Artificial Intelligence for the Internet of Health Things

K. Shankar, Eswaran Perumal, Deepak Gupta

• Publisher: CRC Press
• Year: 2021
• ISBN:
  9781000374339
• Edition: 1
• Topic: Computer Science
• Subtopic: Computer Science General
• Index: Computer Science

This book discusses research in Artificial Intelligence for the Internet of Health Things. It investigates and explores the possible applications of machine learning, deep learning, soft computing, and evolutionary computing techniques in design, implementation, and optimization of challenging healthcare solutions. This book features a wide range of topics such as AI techniques, IoT, cloud, wearables, and secured data transmission. Written for a broad audience, this book will be useful for clinicians, health professionals, engineers, technology developers, IT consultants, researchers, and students interested in the AI-based healthcare applications.

• 
  
  

• Provides a deeper understanding of key AI algorithms and their use and implementation within the wider healthcare sector

• 
  
  

• Explores different disease diagnosis models using machine learning, deep learning, healthcare data analysis, including machine learning, and data mining and soft computing algorithms

• 
  
  

• Discusses detailed IoT, wearables, and cloud-based disease diagnosis model for intelligent systems and healthcare

• 
  
  

• Reviews different applications and challenges across the design, implementation, and management of intelligent systems and healthcare data networks
• Introduces a new applications and case studies across all areas of AI in healthcare data

K. Shankar (Member, IEEE) is a Postdoctoral Fellow of the Department of Computer Applications, Alagappa University, Karaikudi, India.

Eswaran Perumal is an Assistant Professor of the Department of Computer Applications, Alagappa University, Karaikudi, India.

Dr. Deepak Gupta is an Assistant Professor of the Department Computer Science & Engineering, Maharaja Agrasen Institute of Technology (GGSIPU), Delhi, India.

1 Artificial Intelligence (AI) for IoHT – An Introduction

1.1 ARTIFICIAL INTELLIGENCE (AI) IN THE HEALTHCARE DOMAIN

Presently, there is a need for a model with its linked gadgets, persons, and networks completely integrated on the Internet of Health Things (IoHT), which offers medicinal services to patients, monitoring patients, and offering drug recommendations. The IoHT incorporates both technological expertise and electronics. Artificial intelligence (AI) is the application of science-based research and the development of smart machines. The people without the knowledge of intelligent machines conjure images of charismatic human-based systems and robots, as depicted in science ­fiction [1]. The most familiar media reports of applying aerial surveillance drones, driverless cars, and other features of perils in emerging smart machines, which has been improved with common awareness. AI models and approaches are predefined from other formulations. Several domains of AI schemes are available in the ­literature [2].

AI models are applied in automobiles, aircraft guidance fields, smartphone equipment like audio analysis applications such as Apple's Siri, Internet web browsers, and a plethora of alternate practical actions. AI methodologies tend to resolve problems and perform events in stable, effective, and productive types when compared with other possibilities. The nature of mental healthcare domains provides the merits and advancements in AI [3]. For instance, processing models to learning, understanding, and reasoning helps the experts in clinical decision-making, analysis, diagnostics, and so on. AI approaches could be more advanced in self-care devices to enhance people's lifestyles, such as communicative mobile health fields that know the ­patterns as well as priorities of customers. AI results in the enhancement of public health under the assistance of detecting health risks and data inventions. An alternate instance of AI is that virtual humans are capable of communicating with care seekers and can give appropriate remedies to cure the disease. This chapter depicts the chance of applying AI models and methods for healthcare operations in future advancements [4, 5].

The main objective of AI is to develop machines with the potential to perform tasks such as essential intelligence, like reasoning, learning, planning, problem resolving, as well as perception. The relevant fields involved in AI are shown in Fig. 1.1. This domain was named by computer scientist John McCarty, and Marvin Minsky, Nathan Rochester, and Claude Shannon implied it at the Dartmouth Conference. The key objective of this conference was to set a novel domain of science that contributes to the study of modern devices. Each perception of learning intelligence could be so precisely defined that a machine is made to develop it. At the time of the conference, Allen Newell, J.C. Shaw, and Herbert Simon illustrated the Logic Theorist (LT), as the initial computer program manufactured to reflect problem resolving techniques [6].

FIGURE 1.1Relevant fields of AI.

In the last few decades, AI has been developed into multidisciplinary fields with computer science, engineering, psychology, and so on. Few objectives can be accomplished by the application of AI, to develop a framework to achieve remarkable events such as computer vision, audio recognition, and detection of patterns that exist in data [7]. It mainly concentrates on specialized intelligent actions that have been named as Weak AI, also termed as Applied AI. Instead of thinking that humans can play chess, Deep Blue employs the application of brute force approaches to estimate the possibilities to compute the offensive as well as defensive movements. The term “Strong AI” was coined by philosopher John Searle in 1980, and defines the aim of deploying machines with common AI. The major intention of Strong AI is to deploy machines with smart capability that is indistinguishable from humans. Such resources are typically narrow and accurate tasks, namely the function of arithmetic task. AI is utilized to invent the intelligent nature of machines to perform the events of human behavior. AI might be the form of either hardware or software that can stand alone, distributed over the computer networks, or embedded into a robot. Besides, it is in the form of intelligent and independent agents that are capable of communicating with the corresponding platform in the decision-making process. AI is combined with biological operations for brain computer interfaces (BCIs), which is manufactured with biological objectives (biological AI), and small molecular structures referred to as nanotechnology [8].

Several clinical decision support systems (CDSSs) were presented by the use of AI approaches and finds useful in several domains [9]. This chapter offers an introductory explanation of AI concepts, evolution, clinical data generation, and AI techniques in healthcare. This chapter also discusses several applications of AI techniques developed in the healthcare sector.

1.2 EVOLUTION OF AI IN THE MEDICAL SECTOR

Rule-based methods achieved most of the success in the 1970s, which includes interpreting electrocardiograms (ECGs), diagnosing diseases, selecting proper remedies, offering the interpretations of clinical reasoning, and helping doctors in producing diagnostic statements for severe patients. Therefore, rule-based systems are the most expensive to develop and can also be vulnerable since they need explicit presentations of decision rules as well as human-relied updates. Additionally, it is very complex to encode higher-order communications between various pieces of knowledge provided by diverse professionals, and the working function of systems is reduced by an extensive advancement of medical knowledge. Therefore, it is very hard to execute a system that incorporates the deterministic as well as probabilistic reasoning to narrow down the related clinical content, prefer the diagnostic statement, and advanced therapy [10].

In contrast to the first generation of AI, which is based on the advice of medical experts and a formulation of effective decision rules, current AI studies have alleviated machine learning (ML) techniques, which consider the difficult ­interactions to discover the patterns that exist in data. Based on the types of functions that have been induced to resolve, fundamental ML techniques were classified into two classes: supervised and unsupervised [11]. The supervised ML approach was operated by gathering a large number of “training” cases that were comprised with inputs like fundus images and target output labels such as existence or lack of diabetic retinopathy (DR). By the pattern recognition of labeled input and output pairs, the method attempts to generate specific output for the provided input. Also, it has been developed for identifying the best parameters to reduce the deviations among the predictions for training cases and monitored results. The generalizability of a method could be evaluated under the application of the test set. Classification, regression, and characterization of similarity between instances of the same results are vastly applied functions of supervised ML techniques.

Unsupervised learning acquires basic patterns from unlabeled data to discover subclusters of actual data, to find outliers of data, and generate low-dimensional presentations of data. It is pointed out that the discovery of minimum dimension representations for labeled instances has to be more efficient to attain a supervised model. Here, ML approaches tend to develop the AI domains that have been served as the discovery of existing unknown patterns in data with no requirement of particular decision rules for every specific task that assumes for difficult communications between the input features. ML has been one of the applied systems to deploy AI units.

Several types of intelligent neural networks (NNs) are comprised of a maximum number of layers. NNs have a massive number of layers capable of modeling tedious relations from input and output where it requires more data, processing duration, and latest structural designs to reach optimized functions. Various layers, numerical tasks for neurons, and regularizing models were deployed. For instance, convolutional layers are more helpful for extracting spatial or temporal relations, while recurrent layers apply circular links to design the temporal actions. Besides, diverse types of initialization and activation functions are used in improving the method's function. The integration of such units activates the NN to manage multiple inputs with and without spatial or temporal basis. A smart NN could be constrained with a higher number of attributes and consume large computational resources for training purposes.

The improvising application of electronic health record (EHR) systems is comprised of a set of large-scale medicinal data and enables a smooth combination of AI models as a clinical workflow (Fig. 1.2). Traditionally, medical experts gather the medical data from patients to make a clinical decision and save the reports and treatment plans as health records (Fig. 1.2a). The decision support system (DSS) gathers the medical-related data to provide suggestions for doctors (Fig. 1.2b). There are several methods to combine the DSS into the clinical workflow; for example, the DSS is capable of collecting data from patients and EHRs, provides suggestions to physicians, and archives in a system output of EHRs (Fig. 1.2c). Several approaches are associated with complete automatic clinical systems, such as the independent tools gather data from patients to make decisions and produce the outcome into EHRs (Fig. 1.2d). Data from EHR models give brief data regarding patients, such as medicinal notes as well as laboratory measures, activating the field of natural language processing (NLP) approaches to obtain codified vocabularies [12].

FIGURE 1.2Flow of information in healthcare.

1.3 USE OF AI DEVICES FOR CLINICAL DATA GENERATION

Before the existence of an AI system in a healthcare field, it has to undergo training with the help of data produced from clinical events, such as screening, diagnosis, treatment schedule, and so on; thus, it can acquire the data from the same set of subjects, corre...

Table of contents

• Cover
• Half Title
• Series Page
• Title Page
• Copyright Page
• Contents
• Author Biographies
• Preface
• Chapter 1 Artificial Intelligence (AI) for IoHT – An Introduction
• Chapter 2 Role of Internet of Things and Cloud Computing Technologies in the Healthcare Sector
• Chapter 3 An Extensive Overview of Wearable Technologies in the Healthcare Sector
• Chapter 4 IoHT and Cloud-Based Disease Diagnosis Model Using Particle Swarm Optimization with Artificial Neural Networks
• Chapter 5 IoHT-Based Improved Grey Optimization with Support Vector Machine for Gastrointestinal Hemorrhage Detection and Diagnosis Model
• Chapter 6 An Effective-Based Personalized Medicine Recommendation System Using an Ensemble of Extreme Learning Machine Model
• Chapter 7 A Novel MapReduce-Based Hybrid Decision Tree with TFIDF Algorithm for Public Sentiment Mining of Diabetes Mellitus
• Chapter 8 IoHT with Artificial Intelligence–Based Breast Cancer Diagnosis Model
• Chapter 9 Artificial Intelligence with a Cloud-Based Medical Image Retrieval System Using a Deep Neural Network
• Chapter 10 IoHT with Cloud-Based Brain Tumor Detection Using Particle Swarm Optimization with Support Vector Machine
• Chapter 11 Artificial Intelligence-Based Hough Transform with an Adaptive Neuro-Fuzzy Inference System for a Diabetic Retinopathy Classification Model
• Chapter 12 An IoHT–Based Intelligent Skin Lesion Detection and Classification Model in Dermoscopic Images
• Chapter 13 An IoHT-Based Image Compression Model Using Modified Cuckoo Search Algorithm with Vector Quantization
• Chapter 14 An Effective Secure Medical Image Transmission Using Improved Particle Swarm Optimization and Wavelet Transform
• Chapter 15 IoHT with Wearable Devices–Based Feature Extraction and a Deep Neural Networks Classification Model for Heart Disease Diagnosis
• Index