thesein short timeframes, in spite of their very busy schedule. Because of their qualitywork, it has been possible to maintain the high academic standard of theproceedings.

The conference may meet its completeness if it is able to attract elevated par-ticipation in its fold. A conference with good papers accepted and devoid of anyparticipant is perhaps the worst form of curse that may be imagined of. The editorswould like to appreciate the participants of the conference, who have considered theconference a befitting one in spite of all the hardship they had to undergo.

vi Preface

Last but not least, the editors would offer cognizance to all the volunteers fortheir tireless efforts in meeting the deadlines and arranging every minute detailmeticulously to ensure that the conference achieves its goal, academic, or otherwiseand that too unhindered.

Happy Reading!!!

Nadia, India Jyotsna Kumar MandalSilchar, India Somnath Mukhopadhyay

Preface vii

Contents

Detection of Leukemia in Blood Samples Applying Image ProcessingUsing a Novel Edge Detection Method . . . . . . . . . . . . . . . . . . . . . . . . . . 1Megha Dutta, Srija Karmakar, Prithaj Banerjee and Rit Ghatak

Biometric-Based Unimodal and Multimodal Person Identificationwith CNN Using Optimal Filter Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17Goutam Sarker and Swagata Ghosh

Bed Expansion in Two-Phase Liquid–Solid Fluidized Bedswith Non-Newtonian Fluids and ANN Modelling . . . . . . . . . . . . . . . . . . 33Samit Bikas Maiti, Nirjhar Bar and Sudip Kumar Das

M-UNet: Modified U-Net Segmentation Framework with SatelliteImagery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47Ashish Soni, Radhakanta Koner and Vasanta Govind Kumar Villuri

Study and Analysis of Various Heuristic Algorithms for SolvingTravelling Salesman Problem—A Survey . . . . . . . . . . . . . . . . . . . . . . . . 61Roneeta Purkayastha, Tanmay Chakraborty, Anirban Sahaand Debarka Mukhopadhyay

Question Answering System-Based Chatbot for Health care . . . . . . . . . 71Sharob Sinha, Suraj Mandal and Anupam Mondal

Improvement of Packet Delivery Fraction Due to Wormhole Attackby Modified DSR and AODV Algorithm . . . . . . . . . . . . . . . . . . . . . . . . 81Sayan Majumder

Air Pollution Forecasting Using Multiple Time Series Approach . . . . . . 91K. N. Tejasvini, G. R. Amith, Akhtharunnisa and H. Shilpa

Deep Learning-Based Smart Attendance Monitoring System . . . . . . . . . 101Rohit Halder, Rajdeep Chatterjee, Debarshi Kumar Sanyaland Pradeep Kumar Mallick

ix

Mutation-Based Chaotic Gravitational Search Algorithm . . . . . . . . . . . 117Moujinjir Mukherjee, Suman Mitra and Sriyankar Acharyya

SCap Net: A Capsule Network Based Approach for PersonRe-identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133Nirbhay Kumar Tagore and Arnab Mondal

An Embedded System for Gray Matter Segmentationof PET-Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145Khakon Das, Dipankar Khorat and Samarendra Kumar Sharma

Estimation of Resemblance and Risk Level of a Breast Cancer Patientby Prognostic Variables Using Microarray Gene Expression Data . . . . . 159Madhurima Das, Biswajit Jana, Suman Mitra and Sriyankar Acharyya

A Model for Classification of Skin Disease Using PretrainedConvolutional Neural Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173Riddhi Kumari Bhadoria and Suparna Biswas

Improvement of Packet Delivery Fraction Due to Discrete Attacksin MANET Using MAD Statistical Approach . . . . . . . . . . . . . . . . . . . . . 187Sayan Majumder and Debika Bhattacharyya

IoT Based Smart Posture Detector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197Greeshma Karanth, Niharika Pentapati, Shivangi Gupta and Roopa Ravish

Sensor-Based Traffic Control System . . . . . . . . . . . . . . . . . . . . . . . . . . . 207Roopa Ravish, Datthesh P. Shenoy and Shanta Rangaswamy

Addressing Grain-Matrix Differentiation in Sedimentary RockPhotomicrographs in the Light of Brightness Perception Modelling . . . . 223Rajdeep Das, B. Uma Shankar, Tapan Chakraborty and Kuntal Ghosh

Identification of Interfaces of Mixtures with Nonlinear Models . . . . . . . 237Srirupa Das and Somdatta Chakravortty

Random Forest Boosted CNN: An Empirical Technique for PlantClassification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251Somnath Banerjee and Rajendra Pamula

Edge–Texture-Based Characteristic Attribute on Local Radiusof Gyration Face for Human Face Recognition . . . . . . . . . . . . . . . . . . . 263Pinaki Prasad Guha Neogi

Centroid-Based Hierarchy Preserving Clustering Algorithm UsingLighthouse Scanning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275Soujanya Ray and Anupam Ghosh

Design and Implementation of an Automatic Summarizer UsingExtractive and Abstractive Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289Adhiraj Chattopadhyay and Monalisa Dey

x Contents

Context-Aware Conversational Agent for a Closed Domain Task . . . . . 303Srimoyee Bhattacharyya, Soumi Ray and Monalisa Dey

An End-to-End Approach for Benchmarking Time-Series ModelsUsing Autoencoders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319Abhirup Das, Shramana Roy, Soham Chattopadhyay and Soumik Nandi

Load Balancing for Solving of Supplementary Machines Over Jobswith Max–Min Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329Ranjan Kumar Mondal, Enakshmi Nandi, Payel Rayand Debabrata Sarddar

Load Balancing with Minimum Makespan in Cloud Computing . . . . . . 339Ranjan Kumar Mondal, Payel Ray, Enakshmi Nandiand Debabrata Sarddar

Circulation Desk-less Library Management System UsingRadio-Frequency Identification(RFID) Technologyand Computer Vision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349Abhinaba Basu and Arnab Biswas

Intelligent Water Drops-Based Image Steganography . . . . . . . . . . . . . . 363Pinaki Prasad Guha Neogi, Saptarsi Goswami and Joy Mustafi

Hierarchical Multi-objective Route Optimization for SolvingCarpooling Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377Romit S. Beed, Sunita Sarkar, Arindam Roy and Durba Bhattacharya

Bio-molecular Event Trigger Extraction by Word SenseDisambiguation Based on Supervised Machine Learning UsingWordnet-Based Data Decomposition and Feature Selection . . . . . . . . . . 391Amit Majumder, Asif Ekbal and Sudip Kumar Naskar

Multimodal System for Emotion Recognition Using EEGand Customer Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399Debadrita Panda, Debashis Das Chakladar and Tanmoy Dasgupta

A Modified Dragonfly Algorithm for Real Parameter FunctionOptimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411Sabari Pramanik and S. K. Setua

Design Considerations of a Medical Expert System for DifferentialDiagnosis of Low Back Pain (ESLBP) . . . . . . . . . . . . . . . . . . . . . . . . . . . 425Debarpita Santra, J. K. Mandal, S. K. Basu and Subrata Goswami

Survey: Classification and Reconstruction of ArchaeologicalArtefacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439R. Priyadarshini and A. Soumya

Contents xi

Assessment of Black Tea Using Low-Level Image FeatureExtraction Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453Amitava Akuli, Abhra Pal, Tamal Dey, Gopinath Bej, Amit Santra,Sabyasachi Majumdar and Nabarun Bhattacharyya

A Comparison Among Three Neural Network Models for Silk ContentEstimation from X-Ray Image of Cocoons . . . . . . . . . . . . . . . . . . . . . . . 469Gopinath Bej, Tamal Dey, Abhra Pal, Sabyasachi Majumdar,Amitava Akuli and Nabarun Bhattacharyya

A Comparative Study on Disaster Detection from Social Media ImagesUsing Deep Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 485Arif, Abdullah Omar, Sabah Ashraf, A. K. M. Mahbubur Rahman,M. Ashraful Amin and Amin Ahsan Ali

Rank Based Pixel-Value-Differencing: A Secure SteganographicApproach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 501Debashis Das and Ratan Kumar Basak

Forest Covers Classification of Sundarban on the Basis of FuzzyC-Means Algorithm Using Satellite Images . . . . . . . . . . . . . . . . . . . . . . 515K. Kundu, P. Halder and J. K. Mandal

Artificial Neural Network: An Answer to Right Order Quantity . . . . . . 529Saurav Dey and Debamalya Ghose

Multi-Arc Processor—Harnessing Pseudo-concurrent MultipleInstruction Set Architecture (ISA) Over a Single HardwarePlatform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 535Vishal Narnolia, Uddipto Jana and Tufan Saha

Uncertainty Estimation Using Probabilistic Dependency: An ExtendedRough Set-Based Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 549Indrajit Ghosh

Extension of TOPSIS and VIKOR Method for Decision-MakingProblems with Picture Fuzzy Number . . . . . . . . . . . . . . . . . . . . . . . . . . 563Amalendu Si, Sujit Das and Samarjit Kar

A Study of Gathering of Location-Aware Mobile Robots . . . . . . . . . . . . 579Soumik Banerjee and Sruti Gan Chaudhuri

An Energy Efficient Autonomous Street Lighting System . . . . . . . . . . . . 589Pragna Labani Sikdar and Parag Kumar Guha Thakurta

A Survey on Gathering of Distributed Mobile Agent in DiscreteDomain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 601Himadri Sekhar Mondal and Sruti Gan Chaudhuri

xii Contents

Restaurant Recommendation System Based on Novel ApproachUsing K-Means and Naïve Bayes Classifiers . . . . . . . . . . . . . . . . . . . . . . 609Shreya Joshi and Jigyasu Dubey

Low-Cost Optical Illusion Fluid Display Device . . . . . . . . . . . . . . . . . . . 621Arjun Dutta and Debika Bhattacharyya

Smart Grid Demand-Side Management by Machine Learning . . . . . . . . 633Rebeka Bhattacharyya and Avijit Bhattacharyya

An Integrated Domestic Sensing and Control Systemwith Supervisory Check and Real-Time Data Acquisition . . . . . . . . . . . 645Tanishq Banerjee, Purbayan Chowdhury, Shuvam Ghosal,Suvrojit Kumar Saha and Ranit Bandyopadhayay

Development of a BCI-Based Application Using EEG to AssessAttentional Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 659Sayan Dutta, Tanishq Banerjee, Nilanjana Dutta Royand Bappaditya Chowdhury

Framework for Appraisal of Twenty-Twenty League Players . . . . . . . . 671Sannoy Mitra, Tiyash Patra, Raima Ghosh, Souham Ghosh and Avijit Bose

Analysing Pearl’s Do-Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 681Shreya Guha

A Dynamic Threshold-Based Trust-Oriented Intrusion DetectionSystem in MANET . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 699Khondekar Lutful Hassan, J. K. Mandal and Shukla Mondal

Secured Session Key-Based E-Health: Biometric Blended with SalpSwarm Protocol in Telecare Portals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 713Arindam Sarkar, Joydeep Dey and Sunil Karforma

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 727

Contents xiii

About the Editors

Jyotsna Kumar Mandal is Former Dean of the Faculty of Engineering, Tech-nology and Management, and a Senior Professor at the Department of ComputerScience & Engineering, University of Kalyani, India. Holding a Ph.D. (Eng.)from Jadavpur University, Professor Mandal has co-authored six books: AlgorithmicDesign of Compression Schemes and Correction Techniques—A PracticalApproach; Symmetric Encryption—Algorithms, Analysis and Applications: LowCost-based Security; Steganographic Techniques and Application in DocumentAuthentication—An Algorithmic Approach; Optimization-based Filtering ofRandom Valued Impulses—An Algorithmic Approach; and Artificial NeuralNetwork Guided Secured Communication Techniques: A Practical Approach; allpublished by Lambert Academic Publishing, Germany. He has also authored morethan 350 papers on a wide range of topics in international journals and proceedings.His areas of research include coding theory, data and network security, remote sensingand GIS-based applications, data compression, error correction, visual cryptographyand steganography, distributed and sharedmemory, and parallel programming.He is aFellow of the Institution of Electronics and Telecommunication Engineers, and amember of the IEEE, ACM, and Computer Society of India.

Somnath Mukhopadhyay is currently an Assistant Professor at the Department ofComputer Science&Engineering, AssamUniversity, Silchar, India. He completed hisM.Tech. and Ph.D. in Computer Science & Engineering at the University of Kalyani,India, in 2011 and 2015, respectively. He has co-authored one book and edited fiveothers. He has published over twenty papers in various international journals andconference proceedings, as well as three book chapters in edited volumes. His researchinterests include digital image processing, computational intelligence, and patternrecognition. He is a member of the IEEE and IEEE Young Professionals – KolkataSection, life member of the Computer Society of India, and currently the ComputerSociety of India’s regional student coordinator (RSC) for Region II.

xv

Detection of Leukemia in Blood SamplesApplying Image Processing Using a NovelEdge Detection Method

Megha Dutta, Srija Karmakar, Prithaj Banerjee and Rit Ghatak

Abstract In current times, pathologists visually inspect blood cell images underthe microscope for the purpose of identifying blood disorders. Identified blood dis-orders are classified into several blood diseases. Our work aims at studying anddesigning a model (framework) for the detection of leukemia (blood cancer) and itstypes using microscopic blood sample images and analyzing them for diagnosis at anearlier stage. Earlier, hematologists used to visually inspect the microscopic images,therebymaking the diagnosis process error-prone and time-consuming. However, theuse of newly developed automatic image processing systemsmanages to successfullyovercome most of the visual inspection related drawbacks. The early diagnosis ofblood cancer will greatly aid in better treatment. In this process, the acquired datasetimages are taken as inputs and the images are sent through different image process-ing techniques such as image enhancement (preprocessing), segmentation, featureextraction and classification. The method proposed is applied to a large number ofimages of varying quality and is found to provide satisfactory results.

Keywords Leukemia · Image processing · Segmentation · Feature extraction

1 Introduction

Leukemia is a hematologic cancer which develops in the blood tissue and causesrapid generation of abnormal (large) shaped and immature WBC’s. The primarycause behind leukemia is the production of a huge number of abnormal lymphocytes(WBCs) by the bone marrow. There is always a healthy balance between red bloodcells, white blood cells and platelets in human blood. The onset of leukemia, however,disrupts this balance. Damaged or old blood cells always die, only to be replaced bynew ones [1]. The bone marrow of a blood cancer patient produces much more large-sized WBCs compared to that of a normal person’s bone marrow. Unlike normalblood cells, they do not get replaced in due time, resulting in a huge abnormal WBCcount. Normal functions of the different components of blood are affected due to this

M. Dutta (B) · S. Karmakar · P. Banerjee · R. GhatakDepartment of CSE, Institute of Engineering & Management, Kolkata, India

© Springer Nature Singapore Pte Ltd. 2020J. K. Mandal and S. Mukhopadhyay (eds.), Proceedings of the GlobalAI Congress 2019, Advances in Intelligent Systems and Computing 1112,https://doi.org/10.1007/978-981-15-2188-1_1

1

2 M. Dutta et al.

Fig. 1 Example showingdifferent components of ablood stem cell (ImageCourtesy [26])

imbalance. Based on the speed of infection development and the severity, leukemiacan be classified as either acute (rapid growth and fatal in a short time) or chronic(at nascent treatable stage, not that severe yet). Further classifications can also bemade [2]. Acute lymphocytic leukemia [3] will be our main focus of discussion inthis paper, as we have succeeded in acquiring dataset sample images for the same.From a biological standpoint, the most widely practiced blood cancer classificationby structural analysis is the FAB procedure [4]. An improved variant of the samecalled immunologic classification [5] is more common presently. Morphologically,healthyWBCs (lymphocytes) have a regular shape, a headphone-shaped nucleus andabundant cytoplasm while infected ones (lymphoblasts) are characterized by scantycytoplasm and irregular-shaped nucleus. FAB analyzes ALL lymphoblasts into threetypes, namely L1, L2 and L3. L1 has a regular round nucleus with negligible clefting,scanty and vacuole-less cytoplasm. Clefted, irregular, multiple nuclei with variablecytoplasm are present in L2. Prominent distinguishable vacuoles are present in L3(Fig. 1).

Statistical data shows that in the last few years, leukemia has been one of theprimary causes of death worldwide, irrespective of gender. A delay in detectionmakes it highly unlikely for ALL patients to survive. This makes early and correctdiagnosis all the more important. A traditional setup places heavy importance onthe pathologist for accuracy in diagnosis. Bone marrow aspiration, medical historycheckup, CBCand cytogenic analysis are some of the commonly usedmethods in thisregard. However, manual detection methods often yield inconsistent and inaccurateresults. To remedy this, there has been a lot of research in terms of applying imageprocessing to this field. In addition, image processing enhanced techniques can alsohelp in detecting how far the disease has progressed (the stage it is currently in).

The contributions of the paper are as follows: Firstly, we develop a novel seg-mentation or rather edge detection approach which provides more accurate resultscompared to existing approaches. Then we follow the traditional steps of an imageprocessing algorithm and apply our own method in the algorithm implementation.

Detection of Leukemia in Blood Samples Applying Image … 3

The result is the detection of the input sample cell to be either healthy or leukemiainfected (further stage classification into chronic and acute is done).

The organization of the rest of the paper is as follows:In Sect. 1.1, we discuss some recently developed related works for leukemia

detecting using image processing. We also survey the conventional steps in the algo-rithm related to our proposed framework namely segmentation, thresholding, etc. InSect. 2, we developed our own edge detection method. In Sect. 3, we discuss the pro-posed framework and a unique algorithm to perform the task of detection. Section 4deals with the results obtained in terms of image outputs and the result accuracy interms of other methods. Finally, Sect. 5 provides a conclusion to our work, brieflytouching upon its future scope.

1.1 Related Work

The introduction already focused on the biological aspects of leukemia and con-ventional nonimage processing methods related to leukemia detection in blood cellsamples. In our literature survey, the focus will be more on the image processing-oriented leukemia detection methods. The steps of the general workflow and most ofthe important works done to explain and improve these methods are also included.The survey will also encompass most of the novel methods or algorithms tried toimprove the steps of the traditional processes or in some cases, the process as awhole.

References [1–3] will elaborate on the biology and morphology of blood cancer.Bennett et al. [4] and Biondi et al. [5] are basically nonimage processing relatedtechniques used for leukemia detection.

Coming to the application of image processing enhanced methods in leukemiadetection, thresholding is a very vital step. Many unique works have been designedwith a view to modify or improve this step to suit our needs. In [6], Minal D. Joshi,et al. have suggested automatic Otsu’s thresholding, along with image enhancement,for segmentation of white blood cells. Lymphocytes and lymphoblasts are distin-guished using k-NN classifier. In [7], Ghosh et al. applied the concept of fuzzydivergence to find out a threshold for leukocyte segmentation. Various functions likeCauchy, Gaussian and Gamma have been compared for their benefits and drawbacksand the most suitable ones have been used further. In [8], Dorini et al. have proposeda nucleus extraction scheme using watershed transform, which is further based onthe image forest transform. The method, however, gives good results only for roundcytoplasm.

Thresholding is closely related to segmentation. The countless number of literaryworks have been done regarding image segmentation.

In [9], G. EvelinSuji, Y. V. S. Lakshmi, G. Wiselin Jiji et al. have provided acomparative study of different image segmentation algorithms.

4 M. Dutta et al.

In [10], Lim Huey Nee et al. have proposed methods like morphological opera-tions, watershed transformations and thresholding for the purpose of cell segmen-tation. The method provided satisfactory segmentation results for a sample of 50images.

In [11], Monica Madhukar, Sos Agaian and Anthony T. Chronopoulos et al. tryto segment AML blood smears by applying the popular KMeans algorithm.

In [12], N. H. Abdul Halim et al. have introduced segmentation and contraststretching by taking into consideration the HIS color space. The same thresholdvalue is used for enhancement purposes.

Feature extraction is also a very important facet of the entire detection process.The accuracy of this step often determines the accuracy of the actual result.

In [13], Fauziah Kasmin et al. feature changes act as classifier inputs. Usually,features such as color, texture and geometrical shapes are considered.

A feasible method to detect ALL has been proposed in [14] by Lalit Mohan Sainiand Romel Bhattacharjee et al., which uses features such as circularity, perimeter, acell’s form factor and area for classification purposes.

Speaking of classification, it is the final andmost vital step in an image processingrelated detection setup. There are numerous works on both existing as well as novelclassification concepts.

Reta et al. [15] proposed a system for the sole purpose of white blood cell classifi-cation. A complete classification system forALL andAMLdetection using grayscalelevel images only has been proposed in [16].

Combining the steps and by performing slightmodifications to themor by improv-ing them greatly, entirely new algorithms and methods have been suggested aswell.

In [17], Hossein Ghayoumi Zadeh et al. have worked on an image analy-sis approach for enhancement, smoothing, segmentation, automatic detection andclassification of infected cells from normal cells.

In [18], Jaya Sharma, Mashiat Fatma et al. propose application of convolutedneural networks (CNN) to classify healthy and infected cells. It is quite a complexmethod uniting the fields of computer vision and image processing, and though it isvery difficult to execute, the results are great.

In [19], Ruggero Donida Labati et al. have used designed a new blood samplepublic dataset for the purpose of comparing various existing classification and seg-mentation algorithms and selecting the best choice out of them. Each dataset imageprovides figures of merit to compare various algorithm performances.

Our entire work is based on the ALL-IDB image database, which is freelydownloadable from [20].

2 Main Contribution

While performing our algorithm for infectedWBC cell detection, the main challengewe faced was separating individual WBC cells from clusters. The existing edge

Detection of Leukemia in Blood Samples Applying Image … 5

detection methods like Roberts [21], Prewitt [22], Sobel [23, 24] or Canny [25]are proposed for transition detection in images. These methods do not solve thisproblem which led us to device a new modified edge detection algorithm that solvedour problem (Fig. 4f).

2.1 Novel Edge Detection Method

Previous edge detection methods [21–25] when applied on our image could notdetect the edge separating two WBCs. In [23, 24], a 3 × 3 matrix was used as theestimate vector for directional derivativewhose elements are the difference of densityto neighbor. In contrast to previous methods, we used a 5 × 5 matrix which satisfiedour need. Moreover, our edge detector is specially designed to meet our problem, soit works better specific to our image than any existing edge detection methods. Inour proposed methods, we use two 5 × 5 masks along with both X and Y directionfor both horizontal and vertical edges.

This mask when kernel convoluted to our image matrix marks the sharp changesin our image and generates an edge matrix.

Since our image of concern is a black and white image converted to its grayscale,the image pixel values are either 0 (black) or 255 (white). Hence, our proposedmethod does not require any complex directional derivative calculation for the maskcell values. So, our mask cell consists of values −1, 0, 1.

As in Fig. 2,

Ii x = Ix ∗ Mx (1)

Similarly,

Iiy = Iy ∗ My (2)

Here, * is the kernel convolution function. It multiplies each cell of the mask withits corresponding cell value of the image pixel and the summation of all such productis assigned to the central cell of the mask (I i).

Themask in both the directions can give absolutemaskmagnitudewhen combinedtogether as per the equation.

|M | =√M2

x + M2y (3)

I i detects the edge in the image if it encounters a sharp change of grayscale value as inFig. 4f. Thus, the mask (M) slides gradually throughout the image-producing edges.The main problem with other edge detection method is that it cannot separate out

6 M. Dutta et al.

Fig. 2 Mx matrix (top left), My matrix (top right), snapshot of Mx on our image

the blobs, so we have adopted filledgegaps1 algorithm in our edge detection methodwhich in turn fills the gaps in between two adjacentWBCs of our experiment resultingan edge. Algorithm for our proposed edge detection method is as given Sect. 2.1.

2.2 Algorithm for Our Novel Edge Detection Method

Input: Image generated from our image processing step.Output: Edges detected using our novel method.

Step 1: Masking with M is done to the input imageStep 2: The 5 × 5 mask is run through the whole imageStep 3: Generated image is then passed to filledgegaps algorithm to fill all the cleftsStep 4: The image from Step 3 is our output edge image.

1https://www.peterkovesi.com/matlabfns/LineSegments/filledgegaps.m.

Detection of Leukemia in Blood Samples Applying Image … 7

3 Proposed Framework

A block diagram of the proposed method is presented in Fig. 3 and an algorithm forthe same is given below. Before describing the algorithm in a stepwise manner, letus give a brief overview about the steps involved:

The first step in this process is image acquisition. In this step, blood image fromslides will be obtained either from online sources or from a nearby hospital witheffective magnification. The obtained images are called datasets. It is necessary thatthe process must yield satisfactory results over an entire dataset of images in orderto certify it as a feasible method. It is recommended that the acquired images be ofhigh pixel quality so that they can yield better results.

The next step is image enhancement, also referred to as image preprocessing.While acquiring dataset images, due to errors in staining, the images are likelyto be disturbed by noise. The noise might blur the region of interest, leading toimproper segmentation. To remove this noise and to improve quality of image, imagepreprocessing (enhancement) is required.

In the noise removal process, to ensure amore compact and better study, allWBCsat the image edges and all other non-leukocyte components (RBCs, platelets, etc.)are removed.

Image enhancement is followed by image segmentation. Segmentation refers tothe process of distinguishing colors, objects, patterns or textures within an imageitself. The main purpose of segmentation in image processing is to obtain a particularpart or objects from the entire image so that it can be utilized for subsequent steps. Thecorrectness of the subsequent steps namely feature extraction and image classificationdepends on the quality of the segmented output.

Segmentation focuses on the nucleus of WBC only as cytoplasm is scanty ininfected cell images. Only lymphocytes and myelocytes are examined to determinewhether they are blast cells or not.

Our approach has considered different existing segmentation techniques [9] andapplied them to both original images and enhanced (noise removed) images to arriveat the most suitable choice. This is arguably the most important process of the entirealgorithm and needs to be performed as neatly as possible for accurate end results.

Fig. 3 Flowchart of our framework

8 M. Dutta et al.

In fact, our paper’s main contribution lies in proposing a novel segmentation (edgedetection) approach which provides greater accuracy when compared to the existingsegmentation algorithms or methods. Then we move to feature extraction based ongeometrical as well as texture features and postextraction we classify the image to beeither that of an infected or a healthy cell. Lastly, based on study of morphologicalfeatures especially cytoplasm, we classify the cancer to be either chronic or acute.

The algorithm is given in two parts. Part 1 basically deals with acquiring theimage sample, breaking it down into its three component color channels, performingthresholding separately in each of the channels and expressing it as the sum of thecolor planes. Part 2 deals with complementing the image, processing it and thenapplying edge detection (the improved novel method) on the image. Then featureextraction is carried out and the image is classified into healthy or infected (furtherstage classified into acute or chronic).

A detailed step by step overview is given below:Part 1:

Input: A colored blood sample image from datasetOutput: Image expressed as the sum of component planes

• The colored blood slide image is taken as system input. The cell might be healthyor infected; we do not know it yet. Our purpose is to identify whether there areblast cells in the original image and thereby determine if it is healthy or infected.

• Conventional algorithms usually convert the color image into grayscale image inthe next step. However, we will not be doing the same.

• Instead, we break down the color image in three color planes namely the red plane,the green plane and the blue plane.

• Then we perform thresholding separately in each of the three channels, and onlyafter it is done, the image is converted to its black and white forms.

• We then express the image as the sum of all color planes by using the ANDoperation.

Part 2:

Input: Output image from Part 1Output: Image which has been classified as healthy or infected and further stageclassification is done.

• The entire image is complemented (white parts turn black and vice versa). Also,the holes in the image are filled with unwanted small pixels.

• We perform edge detection (a segmentation approach) on the unprocessed image.Generally, state of the art methods are applied but our own research has foundour own designed edge detection method to be more accurate than those. So, weuse it for edge detection. The underlying purpose of using edge detection on anunprocessed image is to elucidate that applying processing is indeed advantageousin this context.

• Small unwanted pixel groups are removed using morphological operations suchas morphological erosion.

Detection of Leukemia in Blood Samples Applying Image … 9

• Again, edge detection is used, this time on the processed image.• The geometrical features such as area and perimeter of segmented cells and texturefeatures are computed. Based on the features extracted in the last step, the cell canbe classified as a healthy cell or infected cell.

• For a healthy cell, the image background will be entirely black depicting that thereare no blast cells in the image. However, for an infected cell, there will be thepresence of white WBC outlines on the black backdrop, depicting the presence ofblast cells and thereby, presence of cancer. In fact, the number of white outlinesis nothing but the number of blast cells in the sample image. Flowchart of thealgorithm is shown in Fig. 3.

• Further stage classification of the detected cancer cell is performed to know theacuteness of the cancer cell.

The entire algorithmwill bemuch easier to understandwhen visual representationof the obtained results (for both cases) is provided in the result and analysis section.

4 Experiment and Result Analysis

4.1 Datasets

A Canon G5 Powershot camera, aided by an optical microscope, has been used tocapture the images for our working dataset. The microscope is magnified withina 300–500 range to capture the images in different zoom angles. Each image has a2592 * 1944 resolution. The image format is JPGwith a color depth of 24 bits. A totalof 108 images clicked during 2005 are present in the dataset. A collection of 39,000blood elements is available in ALL-IDB1 where expert cancer doctors have markedthe lymphocytes. A rough estimate of around 510 candidates infected WBC’s arepresent. The dataset is freely downloadable from [20]. The notation is XXX_Y.jpgwhere a triple-digit integer counter depicts XXX and a Boolean variable (either 0or 1) depicts Y. For healthy lymphocytes, Y takes a value of 0, and for infectedlymphocytes (lymphoblasts), Y takes a value of 1. Thus, it is easy to classify theimages as healthy or infected just by looking at the Boolean Y value. The centroidsof the lymphoblasts (if any) are documented in a text file IMXXX_Y.xyc, which islinked to the main jpg file.

4.2 Results Related to Own Edge Detection Method

There aremanywell-known states of art edge detectionmethodswhich yield accurateresults for any image segmentation process. However, when we tried to apply them,for some images in our dataset, we encountered a problem Fig. 4b–f. In the originalprocessed image shown in Fig. 4a, there is a dumbbell-shaped structure in the middle

10 M. Dutta et al.

Fig. 4 a Original processed image. b Roberts on original processed image. c Prewitt on originalprocessed image. d Sobel on original processed image. e Canny on original processed image.f Proposed method on original processed image

portion of the image sample. To the naked eye, it appears there is a single large entity(WBC)but in reality, there are seven separate lymphocytes. The only reason it appearsto be a single entity is that the separate WBCs are clustered together with no wellvisible partition (cleft) or blobs separating their edges. When we apply the existingwell-known methods on this particular image, they fail to detect the edge outlinesof the separate cells and consider the entire collection of seven leukocytes to bea single one. As a result, they fail to trace the separate cell outlines and the edgedetection process becomes inaccurate and to a large extent, incorrect. This leadsto the necessity of a new edge detection method that can overcome this particulardrawback. The main objective of our novel method should be to detect the separatecells and perform separate outline detection on each of them.

Figure 4a–e all display the results of applying various well-known edge detectionalgorithms on a processed image. However, it clearly shows in the results that the

Detection of Leukemia in Blood Samples Applying Image … 11

methods have failed in treating the collection of sevenWBCs as separate entities andhave considered only the outer layer of the collection for outline detection. Figure 4fdisplays the result of our own approach. It is clearly visible that our algorithm hasbeen successful in considering each leukocyte as a separate entity and edge outlineshave been computed for each of them, unlike other methods.

This observation leads to the conclusion that our novel edge detection methodprovides better accuracy than existing methods in this regard.

4.3 Results Related to Cancer Cell Detection for an InfectedCell

In Fig. 5b, the acquired image is broken up or segregated into its three constituentcolor planes. In Fig. 5c, thresholding is separately performed for each plane and theimage is converted to grayscale form.

A resulting image in Fig. 6A with white outlines on a dark background looksbetter than the other way round. So, for clarity purposes, the image is complementedand image holes are filled with unwanted small pixels. In Fig. 6B, morphological

Fig. 5 a Original infected cell. b Image in three color planes. c Thresholding in three planes andsummation

12 M. Dutta et al.

Fig. 6 A Complemented image with holes filled along with unwanted small pixels. B Image afterremoving unwanted pixels. C Edge detection on processed image using own method. D Featureextraction and eliminating healthyWBCs.E Identification and count of blast cells in original imageand stage classification

operations are used to remove unwanted pixels. Our earlier experimental result hasalready established that using our own edge detectionmethod instead of conventionalmethods will give us better accuracy. So, we apply our own edge detection approachon the processed image (as in Fig. 6C, D).

Since the input imagewas that of a blood cancer infected cell, the feature extractionresult picture successfully identified the presence of blast cells in the original image.It also detected the number of blast cells in the image. In fact, the number of white

Detection of Leukemia in Blood Samples Applying Image … 13

Fig. 7 a Original healthy image cell. b Image on three color planes. c Thresholding in three planesand summation. d Complemented image with holes filled with unwanted small pixels

outlines is nothing but the number of blast cells in the sample image (which is five inthis case). Figure 6E(b) depicts only the presence of lymphoblasts and Fig. 6E(c–e)represents Stage 1 blood cancer, Stage 2 blood cancer and Stage 3 blood cancer,respectively.

4.4 Results Related to Cancer Detection in Healthy Cells

The same sequence of steps that we applied to the infected cell is applied to thehealthy image sample as well (Fig. 7).

Since the input was a healthy cell, after feature extraction and WBC elimination,there were zero blast cells found in Fig. 8d. This goes on to show that the algorithmis capable of producing accurate results.

4.5 Accuracy Calculation in Tabular Form

In the context of our work, accuracy calculation is done by processing some tech-nical parameters. We can either employ a cell test benchmark approach (test returnspositive if the cell in question is infected) or an image-level benchmark (test returnspositive if image in question has even a single infected cell). Some term elementsthat need to be discussed in this regard.

14 M. Dutta et al.

Fig. 8 a Image after removing unwanted pixel. b Edge detection on processed image using ownmethod. c Feature extraction and eliminating healthyWBCs.d Identification of blast cells in originalimage

The number of elements which our test correctly classifies as positive is said tobe true positive (tp) and the elements which the test incorrectly classifies as positiveare called false positive (f p).

Similarly, the number of elements which our test correctly classifies as negativeis said to be true negative (tn) and the elements which the test incorrectly classifiesas negative are called false negative (f n).

The probability that every ALL element has been classified correctly is referredto as sensitivity. In contrast, specificity is defined by the probability of correct clas-sification of non-ALL elements. Mathematically, these terms can be representedas:

Sensitivity = tp/(tp + fn

)(4)

Specificity = tn/(tn + fp

)(5)

All these terms are collectively referred to as figures of merit and can be used foraccurate calculation for both cell and image tests.

Another mode of calculation using the same parameters is by using the conceptsof precision and recall. Precision refers to the percentage of our relevant results.Recall, on the other hand, refers to the percentage of total relevant results that theused algorithm successfully classifies.

Detection of Leukemia in Blood Samples Applying Image … 15

Table 1 Accuracy comparison

Method Accuracy (in %)

K-means clustering 73.1

Watershed transform 71.6

Edge detection using histogram equalizing and linear contrast stretching 74.5

Proposed novel method 86.4

Precision = tp/(tp + fp) (6)

Recall = tp/(tp + fn

)(7)

Accuracy = (tp + tn

)/(tp + tn + fp + fn

)(8)

Taking help from these formulae, we compare the accuracies of results obtainedfrom different methods as well as our own proposed method.

The results from Table 1 clearly show that our own proposed method is moreaccurate than other methods in this regard.

5 Conclusion

It has been stated several times, in this paper, that the possibility of successfultreatment of leukemia is highly dependent on its early detection and diagnosis.

The manual inspection process has been found to be tiring and error-prone andthereby image processing algorithms have been employed to speed up the processand improve accuracy at the same time. We have performed a thorough study ofthe existing algorithms; the processes involved and even designed our own uniqueedge detection algorithm to counter a problem that we faced during processing. Thealgorithm has been found to obtain highly accurate results, thereby assuring that ourproposed method is indeed feasible. However, several improvements can be made infuture to this work (such as improving dataset quality and extending the process tocancer types other than ALL), in order to allow the work to make a greater impact.

References

1. National Cancer Institute: http://www.cancer.gov/cancertopics/wyntk/leukemia. Last accessed3 Oct 2013

2. Through the Microscope: Blood Cells—Life’s Blood. http://www.wadsworth.org/chemheme/heme. Last accessed 3 Oct 2011

3. https://cancer.org. Last accessed

16 M. Dutta et al.

4. Bennett, J.M., Catovsky, D., Daniel, Marie-Theregse, Flandrin, G., Galton, D.A.G., Gralnick,H.R., Sultan, C.: Proposals for the classification of the acute leukemias French-American-British (FAB) co-operative group. Br. J. Haematol. 33(4), 451–458 (1976)

5. Biondi, A., Cimino, G., Pieters, R., Pui, C.-H.: Biological and therapeutic aspects of infantleukemia. Blood 96(1), 24–33 (2000)

6. Joshi, M.D., Karode, A.H., Suralkar, S.R.: White blood cells segmentation and classificationto detect acute leukemia. Int. J. Emerg. Trends Technol. Comput. Sci. (2013)

7. Ghosh, M., Das, D., Chakraborty, C., Ray, A.K.: Automated leukocyte recognition using fuzzydivergence. Micron 41(7), 84 (2010)

8. Dorini, L.B., Minetto, R., Leite, N.J.: White blood cell segmentation using morphologicaloperators and scale space analysis. In: Proceedings of the Brazilian Symposium on ComputerGraphics and Image Processing, pp. 294–304, Oct 2007

9. Evelin Suji, G., Lakshmi, Y.V.S., Wiselin Jiji, G.: Comparitive study on image segmentationalgorithms. IJACR 3(3, 11) (2013)

10. Nee, L.H., Mashor, M.Y., Hassan, R.: White blood cell segmentation for acute leukemia bonemarrow images. In: International Conference on Biomedical Engineering (ICoBE), Penang,Malaysia, 27–28 Feb 2012

11. Agaian, S., Senior Member, IEEE, Madhukar, M., Chronopoulos, A.T., Senior Member, IEEE:Automated screening system for acute myelogenous leukemia detection in blood microscopicimage. IEEE Syst. J. (2014)

12. Abd Halim, N.H., Mashor, M.Y., Abdul Nasir, A.S., Mokhtar, N.R., Rosline, H.: Nucleussegmentation technique for acute leukemia. In: IEEE 7th International Colloquium on SignalProcessing and its Applications (2011)

13. Kasmin, F., Prabuwonw, A.S., Abdullah, A.: Detection of leukemia in human blood samplebased on microscopic images: a study. J. Theor. Appl. Inf. Technol. (2012)

14. Bhattacharjee, R., Saini, L.M., Robust technique for the detection of acute lymphoblasticleukemia. In: 2015 IEEE Power, Communication and Information Technology Conference(PCITC) Siksha O Anusandhan University, Bhubaneswar, India

15. Reta, C., Robles, L.A., Gonzalez, J.A., Diaz, R., Guichard, J.S.: Segmentation of bone marrowcell images for morphological classification of acute leukemia. In: Proceedings of the Twenty-First International Florida Artificial Intelligence Research Society Conference (2010)

16. Taylor, K.B., Schorr, J.B.: Blood 4 (1978)17. Zadeh, H.G., Janianpour, S., Haddadnia, J.: Recognition and classification of the cancer cells

by using image processing and LabVIEW. Int. J. Theory Eng. 5(1) (2013)18. Fatma, M., Sharma, J.: Identification and classification of acute leukemia using neural net-

work. In: 2014 International Conference on Medical Imaging, m-Health and EmergingCommunication Systems (MedCom)

19. Labati, R.D., Piuri, V., Scotti, F.: A1I-Idb: the acute lymphoblastic Leukemia image databasefor image processing (2011)

20. https://homes.di.unimi.it/scotti/all/21. Aybar, E.: Sobel Edge Detection Method For Matlab. Anadolu University, Porsuk Vocational

School, 26410 Eskisehir22. Vincent, O.R., Folorunso, O.: A descriptive algorithm for sobel image edge detection. In:

Proceedings of Informing Science & IT Education Conference (InSITE) (2009)23. Sobel, I.: An Isotropic 3× 3GradientOperator,MachineVision for Three-Dimensional Scenes,

pp. 376–379. Freeman, H., Academic Pres, New York (1990)24. Prewitt, S., Scharr gradient 5 × 5 convolution matrices Guennadi (Henry) Levkine Email:

hlevkin at yahoo.com Vancouver, Canada. First draft, Feb 2011, Second Draft, June 201225. Gao, W., Yang, L.: An Improved Sobel Edge Detection. 978-14244-5540-9/10/$26.00 ©2010

IEEE26. TereseWinslowLLC:Medical andScientific Illustration. https://www.teresewinslow.com.Last

accessed 3 Jan 2020