Deep Learning and IoT in Healthcare Systems

Paradigms and Applications

Krishna Kant Singh, Akansha Singh, Jenn-Wei Lin, Ahmed A. Elngar

• Publisher: Apple Academic Press
• Year: 2021
• ISBN:
  9781000089189
• Edition: 1
• Topic: Informatique
• Subtopic: Intelligence artificielle (IA) et sémantique

This new volume discusses the applications and challenges of deep learning and the internet of things for applications in healthcare. It describes deep learning techniques in conjunction with IoT used by practitioners and researchers worldwide.
The authors explore the convergence of IoT and deep learning to enable things to communicate, share information, and coordinate decisions. The book includes deep feedforward networks, regularization, optimization algorithms, convolutional networks, sequence modeling, and practical methodology. Chapters look at assistive devices in healthcare, alerting and detection devices, energy efficiency in using IoT, data mining for gathering health information for individuals with autism, IoT for mobile applications, and more. The text also offers mathematical and conceptual background that presents the latest technology as well as a selection of case studies.

CHAPTER 3 Case Studies: Healthcare and Deep Learning

ASHISH TRIPATHI¹*, ARUN KUMAR SINGH², K. K. MISHRA¹, PUSHPA CHOUDHARY¹, and PREM CHAND VASHIST¹

¹Department of Information Technology, G. L. Bajaj Institute of Technology and Management, Greater Noida, India

²Department of Computer Science & Engineering, Motilal Nehru National Institute of Technology, Allahabad, India

*Corresponding author. E-mail: [email protected]

ABSTRACT

As the size of the medical data is increasing day by day, the traditional techniques of data analysis are becoming inefficient to provide accurate and valid information on time. Therefore, to overcome the limitations of the traditional techniques, deep understanding and analysis capabilities are required to detect and diagnose the disease in the early stage. In recent years, several artificially intelligent techniques have been developed for the analysis of different diseases, such as cancer, diabetes, Alzheimer’s diseases, and lots more. In this context, deep learning techniques are playing a significant role in analyzing the medical dataset for faster detection and diagnosis of diseases. In this way, these techniques are helping medical research and practitioners working in the healthcare industry. This chapter presents the role of deep learning techniques in the healthcare system. The role of deep learning has been discussed in the analysis and support in clinical decisions and diagnosis of medical images. A tabular detail of the various deep learning techniques applied in the healthcare system has been shown. In the end, we have presented the case study of cancer, diabetes, and Alzheimer’s diseases.

3.1 INTRODUCTION

In the current era, healthcare requires technology-enabled smart techniques to manage and analyze abundant biomedical data to ensure the correct measurement of diseases and also recommend the right treatment to the right patient at the right time (Collins and Varmus, 2015). For this, complete information containing several aspects of patient’s data is required.

In health-care system, the availability of the huge amount of medical data presents ample opportunities and challenges for biomedical researchers to ensure the availability of better treatment to the patient at minimum cost with improved efficiency. In this context, exploring the association among all the information obtained from different data sets plays a significant role in developing reliable techniques based on machine learning (ML). Previously, it had been tried to build joint knowledge base based on linking multiple data sources with the aim to do predictive analysis and discovery of diseases, and to provide the best treatment for them (Xu et al., 2014; Chen et al., 2015; Wang et al., 2014).

Although, existing deep learning techniques show good result on extracting real-time and valid information with accuracy from medical data set, but still these techniques have not been applied widely on the available medical dataset to get the expected outcome (Bellazzi and Zupan, 2008).

In reality, it has been found that the full use of data is still a big challenge due to its high & multidimensional nature, heterogeneity, temporal dependency and, irregularity (Hripcsak and Albers, 2012; Jensen et al., 2012). Also, it becomes more complicated when the same data is represented in different ways at different places across the data and cause of conflict and inconsistency in decision making (Mohan et al., 2011). In this contest, a continuous effort and innovative techniques are required, which automatically discover the knowledge from the heterogeneous and novel data to ensure the quality of diagnosis in terms of timeliness and accuracy (Bengio et al., 2013).

In recent years, a dramatic growth has been found in the application of deep learning due to massive growth in new datasets and need of computational power (LeCun et al., 2015). Like other domains, healthcare and medicine require deep learning models to process the huge amount of data being generated as well as use of medical devices and digital record system on a large scale (Russakovsky et al., 2015; Hirschberg and Manning, 2015; Hinton et al., 2012; Cireşan et al., 2013).

Deep learning methods could be more appropriate for solving healthcare problems by using its different inbuilt features such as end-to-end learning, learning with multiple levels of representation, good in exploring the high-dimensional data, and capable of handling complex and heterogeneous data. Unlike traditional ML, deep learning does not require human intervention to guide machines, thus no concern of human error and thus it improves accuracy (LeCun et al., 2015; Miotto et al., 2017).

In recent studies, various capabilities of advanced deep learning techniques have been mentioned such as learning from complex and unstructured data (Miotto et al., 2017; Wei et al., 2017), image recognition using deep convolutional neural networks (CNNs), and text categorization by deep belief networks. Major applications of deep learning in medical diagnosis are health informatics, biomedicine (Mamoshina et al., 2016), magnetic resonance imaging (MRI) and CT scans, and ECG (Kumar et al., 015; Pyakillya et al., 2017), and all these are useful in the diagnosis of severe diseases like cancer, heart disease, and brain tumor. Medical applications such as classification, detection, localization, registration, segmentation, image reconstruction, post-processing, and regression are some specific uses of deep learning in medical field.

Deep learning gives better performance in heterogeneous environment as it learns from raw data as compared to ML and other traditional techniques, and its hidden layers support it to learn abstractions from the given inputs (Miotto et al., 2017). The basis of deep learning is the working capability of neural network that applies general purpose learning procedures to learn from data.

3.2 BACKGROUND AND RELATED WORK

3.2.1 PRESENT SCENARIO

As per the report of the Department of Industrial Policy and Promotion, from financial year 2014 (1.2% of GDP [gross domestic product]) to 2018 (1.4% of GDP), approximately 0.2% growth in the expenditure on public health of the total GDP has been found. In spite of that, in India, expenditure on public health is still at the lowest level as compared to many low-income countries like Sri Lanka and Indonesia. So, Government of India is planning to spend 2.5% of GDP on public health by 2025.

As we know the healthcare sector is a fastest growing industry in India and lots of challenges have been found in recent years due to the complex data, such as insufficient domain knowledge, lack of timeliness, and accuracy in diagnosis of diseases (Esteva et al., 2019). Healthcare sector covers many things such as hospitals, medical research, drug discovery, health insurance, telemedicine, and medical equipment. In fact, this sector generates huge revenue and employment and emerging as one of the fastest and largest growing sectors in India. But here one of the major problems is that this also increases the unmanaged biomedical data, which is complex, unstructured, high-dimensional and heterogenous in nature. Thus, to get the fruitful information from the biomedical data, many artificial intelligence (AI) and ML techniques have been applied previously. But due to distinct nature and rapid growth in data, deep learning methods are very much applicable. Deep learning offers analysis of data with speed and precision which is very much required in the healthcare sector and this ability never seen before in another branch of AI (LeCun et al., 2015).

3.2.2 DEEP LEARNING

ML is a subset of AI that helps systems to learn automatically from the environment as well as self-improvement from experience. It makes predictions or decisions without being explicitly programmed. Deep learning is a specialized branch/subclass of ML that applies supervised and unsupervised learning to learn features directly from the data representations. Deep learning performs feature extraction and transformation based on multiple layers of nonlinear processing units. The connection of AI, ML, and deep learning is shown in Figure 3.1.

FIGURE 3.1 Connection of AI, ML, and deep learning.

ML applies statistical and data driven rules to convert the given input into the required output. These rules are automatically generated from a large set of examples without use of the explicit human specification. For feature extraction, domain expertise and human engineering are required in a ML system to detect patterns. While, deep learning applies representation learning for pattern recognition. Multiple layers of representation are used in deep learning. These layers are arranged in a sequential manner and contain a large number of nonlinear and primitive operations. A representation of one layer is used as input for the next layer and it continues until optimal representation is obtained.

Deep learning techniques are scalable to large datasets and as an input it can accept multiple data types as shown in Figure 3.2.

FIGURE 3.2 Representation of deep learning uses a variety of data types for feature extraction.

In deep learning, the network itself capable to perform many tasks that cover filtering and normalization of data th...

Table of contents

• Cover Page
• Half-Title Page
• Title Page
• Copyright Page
• About the Author
• Table of Contents
• Contributors
• Abbreviations
• Preface
• Deep Learning for Healthcare
• Role of AI in Healthcare
• Case Studies: Healthcare and Deep Learning
• Assistive Devices and IoT in Healthcare Functions
• Impact of IoT on Healthcare-Assistive Devices
• Smart Fall Detection Systems for Elderly Care
• Smart Sensors Transform Healthcare System
• Healthcare Applications of the Internet of Things (IoT)
• Mobile-App-Enabled System for Healthcare
• Energy-Efficient Network Design for Healthcare Services
• Applying Data Mining to Detect the Mental State and Small Muscle Movements for Individuals with Autism Spectrum Disorder (ASD)
• Index