Medical Information Processing and Security: Techniques and applications
2: Biomedical Image Processing Lab, University of Leicester, UK
Smart healthcare systems, made up of advanced wearable devices, internet of things (IoT) technologies and mobile internet connectivity, allow significant medical information to be easily and regularly transmitted over public networks. Personal patient information and clinical records are stored on hospitals and healthcare centres and can be accessed remotely by healthcare workers. Due to the widespread increase in the sheer volume of medical data being collected and created all the time, it has never been more important to ensure that such information is collected, stored and processed in a reliable and secure manner.
This edited book covers the recent trends in the field of medical information processing, including prediction of complications using machine learning and trends in visualization and image analysis. Further chapters focus on information security and privacy solutions for smart healthcare applications, including encryption of medical information, privacy in smart IoT environments, medical image watermarking and secure communication systems.
Medical Information Processing and Security: Techniques and applications can be used as a reference book for practicing engineers, researchers and scientists. It will also be useful for senior undergraduate and graduate students, and practitioners from government and industry as well as healthcare technology professionals working on state-of-the-art security solutions for smart healthcare applications.
Inspec keywords: health care; deep learning (artificial intelligence); medical image processing; feature extraction; medical information systems; Internet; data privacy; cryptography; image watermarking
Other keywords: image watermarking; deep learning (artificial intelligence); data privacy; medical image processing; medical information systems; cryptography; feature extraction; medical information processing; Internet; health care; security
Subjects: Conference proceedings; Computer vision and image processing techniques; Biology and medical computing; Cryptography; Biomedical measurement and imaging; Information networks; Medical administration; General and management topics; General electrical engineering topics; Patient diagnostic methods and instrumentation; Image and video coding; Data security; Medical and biomedical uses of fields, radiations, and radioactivity; health physics; Neural nets
- Book DOI: 10.1049/PBHE044E
- Chapter DOI: 10.1049/PBHE044E
- ISBN: 9781839535253
- e-ISBN: 9781839535260
- Page count: 457
- Format: PDF
-
Front Matter
- + Show details - Hide details
-
p.
(1)
-
1 Introduction to medical information processing and security: techniques and applications
- + Show details - Hide details
-
p.
1
–6
(6)
With continuous technological advances in cloud environments, wearable devices and the Internet of Medical Things (IoMT), the generation and distribution of medical multimedia has increased [1]. These records, often termed as Electronic Health Data (EHD), are highly sensitive and used by remote healthcare professionals to devise more appropriate diagnostic decisions. Figure 1.1 shows the framework of an IoMT-based smart healthcare system which has three major components, including participants, technologies and management [2, 3]. It uses advanced technologies involving sensors and wearable devices to collect medical information, fast internet services, Internet of Things (IoT) devices, big data, cloud computing and artificial intelligence to securely collect, transmit and manage a large number of medical records [4]. The connecting technologies such as Bluetooth, WiFi, internet, Zigbee, NFS and GPS play a key role in developing the applications for which the healthcare system is considered. Health system management under smart healthcare focuses on database, remote monitoring, networks and security management. The joint effort of these components helps in reducing the cost and risks of critical medical processes, enhances the efficiency of clinical resources and makes the healthcare system more intelligent and smart [2, 3].
-
2 Prediction of complications in spine surgery using machine learning: a Health 4.0 study on National Surgical Quality Improvement Program beyond logistic regression model
- + Show details - Hide details
-
p.
7
–25
(19)
With the advancement of the revolutionary artificial intelligence (AI) technologies, health-care services are rapidly moving toward an intelligent cyber physical system referred to as Health 4.0. In essence, the ability to predict surgical complications is all-important for both surgeons and patients. Recently, the use of machine learning (ML) algorithms for predicting complications has gained much attention. Even though many mature and reliable algorithms exist in the field of ML, the logistic regression (LR) algorithm has been the most widely used in complication prediction. In this study, we used the American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) database to compare the performance of LR to other ML algorithms for predicting complications during spine surgery. The database included 177 681 patients who underwent spine surgery. The occurrence of intraoperative morbidity was relatively low (9.4 per cent) in comparison to the total number of the dataset population, and hence, the dataset under study was considered imbalanced. To thoroughly evaluate and compare the proposed ML algorithms, the dataset was balanced and the algorithms were applied on both the balanced and imbalanced dataset. The results indicated that, in general, no significant difference was found between the performance of LR and random forest (RF), boosted tree (BT), and decision tree (DT).
-
3 Recent trends in histopathological image analysis
- + Show details - Hide details
-
p.
27
–55
(29)
Histopathological images provide a plethora of phenotypic information that forms the basis for proven to be the gold standard for cancer diagnosis and monitoring the progression of the disease in cancer patients. However, such images are challenging to analyse, even for experienced pathologists. Moreover, manual analysis is a tedious and costly task in terms of labour, time, etc. The manual analysis is also affected by intra- and inter-observer disagreement, as reported in several studies. Therefore, computer-aided diagnosis (CAD) systems are being explored to speed up the analysis process. Nowadays, artificial intelligence (AI)-based solutions are quite popular in the medical domain, and deep learning (DL) is becoming the most popular methodological choice for researchers to analyse histopathological images. Usually, feature extraction, image segmentation and histopathological image classification are the popular tasks for which several machine learning (ML) approaches and deep models have been developed. There are also few works that are designed for Internet-of-Things-based applications while also addressing security concerns. Therefore, this chapter briefly presents the recent developments in the automated histopathological analysis of cancer. We further summarise different publicly available datasets and also emphasise the key challenges along with limitations of emerging DL techniques for CAD of cancer. We also provide an insight into possible avenues for future research in this area. It helps the researchers working in this area to leverage the opportunities and challenges that direct towards innovative developments in the field.
-
4 Terahertz imaging in healthcare
- + Show details - Hide details
-
p.
57
–86
(30)
Terahertz (THz) is an electromagnetic spectrum with a frequency range from 0.1 to 10 THz, which is located in between the microwave and infrared regions. The unique features of THz waves make them eligible for use in various medical applications. The THz imaging is one of them, which is mainly based on the analysis and processing of the transmission and reflection spectrum information of the sample. This chapter mainly presents the research status and prospects of several THz medical imaging systems and their applications for medical imaging in biological tissues. As the demand for technology grows, high-performance THz imaging systems are becoming indispensable. The ability of the THz time-domain spectroscopy (THz-TDS) system to extract spectra from amplitude and phase information opens virtually unlimited possibilities for imaging applications. The performance achievements of THz-TDS-based imaging with potential research on its fast-imaging components for solving the existing limitations of imaging speed are also presented. Furthermore, the latest developments of several THz-TDS-based imaging methods, including tomography imaging and near-field imaging, are highlighted with their performance improvement. Additionally, the rapid development of THz-TDS as a highly versatile analytical tool for the characterisation of pharmaceutical materials has drawn considerable attention. One of the greatest biomedical potentials of THz imaging is the use of molecular spectroscopy for diagnostics, which is exponentially advanced and moving closer to progress. It is now evident that different types of biomolecules leave distinctive spectral fingerprints in the THz region, which considerably widens the coverage of its technology application including in-vitro and in-vivo measurements of small molecules of clinical importance in point of care and diagnostic systems. In-vivo molecular imaging is considered the next frontier in medical diagnostics, which would be ideally performed non-invasively. Recent achievements in the field of medical imaging have dramatically enhanced the early detection and treatment of many pathological conditions. THz imaging systems can help in detecting early cancer before it is visible or sensitive to any other identification resources. The THz images can distinguish between healthy tissue and basal cell carcinoma and therefore help in mapping the exact margins of early-stage tumours. By obtaining both frequency and time-domain information, the THz imaging can ensure enhanced detection of cancer and provide sharper imaging and molecular fingerprinting. THz biomedical imaging has become a modality of interest due to its ability to simultaneously acquire both image and spectral information. Advanced digital image processing algorithms are greatly needed to assist in screening, diagnosis and treatment. Finally, we summarise the obstacles in the way of the application of THz biomedical imaging application technology in clinical detection, which need to be investigated and overcome in the future.
-
5 The current state of summarization and visualization in Electronic Health Record (EHR) based on EHR interoperability
- + Show details - Hide details
-
p.
87
–123
(37)
Recently, several health-care organizations store heterogeneous health information about patients aiming to improve the quality of health care. The Electronic Health Record (EHR) contains a huge amount of patients' information making it difficult and time-saving to find the most pertinent information. Accurate, concise, and automated summarization and visualization have the potential to save time by increasing patient safety, improving efficiency, helping clinical decision-making, and reducing medical error as well as costs. Although interoperability and standardization are considered keys to improve the quality of care services and to coordinate care and practice effective summarization, several studies have shown the difficulty of improving the quality of health care using the current summarization- and visualization-based systems since they lack interoperability and do not allow to easily express clinician needs. We found that there is no study that discusses the impact of semantic and syntactic interoperability on the EHR summarization approach, which motivated us to provide and discuss studies on the above topics. In this study, we will review health-care summarization and visualization approaches and systems and analyze the proposed studies according to interoperability and clinicians' needs and challenges. To construct our review, we adopted the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodology and examined papers between 1980 and 2021. Selected studies focus on health-care sub-areas, EHR visualization, EHR summarization, interoperability, and standards. Based on the above papers, we provide a systematic view of development in this field and possible future directions. We conclude that most research studies in summarizing systems lack semantic interoperability and do not rely on clinicians' needs. Besides, EHR visualization systems lack the ability to analyze efficiently health data and integrate expert knowledge domains in the decision-making process. This will promote new research to solve these issues.
-
6 EEG signal classification using robust energy-based least squares projection twin support vector machines
- + Show details - Hide details
-
p.
125
–148
(24)
Electroencephalogram (EEG) signals have been successfully employed in the diagnosis of several neurological disorders such as epilepsy and sleep disorders. The classification of EEG signals has been done via support vector machines (SVMs) and twin support vector machines (TWSVMs). However, both SVM and TWSVM solve quadratic programming problems (QPPs) that require an external toolbox for the optimisation. To overcome this limitation, we propose a robust energy-based least squares projection TWSVM (RELSPTSVM) for the classification of EEG signals. Unlike TWSVM-based models that generate hyperplanes for each category of samples, the proposed RELSPTSVM generates a projection axis for each class in a manner that the data points of the corresponding category are proximal to its mean and data points of other categories are as farthest as possible. Unlike least squares TWSVMs (LSTSVMs) that put the samples of another category to be exactly at distance, the proposed RELSPTSVM model relaxes this constraint via energy parameters, which results in more robustness to noise and outliers. The proposed RELSPTSVM model implements the structural risk which results in avoiding the overfitting issues. Experimental results on EEG signal classification and UCI benchmark datasets demonstrate the effectiveness of the proposed RELSPTSVM model.
-
7 Clustering-based medical image segmentation: a survey
- + Show details - Hide details
-
p.
149
–173
(25)
Image segmentation is a popular phenomenon to dividing a digital image into various meaningful and disjoint segments based on certain criteria. It is used to represent the original representation of the image into a more meaningful representation which can be easily analysed. The segmentation of medical images is one of the key steps for analysis of the medical images. It has been gaining much attention due to its application in clinical diagnosis. Computer-aided image segmentation not only reduces the effort of medical practitioners but also reduces the chances of error due to the human intervention. Proper segmentation of medical images leads to accurate diagnosis of the disease which may lead to appropriate treatment. A lot of algorithms have been suggested to segment an image into its most informative form but image segmentation still remains quite a challenging problem. Out of all these segmentation algorithms the clustering-based segmentation algorithms are quite popular to segment both low- and high-dimensional medical images. These algorithms segregate the input image into a finite number of clusters each having a group of pixels based on certain criteria. In this chapter, the different clustering-based algorithms proposed for medical image segmentation have been reviewed. Concise overview of different clustering algorithms proposed for segmentation of medical images is presented along with their key benefits and limitations.
-
8 Artificial intelligence for genomics: a look into it
- + Show details - Hide details
-
p.
175
–189
(15)
The latest progress in genomics and artificial intelligence (AI) sees both disciplines work together to improve results in the relatively new medical area called precision medicine. This chapter aims to provide readers with a review of AI techniques ingesting genomics data to extract patterns and high-level information. Many new and sophisticated AI architectures have been introduced in the scientific community since the release of the famous Human Genome Project. The latter was delivered in 2003 and allowed sequencing and mapping of all the genes of our species (Homo sapiens). Genomics has paved the way for deeper insights into correlations between changes in DNA sequences and diseases. As described throughout the sections in the manuscript, deoxyribonucleic acid (DNA) sequences are big-sized. This feature makes them suitable for investigation through both machine and deep learning (DL) methods. Moreover, the disruptive advent of DL in the scientific community pushed the bar for achievable accuracy rates in many tasks. Genomics makes no exception. AI methods have been primarily employed to tackle some tasks for biomedical image analysis: detection, classification and segmentation of suspicious regions from MRI (Magnetic Resonance Imaging), PET (Positron Emission Tomography) and CT (Computer Tomography), to mention some, have been broadly addressed using machine learning (ML) and DL approaches. Over the last few years, there has been an exponential spike in the number of DL techniques for genomics. Nowadays, genomics and AI are closely twisted in the attempt to achieve ambitious objectives, such as predicting treatment outcomes to deliver patient-tailored therapies, biomarker discoveries, radiotherapy responses and predicting drug effectiveness from cancer genomic signature. The main goal here is to check through the current state-of-the-art AI methods for genomics spanning the most challenging aspects of today's landscape.
-
9 Research on security of anonymous communication in wireless healthcare online system
- + Show details - Hide details
-
p.
191
–227
(37)
The wireless medical sensor network is one of the emerging technologies that brought a revolution in medical image, healthcare monitoring, etc., playing a vital role in wireless healthcare online systems. However, security risks like privacy disclosure may arise since the medical sensors are connected via wireless. Secure anonymous communication is essential to protect patients' privacy. To this end, we give a seminal summary and comparison of the advanced anonymous communication technologies on wireless healthcare online systems. First, in order to solve the problem of patient privacy disclosure caused by health data disclosure, let us look at a privacy-sensitive public key infrastructure (PKI) model with advanced and advanced security that prevents disclosure of user registration keys. Next, to address the problem of spoofing users when accessing wireless sensors and to ensure that an adversary cannot modify sensor data through interception, look at key agreements and anonymous authentication protocols to address spoofing when users access wireless sensors (删掉). Furthermore, given the unreliability of wireless sensors and the hidden peril of health data transmitting in plaintext, a trust-based secure directed diffusion routing protocol is elaborated to establish a reliable routing and secure the transmission of health data. Finally, we expound on a lightweight anonymous communication model in wireless healthcare online systems, which realizes user identity authentication and the health data's confidentiality, so as to protect the privacy of patients.
-
10 A comprehensive study on the security of medical information using encryption
- + Show details - Hide details
-
p.
229
–260
(32)
The exponential rise in the use of the Internet and rapid progress in computer technology has revolutionised healthcare allowing medical information to be shared efficiently across networks. While these advancements make it convenient to store and share information, they also have their limitations. Medical data contain highly sensitive and confidential information about patients and can be subjected to various kinds of passive and active attacks and analyses if transferred over an insecure network. These data leakages can be catastrophic for the respective individual or organisation due to the undesirable scenarios which emanate from them. Hence, information security is a major concern and warrants developing techniques to securely transfer medical data.
This chapter discusses information security techniques and their properties in detail, reviews the research related to cryptography-based medical information security by categorising the techniques into traditional, hybrid and deep learning-based approaches and highlights their limitations. Comparative studies based on the universality of the method and metric-wise performance are also presented. We further discuss the potential challenges and prospective future directions, enabling a path forward to advance the research in this area.
-
11 Electrocardiogram-based dual watermarking scheme for healthcare applications
- + Show details - Hide details
-
p.
261
–277
(17)
The strong need for a copy-protection scheme for healthcare data has been well recognized. Research studies show that invisibility, embedding capacity, and robustness are some of the most common watermarking features, which makes it difficult to balance the contradictions between them. Inspired by dual watermarking where multimarks are concealed inside the carrier media to offer a robust and secure model. In this chapter, we propose a watermarking scheme which embeds dual marks into a host electrocardiogram (ECG) signal using a transform-domain approach. First, we consider the redundant discrete wavelet transform (RDWT) to decompose the 2D signal into non-overlapping sub-bands, and then the selected sub-bands are decomposed by fast Walsh Hadamard transform (FWHT) followed by QR decomposition to compute the coefficient for data embedding purposes. Second, we apply the sequence of RDWT-FHWT-QR to decompose both marks in a similar manner. Third, we modify the QR coefficients of the host signal with the coefficients of both marks. The QR code of the more robust mark (patient report) is generated to ensure its security against attack. Extensive analysis shows good performance of the proposed algorithm. Our experiments on ECG signals demonstrate that the robustness of our algorithm is significantly improved up to 63.94% when compared with traditional schemes.
-
12 Application of autoencoder in craniofacial reconstruction of forensic medicine
- + Show details - Hide details
-
p.
279
–305
(27)
Craniofacial reconstruction is an important task of forensic medicine, which is to estimate the face from its skull. As a popular generation model in deep learning, autoencoder (AE) has attracted the attention of researchers because of their good ability to extract data features and representations. AE network can learn features from the data samples by unsupervised method on the training data. Based on the investigation of generating three-dimensional faces by AE methods, this chapter introduces in detail the methods and frameworks of traditional AE models, as well as some applications in craniofacial reconstruction and face generation after the model has been improved. We summarize the development and research status of AE models in recent years. In addition, this chapter compares and analyzes these AE models from many aspects. Furthermore, the future direction of face generation is pointed out, which will promote the technology of craniofacial reconstruction to be applied in the identification of unknown corpses in forensic medicine, medical plastic surgery and many other fields.
-
13 Security and Privacy in smart Internet of Things environments for well-being in the healthcare industry
- + Show details - Hide details
-
p.
307
–323
(17)
Privacy protection is required when communicating data in the healthcare system. The Internet of Things (IoT) in healthcare has numerous advantages, including the potential to more closely monitor patients' health and the use of data for analytics. IoT is a framework of interconnected, web-connected devices that may collect and transmit data via a wireless server without the need for human involvement. In this context, IoT-based healthcare uses many technological advances to give several services such as quick and efficient treatment, savings, and better communication. Wireless Body Area Network (WBAN) technology can improve the performance of data communication in smart systems. Throughout each stage of smart medical systems, machine learning (ML) can be applied. In this study, the most current research, suggested approaches, and existing smart healthcare system technologies are discussed in terms of technological advances, applications, and difficulties to provide a proper overview of what IoT signifies in the healthcare sector now and in the future.
-
14 A survey of medical image watermarking: state-of-the-art and research directions
- + Show details - Hide details
-
p.
325
–360
(36)
With the growth and popularity of the utilization of medical images in smart health care, the large amount of health records and related images, which needs to be transmitted, is consistently increasing. However, exchanging these medical records among hospitals, doctors, and medical team faces many challenges associated to ownership conflicts, data security, and privacy. Medical image watermarking is an effective copyright protection tool by embedding copyright/secret information/logo into the carrier image. This chapter provides a detailed overview of watermarking along with its characteristics and current applications. We then provide a comparative survey of the different state-of-the-art approaches along with their merits and limitations. We further summarized each of the state-of-the-art approaches in detail, including objectives, goals, dataset used, evaluation metrics, and weaknesses, and discussed the recent challenges and their possible solutions. We believe this chapter will provide the readers a comprehensive insight to the field of watermarking research in the health care domain.
-
15 Secure communication and privacy preserving for medical system
- + Show details - Hide details
-
p.
361
–389
(29)
The remote medical system has been greatly developed with the help of 5G and Internet of things (IoT) technology. People, especially those living in areas lacking excellent medical resources, benefit from the remote medical system. There are a lot of data transmitted in the channel, which leads to many security and privacy anxiety. It may include: (1) the communication between users and doctors should be confident and authenticated; (2) medical equipment needs to authenticate the doctors' identity; (3) doctors and patients need a secure and private way to connect the database; and (4) patients data privacy should be protected. For solving these anxieties, a lot of security and privacy protocols, such as authentication schemes, privacy-preserving schemes, or n-sources anonymity schemes, are proposed. In this chapter, some of these security and privacy schemes are introduced, which are important and excellent. Compared with other similar protocols, the proposed protocols have advantages in terms of efficiency, safety, dynamics, etc. We believe that these security and privacy schemes can solve the above anxiety and inspire more related works.
-
16 Secure medical image encryption algorithm for e-healthcare applications
- + Show details - Hide details
-
p.
391
–420
(30)
In recent years, with the constant evolution of the e-healthcare and in particular cloud services, medical images have taken a major part in data transmission. This medical data must be compressed and protected against illegal access and fraudulent usage. To ensure data confidentiality, a secure medical image encryption algorithm for e-healthcare applications is proposed in this chapter. In the first stage, precision limited logistic map (LM) and skew tent map (STM) is adopted to encrypt plain medical image, obtaining the cipher messages with a relationship to the plain image. Here, the method uses secure hash algorithm (SHA-512) to generate the initial sequences and parameters of the LM and STM. Then, Lempel-Ziv-Welch (LZW) compression scheme is utilized to compress the cipher image, which is used to reduce communication costs and storage space. According to our obtained results, the proposed encryption before compression algorithm is effective and resist to the many attacks on image and Kaggle dataset. Furthermore, extensive experimental results on real-world datasets demonstrate that the proposed algorithm outperforms the state-of-the-art approaches.
-
17 Conclusion and future directions
- + Show details - Hide details
-
p.
421
–424
(4)
With the development of Internet technologies, large volumes of multimedia and more generally, multimodal data, can easily be shared in digital forms among different users or entities for the purpose of diagnosis, analysis and prediction in healthcare. To address these challenges, this book presented some important medical information processing and security studies and approaches for smart healthcare applications.
-
Back Matter
- + Show details - Hide details
-
p.
(1)