Proceedings of international conference on smart computing and cyber security strategic foresight se

Proceedings of International Conference on Smart Computing and Cyber Security Strategic Foresight Security Challenges and Innovation 1st edition by Prasant Kumar Pattnaik, Mangal Sain, Ahmed A AlAbsi ISBN 981157992X 978-9811579929 pdf https://ebookball.com/product/proceedings-of-internationaldownload conference-on-smart-computing-and-cyber-security-strategicforesight-security-challenges-and-innovation-1st-edition-byprasant-kumar-pattnaik-mangal-sain-ahmed-a-alabsi-is/

Explore and download more ebooks or textbooks at ebookball.com

Get Your Digital Files Instantly: PDF, ePub, MOBI and More Quick Digital Downloads: PDF, ePub, MOBI and Other Formats

Proceedings of International Conference on Smart Computing and Cyber Security Strategic Foresight Security Challenges and Innovation 1st edition by Prasant Kumar Pattnaik, Mangal Sain, Ahmed Al Absi, Pardeep Kumar 9811579903 9789811579905 https://ebookball.com/product/proceedings-of-internationalconference-on-smart-computing-and-cyber-security-strategicforesight-security-challenges-and-innovation-1st-edition-byprasant-kumar-pattnaik-mangal-sain-ahmed-al-absi-par/

Proceedings of Third International Conference on Computing Communications and Cybes Security 1st edition by Pradeep Kumar Singh, Sławomir Wierzchoń, Sudeep Tanwar, Joel Rodrigues, Maria Ganzha 9819747996 9789819747993 https://ebookball.com/product/proceedings-of-third-internationalconference-on-computing-communications-and-cybes-security-1stedition-by-pradeep-kumar-singh-sa-awomir-wierzchoa-sudeep-tanwarjoel-rodrigues-maria-ganzha-9819747996/

Advanced Computing Networking and Informatics Volume 1 Advanced Computing and Informatics Proceedings of the Second International Conference on Advanced Computing, Networking and Informatics 1st Edition by Malay Kumar Kundu, Durga Prasad Mohapatra, Amit Konar, Aruna Chakraborty ISBN 3319073532 9783319073538 https://ebookball.com/product/advanced-computing-networking-andinformatics-volume-1-advanced-computing-and-informaticsproceedings-of-the-second-international-conference-on-advancedcomputing-networking-and-informatics-1st-editio/

(Ebook PDF) Foundations and Trends in Smart Learning Proceedings of 2019 International Conference on Smart Learning Environments 1st edition by Maiga Chang, Elvira Popescu 9811369089 9789811369087 full chapters https://ebookball.com/product/ebook-pdf-foundations-and-trendsin-smart-learning-proceedings-of-2019-international-conferenceon-smart-learning-environments-1st-edition-by-maiga-chang-elvirapopescu-9811369089-9789811369087-full-c/

International Conference on Cyber Security, Privacy and Networking 1st edition by Nadia Nedjah, Gregorio MartÃ-nez Pérez, Gupta 3031220188 9783031220180 https://ebookball.com/product/international-conference-on-cybersecurity-privacy-and-networking-1st-edition-by-nadia-nedjahgregorio-martanez-pa-c-rez-gupta-3031220188-9783031220180-20134/

Cyber Defense Mechanisms Security Privacy and Challenges 1st edition by Gautam Kumar, Dinesh Kumar Saini, Nguyen Ha Huy Cuong ISBN 0367540967 9780367540968 https://ebookball.com/product/cyber-defense-mechanisms-securityprivacy-and-challenges-1st-edition-by-gautam-kumar-dinesh-kumarsaini-nguyen-ha-huy-cuong-isbn-0367540967-9780367540968-17068/

Intelligent Computing Networking and Informatics Proceedings of the International Conference on Advanced Computing Networking and Informatics India June 2013 1st Edition by Durga Prasad Mohapatra, Srikanta Patnaik ISBN https://ebookball.com/product/intelligent-computing-networkingand-informatics-proceedings-of-the-international-conference-onadvanced-computing-networking-and-informatics-indiajune-2013-1st-edition-by-durga-prasad-mohapatra-srik/

International Conference on Cyber Security Privacy and Networking ICSPN 2022 1st edition by Nadia Nedjah,Gregorio MartÃ-nez Pérez 9783031220180 3031220188 https://ebookball.com/product/international-conference-on-cybersecurity-privacy-and-networking-icspn-2022-1st-edition-by-nadianedjah-gregorio-martanez-pa-c-rez-9783031220180-3031220188-25752/

Machine Learning Blockchain and Cyber Security in Smart Environments, Applications and Challenges 1st edition by Sarvesh Tanwar, Sumit Badotra, Ajay Rana 1000623912 9781000623918 https://ebookball.com/product/machine-learning-blockchain-andcyber-security-in-smart-environments-applications-andchallenges-1st-edition-by-sarvesh-tanwar-sumit-badotra-ajayrana-1000623912-9781000623918-20020/

Lecture Notes in Networks and Systems 149

Prasant Kumar Pattnaik Mangal Sain Ahmed A. Al-Absi Pardeep Kumar Editors

Proceedings of International Conference on Smart Computing and Cyber Security Strategic Foresight, Security Challenges and Innovation (SMARTCYBER 2020)

Lecture Notes in Networks and Systems Volume 149

Series Editor Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Advisory Editors Fernando Gomide, Department of Computer Engineering and Automation—DCA, School of Electrical and Computer Engineering—FEEC, University of Campinas— UNICAMP, São Paulo, Brazil Okyay Kaynak, Department of Electrical and Electronic Engineering, Bogazici University, Istanbul, Turkey Derong Liu, Department of Electrical and Computer Engineering, University of Illinois at Chicago, Chicago, USA; Institute of Automation, Chinese Academy of Sciences, Beijing, China Witold Pedrycz, Department of Electrical and Computer Engineering, University of Alberta, Alberta, Canada; Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Marios M. Polycarpou, Department of Electrical and Computer Engineering, KIOS Research Center for Intelligent Systems and Networks, University of Cyprus, Nicosia, Cyprus Imre J. Rudas, Óbuda University, Budapest, Hungary Jun Wang, Department of Computer Science, City University of Hong Kong, Kowloon, Hong Kong

The series “Lecture Notes in Networks and Systems” publishes the latest developments in Networks and Systems—quickly, informally and with high quality. Original research reported in proceedings and post-proceedings represents the core of LNNS. Volumes published in LNNS embrace all aspects and subfields of, as well as new challenges in, Networks and Systems. The series contains proceedings and edited volumes in systems and networks, spanning the areas of Cyber-Physical Systems, Autonomous Systems, Sensor Networks, Control Systems, Energy Systems, Automotive Systems, Biological Systems, Vehicular Networking and Connected Vehicles, Aerospace Systems, Automation, Manufacturing, Smart Grids, Nonlinear Systems, Power Systems, Robotics, Social Systems, Economic Systems and other. Of particular value to both the contributors and the readership are the short publication timeframe and the world-wide distribution and exposure which enable both a wide and rapid dissemination of research output. The series covers the theory, applications, and perspectives on the state of the art and future developments relevant to systems and networks, decision making, control, complex processes and related areas, as embedded in the fields of interdisciplinary and applied sciences, engineering, computer science, physics, economics, social, and life sciences, as well as the paradigms and methodologies behind them. Indexed by SCOPUS, INSPEC, WTI Frankfurt eG, zbMATH, SCImago. All books published in the series are submitted for consideration in Web of Science.

More information about this series at http://www.springer.com/series/15179

Prasant Kumar Pattnaik Mangal Sain Ahmed A. Al-Absi Pardeep Kumar •

•

•

Editors

Proceedings of International Conference on Smart Computing and Cyber Security Strategic Foresight, Security Challenges and Innovation (SMARTCYBER 2020)

123

Editors Prasant Kumar Pattnaik School of Computer Engineering Kalinga Institute of Industrial Technology KIIT Deemed to be University Bhubaneswar, India

Mangal Sain Division of Information and Communication Engineering Dongseo University Busan, Korea (Republic of)

Ahmed A. Al-Absi Department of Smart Computing Kyungdong University Global Campus Gangwondo, Korea (Republic of)

Pardeep Kumar Department of Computer Science Swansea University, Bay Campus Swansea, UK

ISSN 2367-3370 ISSN 2367-3389 (electronic) Lecture Notes in Networks and Systems ISBN 978-981-15-7989-9 ISBN 978-981-15-7990-5 (eBook) https://doi.org/10.1007/978-981-15-7990-5 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of 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 transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the 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 this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to 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

The 1st International Conference on Smart Computing and Cyber Security— Strategic Foresight, Security Challenges and Innovation (SMARTCYBER 2020), took place in Kyungdong University Global Campus, Gosung, Gangwondo, South Korea, during July 7–8, 2020. It was hosted by the Department of Smart Computing, Kyungdong University, Global Campus, South Korea. The SMARTCYBER is a premier international open forum for scientists, researchers and technocrats in academia as well as in industries from different parts of the world to present, interact and exchange the state of the art of concepts, prototypes, innovative research ideas in several diversified fields. The primary focus of the conference is to foster new and original research ideas and results in the five board tracks: smart computing concepts, models, algorithms, and applications, smart embedded systems, bio-Inspired models in information processing, technology, and security. This is an exciting and emerging interdisciplinary area in which a wide range of theory and methodologies are being investigated and developed to tackle complex and challenging real-world problems. The conference includes invited keynote talks and oral paper presentations from both academia and industry to initiate and ignite our young minds in the meadow of momentous research and thereby enrich their existing knowledge. SMARTCYBER 2020 received a total of 143 submissions. Each submission was reviewed by at least three Program Committee members. The committee decided to accept 37 full papers. Papers were accepted on the basis of technical merit, presentation and relevance to the conference. SMARTCYBER 2020 was enriched by the lectures and insights given by the following seven distinguished invited speakers: Prof. Prasant Kumar Pattnaik, School of Computer Engineering, Kalinga Institute of Industrial Technology; Professor Ana Hol, Western Sydney University, Australia; Professor Aninda Bose, Senior Editor Springer India; Prof. Evizal Abdul Kadir, UIR, Indonesia; Dr. James Aich S, CEO Terenz Co. Ltd, South Korea; Prof. Mangal Sain, Dongseo University, South Korea; and Prof. Ahmed A. Al-Absi, Kyungdong University Global Campus, South Korea. We thank the invited speakers for sharing the enthusiasm for research and accepting our invitation to share their expertise as well as contributing papers for inclusion in the proceedings. v

vi

Preface

SMARTCYBER 2020 has been able to maintain standards in terms of the quality of papers due to the contribution made by many stakeholders. We are thankful to the General Chairs, Prasant Kumar Pattnaik, KIIT Deemed to be University, India; Ahmed A. Al-Absi, Kyungdong University, South Korea; Mangal Sain, Dongseo University. We futher thank the Program Chairs, Baseem Al-athwari, Kyungdong University Global Campus, South Korea; Pardeep Kumar, Swansea University, UK; Deepanjali Mishra, KIIT Deemed to be University, India, for their guidance and valuable inputs. We are grateful to Prof. John Lee, President of Kyungdong University (KDU) Global Campus, South Korea, and Honorary General Chair, SMARTCYBER 2020, for his constant support and for providing the infrastructure and resources to organize the conference. We are thankful to Prof. Sasmita Rani Samanta, Pro-Vice-Chancellor, KIIT Deemed to be University, India, Honorary General Chair, SMARTCYBER 2020, for providing all the support for the conference. Thanks are due to the Program and Technical committee members for their guidance related to the conference. We would also like to thank the Session Management Chairs, Publications Chairs, Publicity Chairs, Organizing Chairs, Finance Chairs and Web Management Chair who have made an invaluable contribution to the conference. We acknowledge the contribution of EasyChair in enabling an efficient and effective way in the management of paper submissions, reviews and preparation of proceedings. Finally, we thank all the authors and participants for their enthusiastic support. We are very much thankful to entire team of Springer Nature for timely support and help. We sincerely hope that you find the book to be of value in the pursuit of academic and professional excellence. Bhubaneswar, India Gangwondo, Korea (Republic of) Busan, Korea (Republic of) Swansea, UK

Prasant Kumar Pattnaik Ahmed A. Al-Absi Mangal Sain Pardeep Kumar

Contents

Proposal of Pseudo-Random Number Generators Using PingPong256 and Chaos Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ki-Hwan Kim and Hoon Jae Lee

1

Early Detection of Alzheimer’s Disease from 1.5 T MRI Scans Using 3D Convolutional Neural Network . . . . . . . . . . . . . . . . . . . . . . . . Sabyasachi Chakraborty, Mangal Sain, Jinse Park, and Satyabrata Aich

15

Graph Theory-Based Numerical Algorithm to Secure WSAN Network with Low Delay and Energy Consumption . . . . . . . . . . . . . . . . . . . . . . . Ju Jinquan, Mohammed Abdulhakim Al-Absi, Ahmed Abdulhakim Al-Absi, Mangal Sain, and Hoon Jae Lee

29

Decentralized Privacy Protection Approach for Video Surveillance Service . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jeongseok Kim and Jaeho Lee

45

Exploring Generative Adversarial Networks for Entity Search and Retrieval . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Wafa Arsalane

55

Secure Marine Communication Under Distributed Slotted MAC . . . . . . Mohammed Abdulhakim Al-Absi, Ahmadhon Kamolov, Ki-Hwan Kim, Ahmed Abdulhakim Al-Absi, and Hoon Jae Lee IoT Technology with Marine Environment Protection and Monitoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mohammed Abdulhakim Al-Absi, Ahmadhon Kamolov, Ahmed Abdulhakim Al-Absi, Mangal Sain, and Hoon Jae Lee Automatic Detection of Security Misconfigurations in Web Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sandra Kumi, ChaeHo Lim, Sang-Gon Lee, Yustus Oko Oktian, and Elizabeth Nathania Witanto

69

81

91

vii

viii

Contents

Real-Time Access Control System Method Using Face Recognition . . . . 101 Mohammed Abdulhakim Al-Absi, Gabit Tolendiyev, Hoon Jae Lee, and Ahmed Abdulhakim Al-Absi Towards a Sentiment Analyser for Low-resource Languages . . . . . . . . . 109 Dian Indriani, Arbi Haza Nasution, Winda Monika, and Salhazan Nasution DGA Method Based on Fuzzy for Determination of Transformer Oil Quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 Obhi Thiessaputra, Muhamad Haddin, and Sri Arttini Dwi Prasetyowati Deep Learning-Based Apple Defect Detection with Residual SqueezeNet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 M. D. Nur Alam, Ihsan Ullah, and Ahmed Abdulhakim Al-Absi Smart Parking Management System in Shopping Malls . . . . . . . . . . . . . 135 S. Aravinthkumar, Shreya Makkar, and Ahmed Abdulhakim Al-Absi Blockchain-Based Solution for Effective Employee Management . . . . . . 147 Yuli Nurhasanah, Dita Prameswari, and Olivia Fachrunnisa Implementation of Motorcycle Monitoring Using Bluetooth with an Android-Based Microcontroller Using Arduino . . . . . . . . . . . . . 155 Yudhi Arta, Evizal Abdul Kadir, Ari Hanggara, Des Suryani, and Nesi Syafitri A Comparative Analysis of Data Mining Analysis Tools . . . . . . . . . . . . 165 Eugene Istratova, Dina Sin, and Konstantin Strokin Apple Defects Detection Based on Average Principal Component Using Hyperspectral Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173 MD. Nur Alam, Rakesh Thapamagar, Tilak Rasaili, Otabek Olimjonov, and Ahmed Abdulhakim Al-Absi Development of an Information System for the Collection and Processing of Big Data in Construction . . . . . . . . . . . . . . . . . . . . . . 189 Eugene Istratova, Dina Sin, and Konstantin Strokin Genetic Algorithm for Decrypting User’s Personal Information . . . . . . . 197 Fu Rui, Mohammed Abdulhakim Al-Absi, Ki-Hwan Kim, Ahmed Abdulhakim Al-Absi, and Hoon Jae Lee Text File Protection Using Least Significant Bit (LSB) Steganography and Rijndael Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 Apri Siswanto, Yudhi Arta, Evizal Abdul Kadir, and Bimantara Apple Defect Detection Based on Deep Convolutional Neural Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215 MD. Nur Alam, Shahi Saugat, Dahit Santosh, Mohammad Ibrahim Sarkar, and Ahmed Abdulhakim Al-Absi

Contents

ix

Satellite Image Segmentation and Classification Using Fuzzy C-Means Clustering and Support Vector Machine Classifier . . . . . . . . . . . . . . . . 225 P. Manjula, Ojasvita Muyal, and Ahmed A. Al-Absi The Determinants of Internet Financial Reporting for Investor Decision Making: Evidence from Indonesia Companies . . . . . . . . . . . . . 239 Kurnia Rina Ariani and Gustita Arnawati Putri Resource Allocation in the Integration of IoT, Fog, and Cloud Computing: State-of-the-Art and Open Challenges . . . . . . . . . . . . . . . . 247 Baseem Al-athwari and Hossain Md Azam The Application of Technology Acceptance Model to Assess the Role of Complexity Toward Customer Acceptance on Mobile Banking . . . . . 259 Gustita Arnawati Putri, Ariyani Wahyu Wijayanti, and Kurnia Rina Ariani Exploring the Volatility of Large-Scale Shared Distributed Computing Resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267 Md Azam Hossain, Baseem Al-athwari, Jik-soo Kim, and Soonwook Hwang Business Transformations Within Intelligent Eco-Systems . . . . . . . . . . . 275 Ana Hol Detection of Network Intrusion and Classification of Cyberattack Using Machine Learning Algorithms: A Multistage Classifier Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285 Jay Sarraf, Vaibhaw, Sabyasachi Chakraborty, and Prasant Kumar Pattnaik Robotic Process Automation Implementation Challenges . . . . . . . . . . . . 297 Daehyoun Choi, Hind R’bigui, and Chiwoon Cho Blockchain Technology to Support Employee Recruitment and Selection in Industrial Revolution 4.0 . . . . . . . . . . . . . . . . . . . . . . . 305 Happy Rhemananda, Dima Roulina Simbolon, and Olivia Fachrunnisa Android-Based Online Attendance Application . . . . . . . . . . . . . . . . . . . 313 Panji Rachmat Setiawan, Abdul Syukur, Novendra Kurniadi, and Amrizal Amrizal Customer Sentiment Analysis Using Cloud App and Machine Learning Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325 P. Manjula, Neeraj Kumar, and Ahmed A. Al-Absi Mood Enhancer Based on Facial Expression Using Machine Learning and Virtual Assistant Technology—An Android App . . . . . . . . . . . . . . 337 P. Manjula, Akshay Nagpal, and Ahmed A. Al-Absi

x

Contents

Integrating Complete Locomotive Assistance and IoT-Based Health Care for the Disabled . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353 S. Aravinthkumar, Ajayveer Singh Chandel, and Ahmed Abdulhakim Al-Absi Classification of Multiple Steganographic Algorithms Using Hierarchical CNNs and ResNets . . . . . . . . . . . . . . . . . . . . . . . . . . 365 Sanghoon Kang, Hanhoon Park, and Jong-Il Park Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375

Editors and Contributors

About the Editors Prasant Kumar Pattnaik Ph.D. (Computer Science), Fellow IETE, Senior Member IEEE, is a Professor at the School of Computer Engineering, KIIT Deemed University, Bhubaneswar. He has more than a decade of teaching and research experience and awarded half dozen of Ph.D. Dr. Pattnaik has published numbers of research papers in peer-reviewed international journals and conferences and filed many patents. He also edited book volumes in Springer and IGI Global Publication. His areas of interest include mobile computing, cloud computing, cyber security, intelligent systems, and brain–computer interface. He is one of the Associate Editors of Journal of Intelligent & Fuzzy Systems, IOS Press, and Intelligent Systems Book Series Editor of CRC Press, Taylor Francis Group. Mangal Sain received the Master of Application degree from India in 2003 and the Ph.D. degree in Computer Science from Dongseo University, Busan, South Korea, in 2011. Since 2011, he has been an Assistant Professor with the Department of Information and Communication Engineering, Dongseo University, Busan, South Korea. He has published over 40 international publications. His current research interests include wireless sensor network, middleware, cloud computing, embedded system, and the Internet of Things. He is a member of TIIS and has participated as a TPC member in several international conferences. Ahmed A. Al-Absi Ph.D (Computer Science), is an Associate Professor at the Smart Computing Department, Kyungdong University Global Campus, South Korea. He is currently Dean of International Faculty and Director of Global Academic Collaboration Centers at Kyungdong University Global. He has more than ten years of experience in teaching and university lecturing in the areas of database design and computer algorithms. Dr. Al-Absi has published numbers of research papers in peer-reviewed international journals and conferences. His research areas are Big Data, Large Scale Data Process Systems, Cloud Computing,

xi

xii

Editors and Contributors

IoT, VANET, Deep Learning, Parallel Computing, Security, and Bioinformatics. His professional experience includes being a speaker at a number of renowned research conferences and technical meetings such as IEEE, Korea ICT leaders forum, and reviewer for refereed journals and conferences on data-intensive computing as well as an examiner for postgraduate scholars in his research areas. Pardeep Kumar received the B.E. degree in Computer Science from Maharishi Dayanand University, Haryana (India), in 2002, the M.Tech. degree in Computer Science from Chaudhary Devi Lal University, Haryana (India), in 2006, and the Ph.D. degree in Ubiquitous Computing from Dongseo University, Busan (South Korea) in 2012. He is currently a Lecturer/Assistant Professor with the Department of Computer Science, Swansea University, Swansea, UK. From 2012 to 2018, he had held postdoc positions at the Department of Computer Science, Oxford University, Oxford UK (08/2016–09/2018), at the Department of Computer Science, The Arctic University of Norway, Tromso, Norway (08/2015–08/2016), and at Centre for Wireless Communications and the Department of Communications Engineering, University of Oulu, Finland (04/2012 to 08/2015).

Contributors Satyabrata Aich Terenz Co., Ltd., Busan, Republic of Korea Ahmed A. Al-Absi International Faculty (Academic), Global Academic Collaboration Centers, Kyungdong University-Global Campus, Wonju-si, South Korea Ahmed Abdulhakim Al-Absi Dean of International Faculty (Academic), Director of Global Academic Collaboration Centers, Kyungdong University, Yangju, South Korea; Department of Smart Computing, Kyungdong University, Bongpo, Gosung, Gangwondo, Republic of Korea Mohammed Abdulhakim Al-Absi Department of Computer Engineering, Dongseo University, Sasang-gu, Busan, Republic of Korea Baseem Al-athwari Smart Computing Department, Kyungdong University, Goseong-gun, Gangwon-do, South Korea; Department of Computer Engineering, Kyungdong University, Global Campus (Goseong), Goseong, Gangwon-do, Republic of Korea Amrizal Amrizal Department of Informatics, Universitas Islam Riau, Pekanbaru, Indonesia S. Aravinthkumar SRM University, Sonepat, India Kurnia Rina Ariani Muhammadiyah Surakarta University, Surakarta, Indonesia

Editors and Contributors

xiii

Wafa Arsalane Jeonbuk National University, Jeonju, South Korea Yudhi Arta Department of Informatics Engineering, Faculty of Engineering, Universitas Islam Riau, Pekanbaru, Indonesia Hossain Md Azam Smart Computing Department, Kyungdong University, Goseong-gun, Gangwon-do, South Korea Bimantara Department of Informatics Engineering, Faculty of Engineering, Universitas Islam Riau, Pekanbaru, Indonesia Sabyasachi Chakraborty Department of Computer Engineering, Inje University, Gimhae, South Korea; Terenz Co., Ltd., Busan, Republic of Korea Ajayveer Singh Chandel VIT University, Vellore, India Chiwoon Cho School of Industrial Engineering, University of Ulsan, Ulsan, Republic of Korea Daehyoun Choi School of Industrial Engineering, University of Ulsan, Ulsan, Republic of Korea Olivia Fachrunnisa Department of Management, Faculty of Economics, Universitas Islam Sultan Agung, Semarang, Indonesia Muhamad Haddin Electrical Engineering, Universitas Islam Sultan Agung, Semarang, Indonesia Ari Hanggara Department of Informatics, Universitas Islam Riau, Pekanbaru, Indonesia Ana Hol School of Computer, Data and Mathematical Sciences, Western Sydney University, Rydalmere, NSW, Australia Md Azam Hossain Department of Computer Engineering, Kyungdong University, Global Campus (Goseong), Goseong, Gangwon-do, Republic of Korea Soonwook Hwang Korea Institute of Science and Technology Information (KISTI), Daejeon, Republic of Korea Dian Indriani Informatics Engineering, Universitas Islam Riau, Riau, Indonesia Eugene Istratova Department of Automated Control Systems, Novosibirsk State Technical University, Novosibirsk, Russia Ju Jinquan Department of Computer Engineering, Dongseo University, Sasang-gu, Busan, Republic of Korea Evizal Abdul Kadir Department of Informatics Engineering, Faculty of Engineering, Universitas Islam Riau, Pekanbaru, Indonesia Ahmadhon Kamolov Department of Computer Engineering, Dongseo University, Busan, Republic of Korea

xiv

Editors and Contributors

Sanghoon Kang Department of Electronic Engineering, Pukyong National University, Busan, South Korea Jik-soo Kim Department of Computer Engineering, Myongji University, Yongin, Republic of Korea Ki-Hwan Kim Department of Computer Engineering, Dongseo University, Busan, Republic of Korea; Department of Ubiquitous IT, Dongseo University, Busan, Republic of Korea Jeongseok Kim Department of Electrical and Computer Engineering, University of Seoul, Seoul, South Korea; Security Labs, AIX Center, SK Telecom, Seoul, South Korea Neeraj Kumar VIT University, Vellore, India Sandra Kumi Department of Computer Engineering, Dongseo University, Busan, South Korea Novendra Kurniadi Department of Informatics, Universitas Islam Riau, Pekanbaru, Indonesia Hoon Jae Lee Division of Computer Engineering, Dongseo University, Busan, Republic of Korea; Division of Information and Communication Engineering, Dongseo University, Sasang-gu, Busan, Republic of Korea Jaeho Lee Department of Electrical and Computer Engineering, University of Seoul, Seoul, South Korea Sang-Gon Lee Department of Computer Engineering, Dongseo University, Busan, South Korea ChaeHo Lim BITSCAN Co., Ltd., Seoul, South Korea Shreya Makkar VIT University, Vellore, India P. Manjula SRM University, Sonepat, India Winda Monika Library Science, Universitas Lancang Kuning, Riau, Indonesia Ojasvita Muyal VIT University, Vellore, India Akshay Nagpal VIT University, Vellore, India Arbi Haza Nasution Informatics Engineering, Universitas Islam Riau, Riau, Indonesia Salhazan Nasution Informatics Engineering, Universitas Riau, Riau, Indonesia MD. Nur Alam Department of Smart Computing, Kyungdong University, Gosung, Gangwondo, South Korea

Editors and Contributors

xv

M. D. Nur Alam Department of Smart Computing, Kyungdong University, Bongpo, Gosung, Gangwondo, South Korea Yuli Nurhasanah Department of Management, Faculty of Economics, Universitas Islam Sultan Agung, Semarang, Indonesia Yustus Oko Oktian Department of Computer Engineering, Dongseo University, Busan, South Korea Otabek Olimjonov Department of Smart Computing, Kyungdong University, Gosung, Gangwondo, South Korea Hanhoon Park Department of Electronic Engineering, Pukyong National University, Busan, South Korea Jinse Park Department of Neurology, Haeundae Paik Hospital, Inje University, Busan, Republic of Korea Jong-Il Park Department of Computer Science, Hanyang University, Seoul, South Korea Prasant Kumar Pattnaik School of Computer Engineering, KIIT University, Bhubaneswar, India Dita Prameswari Department of Management, Faculty of Economics, Universitas Islam Sultan Agung, Semarang, Indonesia Sri Arttini Dwi Prasetyowati Electrical Engineering, Universitas Islam Sultan Agung, Semarang, Indonesia Gustita Arnawati Putri Veteran Bangun Nusantara Sukoharjo University, Kabupaten Sukoharjo, Indonesia Tilak Rasaili Department of Smart Computing, Kyungdong University, Gosung, Gangwondo, South Korea Happy Rhemananda Department of Management, Faculty of Economics, Universitas Islam Sultan Agung, Semarang, Indonesia Fu Rui Department of Computer Engineering, Dongseo University, Busan, Republic of Korea Hind R’bigui School of Industrial Engineering, University of Ulsan, Ulsan, Republic of Korea Mangal Sain Division of Computer Engineering, Dongseo University, Busan, Republic of Korea; Division of Information and Communication Engineering, Dongseo University, Sasang-gu, Busan, Republic of Korea Dahit Santosh Department of Smart Computing, Kyungdong University, Gosung, Gangwondo, Republic of Korea

xvi

Editors and Contributors

Mohammad Ibrahim Sarkar Department of Electronic Engineering, Chonbuk National University, Deokjin-gu, Jeonju-si, Jeollabuk-do, Republic of Korea Jay Sarraf School of Computer Engineering, KIIT University, Bhubaneswar, India Shahi Saugat Department of Smart Computing, Kyungdong University, Gosung, Gangwondo, Republic of Korea Panji Rachmat Setiawan Department of Informatics, Universitas Islam Riau, Pekanbaru, Indonesia Dima Roulina Simbolon Department of Management, Faculty of Economics, Universitas Islam Sultan Agung, Semarang, Indonesia Dina Sin Department of Construction, Technical Oil and Gas Institute, Sakhalin State University, Yuzhno-Sakhalinsk, Russia Apri Siswanto Department of Informatics Engineering, Faculty of Engineering, Universitas Islam Riau, Pekanbaru, Indonesia Konstantin Strokin Department of Construction, Technical Oil and Gas Institute, Sakhalin State University, Yuzhno-Sakhalinsk, Russia Des Suryani Department of Informatics, Universitas Islam Riau, Pekanbaru, Indonesia Nesi Syafitri Department of Informatics, Universitas Islam Riau, Pekanbaru, Indonesia Abdul Syukur Department of Informatics, Universitas Islam Riau, Pekanbaru, Indonesia Rakesh Thapamagar Department of Smart Computing, Kyungdong University, Gosung, Gangwondo, South Korea Obhi Thiessaputra Electrical Engineering, Universitas Islam Sultan Agung, Semarang, Indonesia Gabit Tolendiyev Department of Computer Engineering, Dongseo University, Busan, Republic of Korea Ihsan Ullah Department of Robotics Engineering, Daegu Gyeonbuk Institute of Science and Technology, Daegu, South Korea Vaibhaw School of Computer Engineering, KIIT University, Bhubaneswar, India Ariyani Wahyu Wijayanti Veteran Bangun Nusantara Sukoharjo University, Kabupaten Sukoharjo, Indonesia Elizabeth Nathania Witanto Department of Computer Engineering, Dongseo University, Busan, South Korea

Proposal of Pseudo-Random Number Generators Using PingPong256 and Chaos Maps Ki-Hwan Kim and Hoon Jae Lee

Abstract Internet of Things (IoT) devices are easily exposed to physical attacks; therefore, their design must consider authentication and encryption. Many authentication and encryption methods use algorithms such as advanced encryption standard (AES) and secure hash algorithm (SHA). A pseudo-random number generator (PRNG) can also be used for authentication and encryption, and linear feedback shift register (LFSR) provides an easy way to generate PRNGs. LFSR allows the mathematical generation of unique values proportional to a given length. However, as LFSR is mathematically predictable, it is not used alone for this purpose. PingPong256 uses a variable clock for LFSR that can generate very long periods. However, LFSRs are still potentially at risk of being attacked by correlation analysis attacks. There are several methods to account these security issues, including chaos maps (such as logistic maps), SHA, and AES. This paper proposes a method of using logistic maps corresponding to PingPong256 and chaos maps. For this purpose, various PingPong256 configurations are proposed and compared to verify the effectiveness of the proposed method. The method was tested using NIST SP800-22. Keywords PRNG · PingPong256 · Logistic map · LFSR · IoT

1 Introduction The demand for wearable equipment is rapidly increasing worldwide every year [1– 4]. The cumulative sales volume of wearable equipment worn on the wrist was one of the world’s highest [5]. Wearable equipment has several advantages. First, wearable equipment has high portability because it is designed to be worn on various parts K.-H. Kim Department of Ubiquitous IT, Dongseo University, Busan 47011, Republic of Korea e-mail: ghksdl90@naver.com H. J. Lee (B) Division of Computer Engineering, Dongseo University, Busan 47011, Republic of Korea e-mail: hjlee@dongseo.ac.kr © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 P. K. Pattnaik et al. (eds.), Proceedings of International Conference on Smart Computing and Cyber Security, Lecture Notes in Networks and Systems 149, https://doi.org/10.1007/978-981-15-7990-5_1

1

2

K.-H. Kim and H. J. Lee

of the body and has low power requirements. Second, biosensors can continuously monitor health status such as heart rate, electrocardiography (ECG) [6], and blood pressure. Third, data transfer can be achieved conveniently using built-in wireless communication sensors (Wi-Fi, Bluetooth, etc.). Various types of data measured by wearable devices allow the observation of different biometric information such as stress and a user’s mental state [7–9]. These devices can also incorporate artificial intelligence (AI) and a dedicated analytics system can monitor and alert a person regarding health readings [10–12]. However, various cyber-attacks on these devices have been occurring, including the use of robots in distributed denial of service (DDoS) attacks, remote control of medical devices, and the capture of privileges and control of security cameras and automobiles [13]. IoT devices have extremely low computing power and storage space and are often placed in environments that are vulnerable to attack due to limited communicability and physical exposure of the equipment. Therefore, lightweight hardware devices require a proper balance between performance and security. The main purpose of the PingPong algorithm is to generate a pseudo-random noise sequence (PN code) with a long period. PN codes have basic requirements including randomness, unpredictability, and incapability of reproducing/repeating the same sequence (non-reproducibility). A linear feedback shift register (LFSR) is used in this method for generating the maximum length sequence (MLS). The MLS feature of LFSR describes the condition that when there is a memory of size m bits, the period is L = 2m −1 bits and all values generated during the cycle are unique. LFSR is mainly used for generating pseudo-random bit sequences (PRBS), signal signatures, and footings of signal sets. The main advantage of the system is that the operation can be interpreted precisely by the algebraic principle, and the digital conversion is very easy to perform. PingPong256 is based on PingPong128 and is a hybrid generator, combining the nonlinear Lee Moon (NLM) generator with a highly secure clock-controlled generator [14, 15]. It consists of two different LFSRs (LFSR255 and LFSR257 bits), LM generators, and two memory bit store allocations. Chaos theory is the discipline dealing with deterministic nonlinear dynamic systems [16, 17]. Fractals are differentiable and unlike regular Euclidean geometric bodies, they have irregularly separated structures [18]. Time-averaged fractals (escape-time fractals) are color images of the speed at which each point emanates, usually on a complex plane. Various encryption schemes using a chaotic system have been studied [19–21]. The logistic map is a discrete-time dynamic system given as a quadratic polynomial of nonlinear differential equations representing chaotic phenomena. According to logistic thought, the n + 1 generation constitutes the function of the population of the n generation, which is represented by multiplying the opposite value of the input by a specific coefficient value. The logistic map according to r is shown in Fig. 1. The organization of this paper is structured as follows. First, we present the method to use the PingPong algorithm and logistic map to construct a pseudo-random number generator (PRNG). In Sect. 2, we study the PingPong256 and logistic map. In Sect. 3, we propose two PRNGs. Section 4 is the conclusion.

Proposal of Pseudo-Random Number Generators …

3

Fig. 1 Periodic change of the logistic map according to r value

2 Background 2.1 PingPong256 The overall structure of the PingPong256 algorithm is shown in Fig. 2. It is simple and easy to implement in hardware and software to ensure a long-term periodic output. The values of clock control called f a and f b are calculated by referring to some state values of different LFSRs. This method will provide different results over time even when initialized with the same value in the two LFSRs [22]. The linear structure of the LFSR can be extended with unpredictable results through the clock controller. This process can be viewed in detail in Fig. 3. Each LFSR in Fig. 3 has approximately 30 taps, making it harder to predict the cycle. The structure of the process can be changed according to the application. Even though two LFSRs with long periods of

Fig. 2 PingPong256 generator structure

4

K.-H. Kim and H. J. Lee

Fig. 3 Polynomial of PingPong256 function

ai and bi are influenced by an irregular clock controller, they will present an irregular periodic output unless the initial value is known. The output bits from the LFSR are inputs for the combined function (zi), carry function (fc), and memory function (fd) to produce the following memory states and key sequence bit: ci is a carry sequence, di is a memory sequence, and the initial values of the two variables are 0. The LM generator is an enhanced sums generator and must compute ci-1 and di-1 to compute zi. Therefore, PingPong256 acts as a random number generator in the LFSR core structure. LFSR has a structure that guarantees the maximum period of a defined size, but it also has a linear structure and can be predicted easily. Therefore, the PingPong256 has the ability to make the linearity of LFSR behave as a nonlinear structure through variable clocks and functions. In this paper, we attempt to improve the existing linearity of this system by replacing the linear disadvantage of LFSR LFSRwith the logistic map. PingPong256 generates a period as defined in Eq. 1 that prevents the reuse of the same key sequence when encrypting long messages. PingPong256 also has linear complexity as shown in Eq. 2 which will withstand attacks using the Berlekamp– Massey algorithm. Finally, a statistical feature of the function is that the frequency of the sequence of keys “0” or “1” should have approximately the same ability to withstand an attack [9]. The calculation of the linear complexity and period of PingPong256 is shown below: LC ≥ 24.6 × 2 P ≥ 24.6 × 2

512−11 2

= 24.6 × 20.5 × 2250 ≈ 2256

(1)

512−11 2

= 24.6 × 20.5 × 2250 ≈ 2256

(2)

Proposal of Pseudo-Random Number Generators …

5

2.2 Logistic Map The experiment was conducted by setting the value of r to 3.999 in the logistic map function xn + 1 = r * xn(1−xn). The r values were varied for each experimental iteration to produce a variety of results. The logistic map shows that to guarantee a random number, it is preferred to use a constant that cannot be predicted and has a decimal value of 20 bits or fewer [23]. Based on these criteria, and as shown in Fig. 4, the data structure corresponding to the double type variable of the C programming language is divided into four regions: sign, integer, prime number 0, and prime number 1. In this experiment, only the 32-bit value corresponding to the prime number 1 is used. A random number generator test investigates the properties and efficacy of a proposed random number generator. To complete the test, many samples are collected, and statistical analysis is used to determine whether random numbers are generated. If the generator does not pass the statistical test, it must not be recognized as a random number generator and should not pass any other complex tests. Generally, if the random number generator passes all statistical tests, it generates a random number. However, occasionally a generator which has passed the tests may not be able to generate a random number. There are two types of errors in statistical hypothesis testing. Let H0 be the hypothesis that a given sample output sequence is generated by a random bit generator. If the significance level (a) for H0 is too large, a test result of “reject” occurs even though the given sequence is random. This error is called a Type I error. However, if the significance level is too small, the test will “accept” the sequence even if the given sequence is not random. This error is called a Type II error. Generally, the significance level is chosen to be 0.001 ≤ a ≤ 0.05. Essentially, when verifying the data set consisting of bits in a random number, the following method is used [24]. The logistic map is constructed as shown in Fig. 4, and the experimental setup is r = 3.999, x = 0.300501–0.300505. As shown in Table 1, most of the experimental results were satisfactory and passed the test. However, in three experiments, it was confirmed that values exceeding the limit range were found. These cases were primarily found in the poker test with m = 5. We have established a similar experimental environment and confirmed the need to extend the scope of these measurements based on this study [25]. Therefore, in this paper, the numerical values deviating from the allowable range are interpreted as inadequate for use as reference values due to the limitation of the measurement range. As a result, we confirmed that

Fig. 4 Experiment 1: random number generation using the logistic map

6

K.-H. Kim and H. J. Lee

Table 1 Experiment 1: randomness test result (r = 3.999, x = 0.300501–0.300505) Test

Case 1

Case 2

Case 3

Frequency

Criterion 3.841

0.124

2.633

0.085

Serial

5.991

0.129

2.709

1.962

T =3

9.488

2.172

6.023

2.070

T =4

15.507

6.192

16.063

5.520

T =5

26.296

11.360

21.979

12.104

Generalize serial test

Poker test M=3

14.067

2.007

6.558

6.842

M=4

24.996

12.063

22.175

9.476

M=5

44.654

46.279

52.804

46.298

0.010404

0.009974

0.008802

Auto-collimation

Max ≤ 0.05

using a decimal value corresponding to the lower 32 bits provides a result approaching a random number.

3 Proposal main model The LFSR outputs 1 bit per round of calculation, but the structure using the logistic map outputs 32 bits per round. Figure 5 shows the structure in which all LFSRs of the PingPong256 are changed to a logistic map to unify the components which have different outputs. Thus, it has a 32-bit register output per round and has a 32-bit output per round compared to the existing PingPong algorithm. The diagram detailing the PingPong256 algorithm using the logistic map is shown in Fig. 5.

Fig. 5 Two logistic map PingPong256

Proposal of Pseudo-Random Number Generators …

7

3.1 Two logistic Maps and XOR Operations. We used a random number generator in a logistic map experiment which used decimals to reduce the 32 bits of the result to 11 bit. This experiment induced chaos and created a random number generator structure that uses the control group to avoid guessing the initial value. The structure of Experiment 1 is shown in Fig. 6, where the original logistic map X and the comparative logistic map Y are generated [26]. The experimental values were r 1 = 3.999, r 2 = 3.999, and X 0 = 0.300501 and Y 0 = 0.300503. To confirm whether the structure is effective, Y was increased by 0.0002 to make eight control groups. Table 2 lists the results of setting the initial values as r 1 = 3.999, r 2 = 3.999, X 0 = 0.300501, and Y 0 = 0.300503 as shown in Fig. 4. However, the poker test did not satisfy the conditions for an allowable value, and it was confirmed that it

Fig. 6 Experiment 1: a selector structure based on the difference between different random number generation

Table 2 Experiment 1: randomness test result Test

Criterion

Experiment 2

Frequency test

3.841

0.588

Serial test

5.991

7166.367

Generalize serial test T =3

9.488

9440.613

T =4

15.507

13982.530

T =5

26.296

19077.599

M=3

14.067

5112.210

M=4

24.996

7617.505

M=5

44.654

8358.768

Auto-collimation test

≤0.05

0.027487

Poker test

8

K.-H. Kim and H. J. Lee

was difficult to verify the randomness. In a similar paper [26], the opposite result was found, showing that the control structure using logistic map produced positive results. References to the paper [26] show that the above structure varies with input values, and it is uncommon to pass all the test results. To account for this condition, we used an exclusive logical OR function as shown in Fig. 7. The setup for Experiment 2 was r 1 = 3.999, r 2 = 3.999, X 0 = 0.300501, and Y 0 = 0.300503. In this experiment, the r values of the different logistic maps were set with the same value, and the initial input values were varied. The results of the experiment are listed in Table 3. When m = 5 in the poker verification, which was the only portion in the previous experiment to fail, the result was acceptable. Thus, experimental results show that the logistic map with different initial values can be improved using the exclusive OR function. Experiments show that a random number generator using various chaotic functions can express desirable results. To overcome

Fig. 7 Experiment 2: a random number generator using two logistic maps and exclusive OR function

Table 3 Experiment 2: logistic map random number verification result Test

Criterion

Case 1

Case 2

Case 3

Frequency test

3.841

3.064

2.391

2.160

Serial test

5.991

4.937

2.417

2.222

T =3

9.488

5.932

2.687

8.144

T =4

15.507

10.264

8.281

9.872

T =5

26.296

13.946

20.171

22.652

M=3

14.067

11.207

6.184

15.351

M=4

24.996

17.78

16.492

20.957

M=5

44.654

34.481

40.706

40.573

Auto-collimation test

≤0.05

0.010404

0.009974

0.008802

Generalize serial test

Poker test

Proposal of Pseudo-Random Number Generators …

9

Fig. 8 400 outputs from each structure a Experiment 1, b Experiment 2

this problem, we added a PingPong256 structure corresponding to a stream cipher [27]. The results of Experiment 1 and Experiment 2 in graphic form are shown in Fig. 8. Experiment 1 produces results of 0 and 1, so the attacker has a 50% chance of choosing the correct answer each time. Experiment 2 produces real numbers from 0 to 1, so it is more sensitive to changes in value.

3.2 Output Change According to Clock Control Function Logistic maps that do not use clock control can be deduced by continuous observation. The generated signal for this model is created using the XOR model proposed in Sect. 3.1. The initial values of the two different logistic maps are 0.5, and the r coefficient values, r 1 and r 2 are set to 3.999 and 3.9999, respectively. The blue line in Fig. 9 is the logistic map result using the x value, and the red-dotted line is the result using the y value. The functions f a and f b are configured to operate using an arbitrary value corresponding to each logistic map. Figure 10 compares the results of the operation using the same initial value as Fig. 9 and referring f a and f b to different logistic maps. In this case, the blue line

Fig. 9 Logistic map output without clock control function

10

K.-H. Kim and H. J. Lee

Fig. 10 Difference between the logistic map without clock control function and logistic map with clock control function

represents the difference value of the logistic map referring to the x value, and the red-dotted line represents the difference value of the logistic map referring to the y value. Because the initial values of both sides are the same, the difference does not occur until approximately counting 13 on the x-axes. Subsequently, the difference between them can be observed, and the output is an actual value between 0 and 1, which makes it difficult to make accurate predictions.

3.3 R-Factor Relationship in Logistic Map Table 4 lists the results of performing the PingPong256 procedure using two logistic maps three times with different initial values. A total of 2,257,600 bits were used for random number verification. The tests were varied by changing the initial value of the logistic map by increasing x 2 of x 1 and using another initial value. The initial values Table 4 Two logistic map PingPong256 random verification results (r 1 = 3.999, r 2 = 3.999, total length = 2,457,600, x 1 = 0.7861, x 2 = 0.859167) Test

Criterion

Case 1

Case 2

Case 3

Frequency

3.841

0.375

0.008

2.305

Serial

5.991

0.626

0.234

2.52

T =3

9.488

1.345

2.793

3.03

T =4

15.507

14.599

6.079

5.391

T =5

26.296

21.697

13.407

6.413

M=3

14.067

3.631

3.996

5.972

M=4

24.996

17.173

8.536

30.682

M=5

44.654

34.86

27.885

42.615

Auto-collimation

≥0.05

0.61717

0.635077

0.639495

Generalize serial test

Poker test

Proposal of Pseudo-Random Number Generators …

11

Table 5 Two logistic map PingPong256 random number verification result (r 1 = 3.999, r 2 = 3.9999, total length = 2,457,600, x1=0.7861, x 2 = 0.859167) Test

Criterion

Case 1

Case 2

Case 3

Frequency

3.841

1.818

0.095

0.549

Serial

5.991

5.089

0.412

4.462

Generalize serial test T =3

9.488

10.665

0.75

6.425

T =4

15.507

13.09

2.676

12.147

T =5

26.296

24.726

9.242

14.489

M=3

14.067

14.836

5.638

13.071

M=4

24.996

13.638

7.025

14.36

Poker test

M =5

44.654

19.291

16.356

32.019

Auto-collimation

≥0.05

0.070461

0.573655

0.047886

used in the experiment passed all the tests listed in Table 4. In case 3, frequency verification showed that the number of 0s returned was 1,226,420 and the number of 1s returned was 1,231,180 which summed to a total of 2,457,600 bits. In the sequence verification process, the number of bits changed from 0-to-0 was 612,051, those changed from 0-to-1 was 614,369, those changed from 1-to-0 was 614,368, and those changed from 1-to-1 was 616,811. However, it was confirmed that the criterion was not passed in a similar manner for all cases and input values. The case results are negative in cases 1 and 2. Table 5 lists the nonlinear element by changing the value of r 2 to 3.9999. When this change is made, cases 1, 2, and 3 each passed all the tests. With more acceptable results achieved, we re-implemented the PingPong256 of the logistic map using the changed r value above and confirmed the results as listed in Table 6. The characteristics of the system confirmed through this experiment are as follows. When generating a random number using multiple logistic maps, the values of r must be different. A logistic map has a wide range of random number generation ability according to r value. Using multiple logistic maps and nonlinear functions can help to ensure adequate randomness. It can be seen from the results of Tables 5 and 6 that the system is sensitive to initial values, and they are recommended to be set as asymmetrically as possible.

4 Conclusions This paper proposes a new PRNG method using the PingPong256 algorithm and logistic maps corresponding to chaos maps. There are three primary benefits in using the proposed algorithm. First, the clock controller of PingPong256 expands

12

K.-H. Kim and H. J. Lee

Table 6 Two logistic map PingPong256 random number verification results (r 1 = 3.9999, r 2 = 3.9999, total length = 2,457,600, x 1 = 0.7861, x 2 = 0.859167) Test

Criterion

Case 1

Case 2

Case 3

Frequency

3.841

0.784

2.759

0.844

Serial

5.991

11.112

3.073

2.352

Generalize serial test T =3

9.488

4.651

6.822

8.499

T =4

15.507

11.112

7.454

12.462

T =5

26.296

14.65

17.907

17.056

M=3

14.067

6.368

5.638

10.088

M=4

24.996

14.161

7.025

12.945

Poker test

M=5

44.654

31.945

32.757

22.624

Auto-collimation

≥0.05

0.800961

0.576845

0.219434

the randomness by introducing misalignment over time even when the same initial value is used on both sides of the process. Second, the carry and memory functions make it easy to create a continuous signal by using the previous and current outputs simultaneously. Finally, the logistic map achieves superior randomness to the LFSR due to its irregular and continuous features. Therefore, the PingPong algorithm using logistic maps can generate random numbers using logistic maps corresponding to a chaotic function using the high sensitivity of the initial value. While some experiments have shown that one logistic map function works successfully, when we tested them we found problems in the poker test including that they were biased and returned specific output patterns. The proposed PRNG guarantees long periodicity of irregular output and can be created by an intuitive hardware and software structure. Acknowledgments This work was supported by Dongseo University, “Dongseo Cluster Project” Research Fund of 2020 (DSU-20200008).

References 1. “Wearable device unit sales worldwide by region from 2015 to 2022 (in millions),” statistic 2. J.E. Mück, B. Ünal, H. Butt, A.K. Yetisen, Market and patent analyses of wearables in medicine. Trends Biotechnol. (2019) 3. IDC, IDC Forecasts Sustained Double-Digit Growth for Wearable Devices Led by Steady Adoption of Smartwatches, 17, Dec, 2018. https://www.idc.com/getdoc.jsp?containerId=prU S44553518 4. Marketer, Wearables 2019, 3, Jan, 2019. https://www.emarketer.com/content/wearables-2019 5. Forecast unit shipments of wearable devices worldwide from 2017 to 2019 and in 2022 (in million units), statistic, [Internet]

Proposal of Pseudo-Random Number Generators …

13

6. T.B. Garcia, D.J. Garcia, Arrhythmia Recognition: The Art of Interpretations (Jones & Bartlett Publishers, 2019) 7. C. Setz, et al., Discriminating stress from cognitive load using a wearable EDA device. IEEE Trans. Inf. Technol. Biomed. 14(2), 410–417 (2009) 8. R. Jerauld, Wearable emotion detection and feedback system. U.S. Patent No. 9,019,174. 28 Apr. 2015 9. V.P. Rachim, W.-Y. Chung, Wearable noncontact armband for mobile ECG monitoring system. IEEE Trans. Biomed. Circ. Syst. 10(6), 1112–1118 (2016) 10. CBINSIGHTS, From Virtual Nurses To Drug Discovery: 106 Artificial Intelligence Startups In Healthcare, 3, Feb, 2017. https://www.cbinsights.com/research/artificial-intelligence-startupshealthcare/ 11. IBM, IBM Wation Health. https://www.ibm.com/watson-health/learn/artificial-intelligencemedicine 12. J. Bresnick, Top 5 Use Cases for Artificial Intelligence in Medical Imaging, Health it Analytics, October, 30, 2018. https://healthitanalytics.com/news/top-5-use-cases-for-artificialintelligence-in-medical-imaging 13. Guest Writer, The 5 Worst Examples of IoT Hacking and Vulnerabilities in Recorded History, Iotforall, May, 10, 2017. 14. L.H. Jae, S.M. Sung, H.R. Kim, NLM-128, An Improved LM-type Summation Generator with 2-bit memories, in 2009 Fourth International Conference on Computer Sciences and Convergence Information Technology (IEEE, 2009) 15. H.J. Lee, S.J. Moon, On an improved summation generator with 2-bit memory. Sig. Proc. 80(1), 211–217 (2000) 16. B. Davies, Exploring Chaos: Theory and Experiment (CRC Press, 2018) 17. R.L. Devaney, A First Course in Chaotic Dynamical Systems: Theory and Experiment (CRC Press, 2018) 18. L. Méhauté, M.G. Alain, C. Tricot, Fractal Geometry (Carbon Black. Routledge, 2018), pp. 245–270 19. R.A. Elmanfaloty, E. Abou-Bakr. Random property enhancement of a 1D chaotic PRNG with finite precision implementation. Chaos, Solitons & Fractals 118, 134–144 (2019) 20. Z. Lin et al., Security performance analysis of a chaotic stream cipher. Nonlinear Dyn. 94(2), 1003–1017 (2018) 21. D. Eroglu, J.S.W. Lamb, T. Pereira, Synchronisation of chaos and its applications. Contemp. Phys. 58(3), 207–243 (2017) 22. H.J. Lee, Chen, K. PingPong-128, a new stream cipher for ubiquitous application, in 2007 International Conference on Convergence Information Technology (ICCIT 2007) (IEEE, 2007) 23. M. François, D. Defour, C. Negre, A fast chaos-based pseudo-random bit generator using binary64 floating-point arithmetic. Informatica 38(3) (2014) 24. A. Rukhin, J. Soto, etc., A Statistical Test Suite for Random and Pseudorandom Number Generators for Cryptographic Applications. National Institute of Standards and Technology. 25. H. Choi, D. Won, On algorithm for finding primitive polynomials over GF(q). Korea Inst. Inf. Secur. Cryptology 11(1), 35–42 (2001) 26. V. Patidar, K.K. Sud, N.K. Pareek, A pseudo random bit generator based on chaotic logistic map and its statistical testing. Informatica 33(4) (2009) 27. K.H. Kim, T.Y. Kim, S.G. Lee, W.T. Jang, H.J. Lee, Proposal of parallelization structure for PingPong 256. J. Eng. Appl. Sci. 13, 1124–1129 (2018)

Early Detection of Alzheimer’s Disease from 1.5 T MRI Scans Using 3D Convolutional Neural Network Sabyasachi Chakraborty, Mangal Sain, Jinse Park, and Satyabrata Aich

Abstract Alzheimer’s disease is a neurodegenerative disease that affects the old age population and is affected by the neurofibrillary tangles and neurotic plagues as they impair the neuron’s microtubule transport system. The onset of this disease leads to a decline in the normal cognitive functioning of a person. The commonly observed symptom of AD is the difficulty in remembering the latest events. Moreover, as the progression occurs in a person, it can include symptoms like issues with the language, mood swings, and behavioral issues. As for the particular disease, no cure has been found out yet, to completely eradicate the disease from the body, therefore detection in advance of the disease has proven to be effective in improving a person’s life. In the study, 1.5 T T1 weighted MRI scans were acquired from the Alzheimer’s disease neuroimaging initiative (ADNI) database of 910 patients, where 336 were healthy control, 307 were mild cognitive impairment(MCI), and 267 for Alzheimer’s disease. The study leverages a 3D convolutional neural network (3D-CNN) for learning the intricate patterns in the magnetic resonance imaging (MRI) scans for the detection of Alzheimer’s disease. The 3D-CNN model performed superiorly by plotting an accuracy of 95.88%, precision of 0.951, recall of 0.9601, and f1-score of 0.9538. Keywords Deep learning · MRI · Alzheimer’s · Disease · Neural network

S. Chakraborty · S. Aich (B) Terenz Co., Ltd., Busan, Republic of Korea e-mail: james@sikonic.io M. Sain Division of Computer Engineering, Dongseo University, Busan, Republic of Korea J. Park Department of Neurology, Haeundae Paik Hospital, Inje University, Busan, Republic of Korea © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 P. K. Pattnaik et al. (eds.), Proceedings of International Conference on Smart Computing and Cyber Security, Lecture Notes in Networks and Systems 149, https://doi.org/10.1007/978-981-15-7990-5_2

15

16

S. Chakraborty et al.

1 Introduction The most common neurological disorder that pertains to the old age population in the world is considered to be Alzheimer’s disease (AD). Alzheimer’s disease is primarily caused due to the neurofibrillary tangles and neurotic plagues as they impair the neuron’s microtubule transport system. Therefore, the cellular collapse happened in the brain cells which leads to shrinkage of the hippocampus and parts of the cerebral cortex and hence causes Alzheimer’s disease. Moreover, Alzheimer’s disease is progressive in nature, and therefore, early detection and monitoring of the disease lead to the improvement in the life of patients. Also, as the old age population is increasing rapidly, a requirement for the development of suitable methods for the detection of Alzheimer’s disease at a very early stage is indeed very important. For the early detection of Alzheimer’s disease, the most widely used diagnostic paradigm is the analysis of magnetic resonance imaging (MRI) scans of the brain. The MRI scans provide anatomical details about the subcortical structures of the brain that are further analyzed to check for any aneurysms, which further deems helpful for the early diagnosis of a particular type of disease. However, as the MRI is a 3D structure, it becomes really difficult for the human eye to analyze the intrinsic details and heterogeneous properties of subcortical structures. Therefore, with the advancement of intelligent technologies, computer-aided detection systems have been proven to be very effective concerning the analysis and diagnosis of diseases by leveraging multidimensional health care data. In the past, many studies have been performed and found out that the textural, morphological analysis of the tissue and cell imaging scans have provided some very astonishing results. The application of textural and morphological analysis was considered to be huge as it was able to perform the quantification of gray level patterns and derive the inter-pixel relationship within the regions of interest. Moreover, it was also observed that different areas in a scan or an image had different textural and morphological patterns which were difficult for human beings to calculate. Therefore, textural and morphological analysis of the imaging scans proved to be very much reliable for neurological studies and applications in the detection and diagnosis of progressive diseases. But with the advances in the field of computer applications and intelligent systems, the research community is now focusing more on data-driven feature representation rather than handcrafted feature engineering which requires domain-specific knowledge. Therefore, with the rapid development of deep learning architectures and technologies, it is proving to lay down some state-of-the-art methodologies for medical image applications.

Early Detection of Alzheimer’s Disease from 1.5 T MRI …

17

2 Data Collection and Preprocessing 2.1 Data Collection The data for the study was collected from the ADNI database (https://adni.loni.usc. edu/). ADNI database for neuroimages is considered to be a landmark, international, and multicenter study to research the biomarkers that are responsible for Alzheimer’s disease progression. The MRI scans selected for the study were based on particular imaging protocols described in Table 1 and also correspond to the baseline visit. Further, all the scans that were considered in the study were obtained from a single type of scanner, i.e., GE medical systems. Moreover, all the acquired scans were based on magnetization prepared-rapid gradient echo (MP-RAGE) sequence. All the scans used in the study were acquired in a time range of 20–30 min field of view (FoV) of all the scans including vertex, cerebellum, and pons. Post applying the filter based on the imaging protocol mentioned in Table 1, a total of 910 MRI scans were selected from the baseline visit of the patients. Out of 910 patients, 423 were female, and 487 were male. The scans that were considered for the study belonged to the subjects aged 68.26 ± 7.2. The scans primarily belonged to two research groups that are healthy control (HC), mild cognitive impairment (MCI), and Alzheimer’s disease (AD). The scans were distributed into the respective research groups as 336 for healthy control, 307 for MCI, and 267 for Alzheimer’s disease. The subjects who were considered for obtaining the scan were selected on certain criteria described in Table 2. Table 3 plotted above shows the specifications of the scans that were obtained from the ADNI database. Table 1 Parameters for choosing MRI scans from the ADNI study Imaging protocol

Values

Modality

MRI

Research group

HC, MCI, and AD

Visit

Baseline

Acquisition plane

Sagittal

Acquisition type

3D

Field strength

1.5 T

Flip angle

8.0°

Scanner manufacturing

GE medical systems

Pixel spacing

0.9–1.5 mm (X&Y)

Slice thickness

1.0 mm

Weighting

T1

18

S. Chakraborty et al.

Table 2 Eligibility criteria for the subject to be included in a group Research group

Criteria

Alzheimer’s disease

1. Mini-mental state exam score between 20 and 26 2. Clinical dementia rating of 0.5 or 1.0 3. NINCDS/ADRDA criteria for probable AD

Control

1. Subjects must be free of memory complaints 2. Mini-mental state exam score between 24 and 30

Mild cognitive impairment 1. Clinical dementia rating of 0.5 2. The subject must have a subjective memory concern as reported by the subject, study partner, or clinician 3. Cognitively normal, based on an absence of significant impairment in cognitive functions or activities of daily living

Table 3 Specification of acquired scans from ADNI Imaging parameters

Values

Dimensions

256 × 256 × 180 pixels

Interslice gap

0.0 mm

Slice thickness

1.0 mm

Spacing

0.9375 × 0.9375 × 0.9375 mm

Plane

Sagittal

2.2 Data Preprocessing The dataset that was used in the study was obtained from the ADNI database, and as mentioned above that ADNI is a multicenter study, therefore, the imaging scans acquired in the study contained temporal and spatial differences. To solve this particular problem and to maintain a constant tendency between all the scans, it is required that all the scans need to be in the same space such as Montreal Neurological Institute (MNI) [1, 2] or Individual Brain Atlases using Statistical Parametric Mapping (IBASPM) [3]. Therefore, to bring all the scans to the same space, an image registration procedure is performed. Image registration is a process that mutates upon a fixed image to find the correct alignment parameters so that an unknown or unseen image can be aligned similarly to the fixed image. Trivially, image registration could be understood as the process of aligning two images to a particular space where one acts as the source image and the other as target image, and the source image is transformed in a method to align with the target image. In the specific study, the MRI scans obtained from the PPMI database are considered as the source image, and the atlas such as MNI or IBASPM are considered as the target image. The registration of the MRI scans obtained from the ADNI database was performed using ICBM-152-T1w-Nonlinear-Symmetric atlas created by Fonov et al. [1, 2]. The specifications of the ICBM-152-T1w-Nonlinear-Symmetric atlas are described in Table 4. The registration of the MRI scans was performed using one

Early Detection of Alzheimer’s Disease from 1.5 T MRI …

19

Table 4 Specification of the acquired scans from ICBM–152- T1w-Nonlinear-Symmetric atlas Image parameters

Values

Dimensions

193 × 229 × 193 pixels

Interslice gap

0.0 mm

Slice thickness

1.0 mm

Spacing

1.0 × 1.0 × 1.0 mm

Plane

Sagittal

of the most effective normalization tools known as Advanced Normalization Tools Python (ANTsPy) [4]. ANTsPy is particularly used in the field of imaging research for extracting important information from complex imaging datasets to perform preprocessing on MRI, fMRI, and SPECT data. The registration of the acquired MRI scans with the ICBM-152-T1w-Nonlinear-Symmetric atlas was performed using symmetric normalization. Figure 1 shown below depicts a particular MRI scan before and after the registration process.

Fig. 1 Sample scan pre and post-registration. a MRI scan before registration b MRI scan after registration

20

S. Chakraborty et al.

3 Materials and Methods The main premise of the study focusses on the detection of Alzheimer’s disease and the classification of MRI scans as healthy control (HC), mild cognitive impairment (MCI), and Alzheimer’s disease (AD) using 3D convolutional neural networks. The complete flow of process and methodology for the detection of Alzheimer’s disease is been described in Fig. 2. The methodology is primarily divided into four (4) stages namely MRI scan acquisition from the ADNI database, data preprocessing, registration and transformation, 3D convolutional neural network architecture, and finally the results and performance evaluation of the CNN architecture based on particular performance metrics. The first two stages of the methodology have been thoroughly discussed in Sect. 2, and further, the third and the fourth stage will be discussed in the following sections.

3.1 3D Convolutional Neural Networks In recent times, supervised learning techniques for solving problems have evolved massively. Moreover, the popularity and effectiveness of deep learning algorithms have also undergone a major paradigm shift in terms of architectural designs and optimizer functions [5]. Particularly, in the field of health care, the deep learning

Fig. 2 Complete process flow

Early Detection of Alzheimer’s Disease from 1.5 T MRI …

21

algorithms have shown much predominance over the previous techniques that were used for imaging analysis, aneurysms detection in images, biosignal analysis, etc. In this study, a 3D convolutional neural network model has been developed for the detection of Alzheimer’s disease from T1 weighted MRI scans. The primary proposition of the work presents a system that can be used to identify Alzheimer’s disease from MRI or rather brain images. Additionally, the second proposition of the study is to determine the plausible regions of interest (ROIs) in the brain MRI images that are responsible for Alzheimer’s disease. Therefore, to solve the primary proposition of the study, a 3D convolutional neural network has been developed as shown in Fig. 3. The CNN network developed in the work consists of 18 layers including the input and the output layer. Further, the network architecture consists of ten 3D convolution layers which allows the model to create the feature representations of the input brain MRI scans. Moreover, all the convolution layers are supported by activation functions. Further, all the feature representations are subjected to maxpooling layers which are responsible for downsampling the input feature matrix and provides an abstract form of the feature representation to avoid overfitting. After the complete process of feature learning, all the feature matrices are flattened so that it can be accepted by the dense layer or the fully connected layer. The representations from the dense layer are further subjected to the output dense layer with three neurons and SoftMax activation which corresponds to the three states that are healthy control, mild cognitive impairment (MCI), and Alzheimer’s disease.

3.2 Hypothesis and Training Procedure For developing statistical, machine learning, and deep learning model, the first step is considered to be the development of the hypothesis of the problem that needs to be solved. Therefore, the primary hypothesis that was devised for solving a particular problem is as follows: 1. The recall of Alzheimer’s disease class must be 100%, and there should not be any mispredictions of the samples belonging to Alzheimer’s disease class to any of the other two classes. 2. For the MCI class, there must not be any mispredictions of samples belonging to the MCI class to the healthy control class. 3. The recall of the healthy control class must be more than 85%. Therefore, based on the above hypothesis, the performance of the 3D convolutional neural network model was evaluated. For the evaluation purpose, five different classification performance metrics were considered namely accuracy, precision, recall, f1-score, and confusion matrix. Also, for determining the generalizability of the model over unseen data, a five-split cross-validation was performed. The details regarding the evaluation of the performance metrics are described in the results section.

22

S. Chakraborty et al.

Fig. 3 3D-convolutional neural network architecture

3.3 Model Optimization Hyperparameters and Loss The development of the 3D CNN architecture is indeed the most important aspect of the work. But more essentially, the component that needs to be considered carefully for creating the learning algorithm is choosing the right set of hyperparameters for

Early Detection of Alzheimer’s Disease from 1.5 T MRI …

23

optimizing the internal set of parameters of the network such as weights and biases, and the loss function. The process of controlling the training process is considered very important while creating a deep learning model. The process is undertaken by the hyperparameters of the optimizer function that is responsible for tuning the optimizer algorithms. For the present study, the primary aspect that lies in the optimization algorithm is to minimize the validation and the testing error of the model. For performing the specific task, the hyperparameters that reside outside the primary deep learning model must be tuned in such a way we generate the perfect internal parameters of the model that are the weights and biases. But the challenge that is faced in the process is that the hyperparameters must be chosen in a particular way that it should be model-specific rather than training set specific to increase the generalizability of the model over unseen data. Therefore, for choosing the perfect set of hyperparameters to maintain the overall model generalizability and optimum objective score, Bayesian sequential model-based optimization (SMBO) is used. Bayesian SMBO is an algorithm used for hyperparameter optimization the works on minimizing an objective function by creating a surrogate model (probability function) based on the evaluation results of the previous objective function. The basic objective function of the Bayesian SMBO is given by. P(score|hyperparameters) =

P(hyperparameters|score)P(score) a. P(hyperparameters)

(1)

The surrogate model that is developed by the Bayesian SMBO is considered to be less expensive than the main optimizer function [6]. Further, the next set of evaluation results is selected by using the expected improvement criterion [7]. The criterion is defined as E I (x) = E max f (x) − f ∗ , 0

(2)

where x belongs to the hyperparameter values and considered to be an improvement in the objective score of f (x) and f * is the maximum value of the objective score found in the process. Further, in the process, AdaDelta [8] is chosen as the optimizer algorithm for optimizing the weights and biases of the network. AdaDelta is considered to be a very robust algorithm relating to the gradient descent algorithm. The algorithm dynamically adapts over due course of the training process by leveraging only first-order information. Moreover, the algorithm does not require any manual tuning of the learning rate and is very robust towards noisy gradient information. Therefore, Bayesian SMBO was applied to the algorithm to generate the optimum hyperparameters and is mentioned below. Learning rate : 0.07423; ρ : 0.751; : 1.0 Another very integral part of the deep learning models is the loss functions. These functions are typically used to determine the variability between the prediction and

24

S. Chakraborty et al.

the true value. The output of the loss functions is non-negative values that increase the generalizability of the model by decreasing the loss [9]. The loss function of a model is given by: L(θ ) =

n 1 L(y (i) , f x (i) , θ n i=1

(3)

where θ represents the parameters of the model, x represents the feature matrix and y represents the true values for a particular set of features. The loss function used in the work is the categorical cross-entropy loss which is also known as the SoftMax loss. The categorical cross-entropy loss determines the performance of a model whose output is a probability. In the categorical crossentropy loss function, each prediction is compared to the actual class value, and a score is calculated. The score is further used to penalize the probability of the prediction based on the difference from the actual value. The penalty that is offered to the predicted value is purely logarithmic in nature where a small score is been allotted to tiny differences and the huge score is allotted to larger differences [10]. The equation for the categorical cross-entropy loss is given by: 1 1 y ∈C log(Pmodel [yi ∈ Cc ]) − N i=1 c=1 i c N

C

(4)

where the double sum has been performed on the ith data samples ranging from 0 to N and the classes which range from 0 to C. The term in the equation,1 yi ∈Cc acts as the indicator function for the ith observation for the Cth category. The term Pmodel [yi ∈ Cc ] is the prediction probability for the ith observation in the Cth class.

4 Results The 3D convolutional neural network model developed in the work presented decent results in terms of detecting Alzheimer’s disease from brain MRI scans. The model developed in the work quantitatively presented effective results by prompting an average recall and precision of 0.9553 and 0.9411 for all three classes, respectively. Also, for the training procedure, a five-split cross-validation with the ratio of 80:20 was performed over the complete dataset, and it was found that all the data splits showcased a constant tendency toward the testing accuracy. Table 5 plots the results of the five-split cross-validation where were used to determine the generalizability of the 3D convolutional neural network model over unseen data. Figure 4 demonstrates the confusion matrix that was generated based upon the results received from the best performing split of the 3D CNN architecture. Also, from the confusion matrix, it can be observed that the results completely align with the initial hypothesis which states the recall of the samples belonging to the Alzheimer’s

Early Detection of Alzheimer’s Disease from 1.5 T MRI …

25

Table 5 Performance evaluation of 3D CNN model Metrics

Split 1

Split 2

Split 3

Split 4

Split 5

Accuracy (%)

90.36

94.65

95.88

95.37

92.70

Precision

0.906

0.9347

0.951

0.9577

0.928

Recall

0.9135

0.9481

0.9601

0.9436

0.9234

F1-score

0.9107

0.938

0.9538

0.95

0.9162

Fig. 4 Confusion matrix of third split cross-validation

disease class needs to be 100%, in MCI class, there should be no mispredictions in the control class, and recall of the healthy class should be more than 85%. Another very important factor that needs to be measured for evaluating the performance of the deep learning models is the interpretability of the models. The field of health care is considered to be a critical field when it comes to the implementation of automated intelligent systems. So, the prime requirement that needs to be provided out of the model is the interpretation behind a particular prediction or causal-effect information that led to a particular prediction. Therefore, to interpret the predictions of the developed 3D CNN model, class activation map was used [11–13]. Figure 5 shows the class activation map on the sample MRI slices that has been predicted as Alzheimer’s disease. The class activation map shows that the model paid much attention to the region of the hippocampus where the degradation took place.

5 Discussion The study presented in the papers concerns the development of a 3D convolutional neural network architecture for the detection of Alzheimer’sdisease from 1.5T-T1 weighted MRI scans. For performing the study, MRI scans were collected from the ADNI database from three different research groups namely, healthy control, MCI,

26

S. Chakraborty et al.

Fig. 5 Class activation maps for sample slices of MRI scan that has been predicted as Alzheimer’s disease a Axial view b Coronal view c Saggital view

and Alzheimer’s disease. Primarily as discussed that the ADNI is a multicenter study, therefore, the acquired MRI scans had spatial and temporal differences. So, to bring all the MRI scans to the same space, an image registration routine was performed over all the MRI scans. The registration of images was performed using ICBM152-T1w-Nonlinear-Symmetric atlas. Post-registration of the brain MRI scans, a 3D convolutional neural network was developed for the learning intricate patterns in the MRI scans for the detection of Alzheimer’s disease and classifying MRI scans into healthy control, MCI and Alzheimer’s disease category, respectively. Before the development of the model, a hypothesis was designed to evaluate the performance metrics of the model. The hypothesis stated that the recall of the samples belonging to the Alzheimer’s disease class needs to be 100%, in MCI class, there should be no mispredictions in the control class, and recall of the healthy