A Review of Medical Image Watermarking Requirements for Teleradiology

Abstract

Teleradiology allows medical images to be transmitted over electronic networks for clinical interpretation and for improved healthcare access, delivery, and standards. Although such remote transmission of the images is raising various new and complex legal and ethical issues, including image retention and fraud, privacy, malpractice liability, etc., considerations of the security measures used in teleradiology remain unchanged. Addressing this problem naturally warrants investigations on the security measures for their relative functional limitations and for the scope of considering them further. In this paper, starting with various security and privacy standards, the security requirements of medical images as well as expected threats in teleradiology are reviewed. This will make it possible to determine the limitations of the conventional measures used against the expected threats. Furthermore, we thoroughly study the utilization of digital watermarking for teleradiology. Following the key attributes and roles of various watermarking parameters, justification for watermarking over conventional security measures is made in terms of their various objectives, properties, and requirements. We also outline the main objectives of medical image watermarking for teleradiology and provide recommendations on suitable watermarking techniques and their characterization. Finally, concluding remarks and directions for future research are presented.

Introduction

Recent technological advances introduced a radical change in the modern health care sector including medical imaging facilities, hospital information system (HIS), and information management systems in hospitals. Changes in medical imaging facilities in radiology have acquired sufficient reliability and cost-effectiveness that the film-based imaging technology has been shifted to filmless techniques for producing digital images on various devices rather than generating hardcopies. With the use of these digital medical images, in addition, HIS comprising radiology information system (RIS) and picture archiving and communication system (PACS) [1] has facilitated offering various eHealth services. These eHealth services are introducing new practices for the profession as well as for the patients by enabling remote access, transmission, and interpretation of the medical images for diagnosis purposes. This has made easy the widespread use of teleradiology with the potential to improve healthcare access, delivery, and standards, where complex and new legal and ethical issues are also raising. These issues include image retention and fraud, privacy, malpractice liability, licensing and credentialing, and contracts for PACS, RIS, and teleradiology [2].

In teleradiology, one of the most successful eHealth services at present, security and privacy protection has become a critical issue [3, 4]. In this study, we mainly focus on the teleradiology that essentially captures a broad range of security requirements along with other radiological information management issues including that of its original medical specialty, radiology. When radiology employs the use of imaging to both diagnose and treat disease visualized within the human body, teleradiology has been for a long time understood to be an eHealth service done through remote transmission of the radiology images and information over electronic networks, and the interpretation of the transmitted images for diagnosis purposes [3]. Remote access and transmission of the images and other radiology information, particularly, electronic personal health information (EPHI), expose them to possible tampering or theft with serious ramifications, since they are sensitive and in most cases EPHI are identifiable. Such radiology images and information not only require protection with integrity and high confidentiality but also appropriate management through different healthcare services.

Providing the required security and privacy of the radiology information requires the following: (1) a standard set of security and privacy profile/policy for teleradiology and (2) a set of security measures by which the security principles in the profile are fulfilled. Various national and international legislative rules and directives define the security and privacy requirements of medical information. These requirements are being achieved by different conventional measures, which are thought to be incapable of providing the required security of the electronic radiology information in the PACS/RIS-based teleradiology [5–7]. On the other hand, recent studies show the possibility of using digital watermarking for improving security in teleradiology [8–16].

Digital watermarking has various attractive properties to complement the existing security measures that can offer better protection for various multimedia applications [17]. However, it is particularly important to know the applicability of digital watermarking from every aspect of radiology information requirements and the suitability of that over other (both the existing and developing) similar measures. Although Coatrieux et al. [7, 18] studied the applicability of digital watermarking in medical imaging, a further justification of the watermarking considering the security requirements in teleradiology is still necessary.

The rest of the paper is organized as follows. General security and privacy requirements of, and expected threats for, the medical information from the perspective of different security and privacy profiles/policies are reviewed and presented in section “Security and Privacy Requirements in Teleradiology.” The limitations of the conventional security measures to handle those threats are also studied and discussed there in. The section on “Digital Watermarking in Teleradiology” introduces briefly the digital watermarking and its various benefits. Justification over other comparable measures, various properties, objectives, suitable types, and their requirements of watermarking for medical images are also given. Concluding remarks and discussion are given in the section on “Discussion and Conclusions.”

Security and Privacy Requirements in Teleradiology

Security and Privacy Standards

Medical information security requirements are generally defined by the strict ethics and legislative rules of the security policy/profile, and concerned entities must adhere to them. There are many widely used guidelines and standards for protecting personal health information. The basic international standard developed for security management of health information is the ISO27799 (Security Management in Health Using ISO/IEC/17799) [19]. The standard itself provides guidance to health organizations and other holders of personal health information on how to protect such information via implementation of ISO17799/ISO27002. It specifically covers the security management needs in this sector, with respect to the particular nature of the data involved.

Some countries have their own security and privacy policy; for example, USA’s Health Insurance Portability and Accountability Act (HIPAA) [20], Code of Federal Regulations number 45 (CFR 45) [21], and Europe’s Directive 95/46/EC [22] are expressions of such a constraint. The HIPAA requires all the cover entities (i.e., health plans, health care clearinghouses, and healthcare providers) to take measures to ensure the security of medical images to protect patient’s privacy. Directive 95/46/EC states the legislative rules on the protection of individuals with regard to the processing of personal data and on the movement of such data. In addition, the CFR 45 (part 164: security and privacy) includes a set of standards for the protection of sensitive EPHI.

There is no specialized standard similar to HIPAA or CFR 45 in Australia at this time, although it does seem likely that a similar set of standards will eventually be required in the future, if online and electronic health records are to be appropriately protected [23]. As the government regulations in relation to privacy grow throughout the world, it forces the security of medical images to grow also. However, the Australian Law Reform Commission [24] produced the Australian Privacy Law and Practice Report that is a comprehensive review of the Privacy Act of 1988 (Australian Law Reform Commission, 2008). That review incorporates privacy regulations on electronic health information systems.

Besides, the Digital Imaging and Communication in Medicine (DICOM) [25] was conceived in 1983 by a joint committee formed by the American College of Radiology (ACR) and the National Electrical Manufacturers Association (NEMA). Early standards did not gain universal acceptance among manufacturers. In 1993, ACR-NEMA version 3.0 was released, and at this time, the standard was renamed DICOM 3.0. This version of that standard has become universal within radiology and been adopted in other medical fields such as dentistry, pathology, and cardiology [26]. It is now commonly known as simply the DICOM standard, an 18-part document. This standard aims to define a technical framework for application entities involved in the exchange of medical data to adhere to a set of security profiles. DICOM also warrants the inclusion of the imaging information for the electronic health record systems and digital signatures for checking the integrity of medical images.

Medical Information Security Requirements

The standards and their technical frameworks, strict ethics, and legislatives rules, as mentioned above, give rights to the patient and duties to the health professionals. Development and implementation of the security and privacy protection services derived from the standards depends upon the model or infrastructure of the teleradiology and its concerned entities. Two widely used models in today’s teleradiology are referred by Ruotsalainen [3] to develop their security requirements. The most common model used in teleradiology is based on offline messaging. The other model incorporates the online delivery of distributed imaging services and allows a radiological information system to be spread over a large distributed area. Irrespective of the communication type (i.e., offline or online), three individual domains, namely: (1) host organization/hospital’s PACS/RIS (domain A), (2) communication network (domain B), and (3) consultant (domain C) can be considered from Fig. 1, which are responsible for providing the required security in a teleradiology system. On the other hand, in radiology, security concerns arise only from the domain A (e.g., from acquisition of medical images to storing them in PACS of the same hospital). Therefore, as we mentioned in “Introduction,” the security requirements of teleradiology also include the security requirements of radiology.

Fig. 1.

Teleradiology model

In an offline model, the security domains are isolated, and communication is made via interfaces whereas online teleradiology maintains communication with a remote consultant allowing access to the local PACS/RIS services of the legacy system [3]. However, based on the technological and organisational models used in teleradiology, their various security requirements can be outlined below [3, 27]:

• All concerned entities/domains (e.g., the PACS/RIS in hospital or clinic, communication network, and consultant/radiologist at distant place) must have the same level of security and protection.

• In all domains, proper authorization process must be employed through various access and user controls, transmission controls, and directive controls.

• Integrity, authenticity, and confidentiality of all radiological information have to be ensured during teleradiology session, consultation process, and information processing, management, and preservation.

The principle of those requirements imposes three mandatory characteristics for security of medical information [7, 27, 28]: confidentiality, reliability, and availability.

• Confidentiality— ensures that only the entitled users have access to the information.

• Reliability based on the outcomes of: (1) integrity—the information has not been modified by non-authorized people—and (2) authenticity—a proof that the information belongs to the correct patient and issued from the right source.

• Availability—warrants an information system to be used in the normal scheduled conditions of access.

On the other hand, the security concept derived from the related standards mentioned in the preceding section can be established through different stages. For example, as outlined by Baur et al. [27], the major stages can be: (1) determination of the appropriate level of security, (2) threat analysis, (3) risk analysis, and (4) establishment of security concept. Determination of the appropriate level of security may include determining security levels of all entities and objects (e.g., IT applications and information sets) linked with the teleradiology system. Threat analysis helps determine the expected threats from the involved objects (e.g., infrastructure, hardware, software, paper-ware). Risk analysis helps quantify the damages for all the identified threats and their occurring frequency. Establishment of security concept deals with either reducing the probability of occurrence of the threats or reducing the damage if an adverse event is unavoidable. This includes selection of suitable measures that reduce the risks to a tolerant level, evaluation of the selected measures, examining the cost–effect relationship as well as analyzing any further risk. All these comprise the security requirements of different domains in teleradiology.

Besides, for computer and network security, various requirements are entitled in different standards such as USA’s Federal Information Processing Standards [4] and Germany’s Bundesdaten-schutzgesetz [27], the general categories of which includes: access control; audit and accountability; certification, accreditation, and security assessment; configuration management; identification, and authentication; media protection; physical and environmental protection; system and communications protection; and system and information integrity. However, as an important aspect of security and risk management in the context of information security [29], we restrict our attention to recognizing the value of information and defining appropriate procedures and protection requirements for the information.

Expected Threats and Conventional Security Measures

Identifying the vulnerability of the system is important to define appropriate procedures or security measures, since the strength of any system is no greater than its weakest link. For example, medical images may pass through various image-information processing systems over the networks, and thereby, the images can be threatened throughout their lifetime in many different ways. A complete protection to those threats means having individual protection mechanisms for each component of the processing system that the images may pass through. With particular attention to the medical information, here we find the suitable measures that provide the required security and privacy services for the information and for the communication services.

Several existing security measures are currently being used such as access control services, firewall, encryption, de-identification services, certification services, etc. Furthermore, the possibilities of new measures such as digital watermarking, digital signature, image hashing, etc. are currently being studied. According to the security requirements discussed in previous section, a review of expected threats and their conventional security measures are summarized in Table 1.

Table 1.

Security requirements of medical information

Security requirement	Threats	Security measures
Confidentiality	Disclosures and re-routing of the information:	Encryption of the data
	During transmission (e.g., when an ill-intentioned person intercepts and illicitly copies files and records)	Limiting lifetime of data
	In the database (resulting in intrusion, identity usurpation, or Trojan horse virus that keeps an open access through the network)	Private communication network (e.g., virtual private network)
		Access control services (against unauthorized person, illegal copy, identity usurpation, etc.) using smart card, firewall, etc.
		User control services for authenticating and identifying the user against identity usurpation, etc.
Reliability: integrity and authentication	Illicit destruction, production, and/modification of the contents of files and records	One-way hash function or robust hash function or digital signature (DS)
		Encryption of the data
		File header, audit logs for recording of data transmission
		Certification of communication partners
		Access control services for writing, reading, and manipulation of data
		User control services for authenticating and identifying the user against identity usurpation
		Software accreditation and use of antivirus and firewall for virus and malicious intrusion
		Non-repudiation services and e-signing
Availability	File management system disablement, destruction of a hard disk, or a malicious pirate who disrupts or alters surreptitiously the organization or content of the data	Access control services for writing, reading, and manipulation of data
		User control services for authenticating and identifying the user against identity usurpation
		Private communication network
		Software accreditation, and use of antivirus and firewall for virus and malicious intrusion

Limitations of the Existing Security Measures

Various existing security measures, as illustrated in Table 1, are being used to protect the medical images and information, and their communications. For example, virtual private network (VPN), firewall, etc., as well as encryption, cryptographic hash function, or their derivatives such as digital signature (DS), machine authentication code (MAC), manipulation detection code (MDC), and perceptual hashing, etc. However, these conventional security measures are considered to have limitations specially in protecting the medical images [6, 7, 13, 18, 30–34], which are summarized in Table 2 and should be properly addressed for the improved security.

Table 2.

Limitations of existing security measures/tools

Measures/tools	Limitations
Firewall and VPN	Only protect the information up to the point of the internal networks [7]
	Provide a certain level of isolation between the intra-net and internet but are easily bypassed by hackers [7]
Encryption	Probably an efficient tool for secure storage and transmission, but once the sensitive data is decrypted, the information is not protected anymore [7, 13, 33]
	Simply using encryption is no guarantee of confidentiality or secrecy [39].
	The randomness of the data for encrypted files stored on media can be used to distinguish the files from other stored data [39].
File-header	Can be easily usurped by a pirate in the plaintext format
	If encrypted, can be very sensitive to bit errors occurring during storage and transmission [7, 32]
Cryptographic hash function and its derivatives (e.g., DS, MAC, MDC, etc.)	Hash function cannot locate where the images have been tampered [31, 32, 47].
	The security of DS largely depends on the strength of the hash functions used to validate the signatures [30].
	It is possible to generate two datasets with different content but having the same message-digest algorithm 5 (MD5) hash [34].
	Cryptographic hash function is extremely bit sensitive to the input [32, 47].
Perceptual hashing	Perceptual hashing usually requires searching for match and access to a central database, where a large amount of pre-computed perceptual hashes are stored [7].
	Most randomization methods in perceptual hashing are linear, which introduces security flaws as known input/hash pairs can be used to recover a secret key [46].
	Their quantization and encoding stages require the learning of appropriate quantization thresholds.
	The quantizer training as well as the storage of thresholds introduces additional security weaknesses.

Firewall and VPN

Among various network security measures, firewalls and VPN are common. Along with intrusion detection systems, antivirus systems, etc., those measures are implemented mainly for protecting the information through securing the communications of a system.

A firewall is usually placed between two networks to act as a gateway, which is a combination of hardware and software that protects the company’s network and computers from possible intrusion by hackers from the external network [35]. Canavan [35] described this as a fundamental component of any perimeter defence that can have the following uses: (1) keeping unwanted and unauthorized traffic from passing (in or out); (2) providing an efficient Internet access to internal users; (3) monitoring for and notifying of intrusions and network problems; (4) maintaining logs of all communication activities between two networks effectively, which can be used to identify abnormal events. Canavan also described three principal requirements of an effective firewall: (1) it must act as a door through which all traffic must pass (incoming and outgoing); (2) it must allow only authorized traffic to pass; and (3) it must be immune to penetration or compromise.

However, a firewall by itself does not assure a secure network, and it represents a single point of failure [35]. Firewall, as only a tool, needs proper configuration and regular monitoring. Firewalls that are not properly configured may allow unauthorized users through. In addition, a denial-of-service attack that effectively shuts down the firewall shuts down the network connection to the outside world [35]. Moreover, a firewall takes time more or less to examine incoming and outgoing traffic, which tends to degrade network performance.

As another significant limitation, firewalls are of no use to track activities on the internal network. While a firewall does make it somewhat more difficult for someone from the outside to get in, the majority of attacks on corporate systems come from the inside, not from the outside [35]. In addition to the threat from inside of an organization, firewalls can be circumvented by outsiders [35]. As a result, critical systems should be configured to monitor logins, failed logins, and all network activity of the internal systems.

A VPN, on the other hand, is a means of transporting traffic in a secure manner over an unsecured network which is achieved by employing some combination of encryption, authentication, and tunnelling [36]. "Tunnelling" refers to the process of encapsulating or embedding one network protocol to be carried within the packets of a second network. There are several different implementations of VPN protocols such as point-to-point tunnelling protocol (PPTP), Internet protocol security (IPSec), secure sockets layer (SSL), secure shell (SSH), etc. Those protocols have different pros and cons from different technical perspective [36]. For example, SSL supports transmission control protocol traffic only; SSL and SSH depend on client port forwarding; some protocols use symmetric or weak encryption (e.g., PPTP), and IPSec supports unicast traffic only, etc. However, considering the general perspective of information security, further to firewalls, a VPN can be used to protect the information up to the point of the communication networks.

Encryption

In order to protect the privacy and confidentiality of electronic health information, encryption has been a commonly accepted technology in health care sector [37]. In cryptography, encryption is the process of transforming information (referred to as plaintext) using an algorithm (called cipher) to make it unreadable to anyone except those possessing special knowledge, usually referred to as a key [38]. The result of the process is encrypted information (called ciphertext). There are two types of encryption: symmetric (private/secret key) encryption (e.g., data encryption standard, Rivest Cipher #4- RC4) and asymmetric (public key) encryption (e.g., Diffie Hellman, digital signature algorithm (DSA)).

The strength of the symmetric scheme is largely dependent on the size of the key and on keeping it secret. Generally, the larger the key, the more secure the scheme [39]. Furthermore, symmetric encryption is relatively fast and widely understood. However, the main weakness of this type of encryption is that the key or algorithm has to be shared [39]. In addition, symmetric key provides no process for authentication or nonrepudiation [39]. Here, nonrepudiation is the ability to prevent individuals or entities from denying that a message was sent or received or that file was accessed or altered, when in fact it was. That is why symmetric cryptosystems are not well suited for spontaneous communication over open and unsecured networks [39].

On the other hand, asymmetric encryption uses two keys as opposed to one key in a symmetric system [39]. One of them is kept secret and called private key, while the other is made public and called public key. A message is encrypted with the private key and decrypted with the public key. The advantages of this type of encryption include no secret sharing and providing a means of authentication and nonrepudiation with the help of digital certificates. Unlike symmetric cryptosystem, public key allows for secure spontaneous communication over an open network. Besides, it is more scalable for very large systems than symmetric cryptosystems. Yet, asymmetric encryption is relatively slower and computationally intensive, and requires certificate authority [39].

File Header

It is a common practice of appending metadata containing owner ID, size, last modified time, and location of all data blocks, etc., as a header with the data block. The size of this header varies depending on how much header information is to be stored. The DICOM standard allows image information object definitions that a DICOM file not only contains pixel data but also key information about the image [40]. Thus, a single DICOM file contains both a header and all of the image data.

Conventionally, each DICOM medical image is associated with a patient’s private data such as patient’s name, age, results of examination/diagnosis, time taken, etc. All these private information are recorded into a meta-data or header file, which is appended to the image. The DICOM standard stores the image data and the meta-data separately. Clearly, this is dangerous as the link between the image and the textual information is practically non-existent [41]. For example, for the images with plain-text file-header, the major threat is the violation of the access rights and of the daily logs by the intruder. Hence, breaking of the confidentiality implies that integrity and authenticity of the data cannot be guaranteed anymore [7]. Furthermore, for an encrypted header, the bit error sensitivity may result in loss of header and/raise further complexity in managing the medical images. Thus, at the least, the patients’ private data in a DICOM image are at risk of happenings of a mismatch (i.e., linking of meta-data with an incorrect medical image) and of disclosure and loss of header or meta-data in an image undergoing some intentional processing.

Cryptographic Hash Function

A cryptographic hash function is a deterministic procedure that takes an arbitrary block of data and returns a fixed-size bit string, the (cryptographic) hash value, such that an accidental or intentional change to the data will change the hash value [32]. The data to be encoded is often called the message, and the hashes are sometimes called the message digest or simply digest [42]. The ideal cryptographic hash function has four main properties [43]: (1) it is easy (but not necessarily quick) to compute the hash value for any given message; (2) it is infeasible to generate a message that has a given hash; (3) it is infeasible to modify a message without changing the hash; and (4) it is infeasible to find two different messages with the same hash.

Cryptographic hash functions have many information security applications, notably in DS, MACs, MDCs, and other forms of authentication [42]. They can also be used for other purposes such that indexing data in hash tables, fingerprinting, detecting duplicate data and accidental data corruption, etc [42]. Indeed, in information security contexts, cryptographic hash values are sometimes called (digital) fingerprints, checksums, or just hash values, even though all these terms stand for functions with rather different properties and purposes [42]. In addition, most of the existing cryptographic hash function schemes unfortunately remain vulnerable to incidental modifications (i.e., even a one bit change in the input will change the output hashes dramatically) [32]. This severely limits their practical utility in robust content authentication for multimedia applications.

Perceptual Hash Function

Perceptual hash functions (or, robust perceptual hash function, or simply, perceptual hashing) are designated hash functions for multimedia contents. This type of hash function takes a large digital image as input, and with constructing a content descriptor of the input, outputs a fixed length binary vector known as hash value. This hash value is required to be invariant under changes to the image that are perceptually insignificant whereas, on perceptually distinct inputs, the hash values need to be approximately independent and hence different with high probability [44]. A good perceptual hash function should have the following properties [45]: (1) Robust: Manipulations that do not change the perceptual information should not change the hash value; (2) Unique: Perceptually different inputs should have completely different hash values; and (3) Secure: It should be very hard to find (forge) perceptually different inputs having similar hash values.

Similar to cryptographic hash functions, perceptual hashing is required to generate different hash values for different inputs. However, here, the definition of difference is changed from bitwise difference to perceptual difference [45]. That is, unlike getting a very different hash value from a single bit change in the input of the cryptographic hash function, perceptual hashes are expected to be different only with the changes in the perceptual content of the input. For instance, the hash value of an image and its JPEG compressed version should be the same for the perceptual hash function, since they have no perceptual difference, although their bit-string representation is completely different [45].

Generally, perceptual hashing consists of feature extraction and randomization that introduces non-invertibility and compression followed by quantization and binary encoding to produce a binary hash output. Most randomization methods are linear, and this introduces security flaws because known input/hash pairs can be used to recover a secret key [46]. Furthermore, the quantization and encoding stages require the learning of appropriate quantization thresholds, and the quantizer training as well as the storage of thresholds that introduce additional security weaknesses.

Moreover, content-based feature extraction methods, developed from a signal processing perspective, are known to be robust but not secure [44]. Kalker [31] described perceptual (or robust) hashing from the perspective of a neural archiving activities using clever signal processing and database techniques. The former is responsible for extracting essential perceptual features (also referred to as perceptual hash values or hash values for short), the latter for storing and searching large amounts of pre-computed hash values. Kalker also exemplified with a typical scenario, where a local client (e.g., a mobile phone) is responsible for capturing the content and transmitting the content (possibly only the hash values if the client is equipped with a feature extractor) to a central database. The central database matches the hash values of the unidentified content with the pre-computed hash values, retrieves the best match, and takes appropriate action (e.g., sending an artist name and song title in an SMS message to the requesting client).

Reviewing some key existing security measures as discussed above, it can be said that they are useful in handling the common security problems of the system. Yet, their limitations suggest that they are no longer sufficient to provide the required security of the medical information in teleradiology. Therefore, as the new security problems are arising from the advances of technology and developments of PACS/RIS mentioned in “Introduction,” new measures are required to be developed and deployed for the improved security of medical images and EPHI. Hence, studies show that digital watermarking can be promising to facilitate sharing and remote handling of that information in teleradiology in a secure manner [16–18], though a reasonable justification of watermarking applicability for medical images is lacking.

Digital Watermarking in Teleradiology

Watermarking nowadays, while well established in a range of applications [48], is only just beginning to be explored for healthcare and medical information systems [49, 50]. Digital watermarking, basically, is a process that principally permits the adding of information as a ‘watermark’ into the object, a digital media (e.g., digital image, audio, etc.) such that the watermark can be detected afterward. Generally, digital watermarking consists of three major components: watermark generator, embedder, and detector [51] as shown in Fig. 2. A watermark generator generates desired watermark(s) for a particular application, which are optionally dependent on some keys. Watermark(s) are embedded into the object by a watermark embedder, sometimes based on an embedding key whereas a watermark detector is responsible for detecting the existence of some predefined watermark in the object. It is sometimes desirable to extract a message as well.

Fig. 2.

Fundamental components of digital watermarking: a watermark generation, b watermark embedding, and c watermark detection

In a target application, digital watermarking objectives can deal with mainly two issues. One is to address security (e.g., authentication and integrity control of the cover object, confidentiality of the information used in watermark, etc.), and the other is to address system considerations (saving memory and bandwidth, avoiding detachment, etc., e.g., annotation of useful information such as electronic patient records (EPR), electronic transaction records (ETR), etc.). Furthermore, based on the processing domain, watermarking schemes can be broadly categorized as (1) spatial domain watermarking and (2) transform domain watermarking. Spatial domain schemes include LSB embedding, spread spectrum technique, etc., and transform domain schemes are based on discrete cosine transform, discrete Fourier transform, and discrete wavelet transforms. Watermarking in spatial and transform domains have different advantages and disadvantages [14, 52], which are illustrated in Table 3 below.

Table 3.

Advantages and disadvantages of watermarking in spatial and transform domains

Types of processing	Advantages	Disadvantages
Spatial domain	Comparatively simple and faster operation	Vulnerable to compression, geometric distortion, and filtering
Transform domain	Compression compatible and robust against many geometric distortions (e.g., rotation, scaling, translation, cropping) and filtering	Comparatively higher computational time and complexity

Advantages of Digital Watermarking

Watermarking has received much attention recently for medical image applications because of its various attractive attributes [7, 53, 54], which are listed below:

Security and Privacy

The fundamental and most attractive property of watermarking is data-hiding capability [55, 56]. The utmost confidentiality can be maintained by hiding the private data into the images. Keeping necessary medical information (e.g., EPR including demographic data, diagnostic results, treatment procedures, etc.) hidden in medical images may provide a better security against malicious tampering, assuming medical images would not be of people’s interest without the patient information [15, 57]. Even that which is tampered intentionally or in an unintended manner can be detected and possibly recovered by using an appropriate watermarking scheme [58, 59]. Hence, Coatrieux et al. [7] outlined three main objectives of watermarking in the medical image applications: data hiding, integrity control, and authenticity, which can provide the required security of medical images. For example, data-hiding objective of watermarking allows inserting meta-data and other information so that the image is more useful or easier to use. Integrity control objective of watermarking ascertains that the image has not been modified in an unauthorized manner. Digital watermarking allows permanent association of image content with proofs of its reliability by modifying [some] image pixel values, independently of the image file format [13]. It can also operate in a stand-alone environment and has a versatile message set. In addition, authenticity traces the origin of an image.

Avoiding Detachment

The data-hiding property of watermarking mentioned above further facilitates annotation of necessary information to avoid detachment. Millions of medical images are being produced in radiology departments around the world, which have immense value to practicing medical professionals, medical researchers, and students [53]. Researches in this field are being accomplished to embed patient data to medical images [55, 60, 61]. If the EPR and the images are separate, the chance of detachment of patient data from the image becomes higher. Misplacing a data will be very crucial in the case of medical image. In order to avoid this misplacing or detachment, watermarking offers necessary data embedding within the image itself.

Indexing

Another benefit stems from data-hiding capability of watermarking is indexing, where relevant keywords or indices can be embedded into the images and used for effective archiving and retrieval of the images from databases [53].

Nonrepudiation

In teleradiology, distribution of the watermarked images between HISs may cause nonrepudiation problem, where both the involved parties (e.g., hospital personnel and clinician) may repudiate that they did not send the data. Along with other advantages, watermarking is also promising to support nonrepudiation in various multimedia applications [62, 63]. Hence, use of a key-based watermarking system may facilitate nonrepudiation in teleradiology such that both parties could be in safer side where the key used by the hospital could be their logos or digital signatures.

Controlling Access

Provision for using keys in watermarking schemes further provides an alternative to access control mechanism, where confidential meta-data can be accessed with the proper authoritative rights given in terms of keys [53, 64].

Memory and Bandwidth Saving

Storage space and bandwidth requirements are important decisive factor for small hospitals’ financial economy. The memory for storage can be saved to a certain extent in HIS by embedding the EPR in the image [61, 65]. On the other hand, a huge amount of bandwidth is required for the transmission of the image data in teleradiology. The additional requirement of bandwidth for the transmission of the metadata can be avoided if the data is hidden in the image itself. Since the EPR and the image can be integrated into one, bandwidth for the transmission can be reduced in telemedicine applications [53].

Choice of Design and Evaluation Parameters

Watermarking requirements for medical images are mainly defined in terms of security and privacy, fidelity, and computational properties. Hence, security and privacy requirements characterize a watermarking scheme to achieve data hiding, integrity control, and authenticity objectives as discussed in previous section. Fidelity requirements guarantee that the watermarked medical images are useable for diagnosis and other clinical uses. Besides, the computational properties help obtain the cost benefit and feasibility analysis for practical implementation. All these watermarking requirements, on the other hand, define various watermarking design and evaluation parameters in an application scenario. Design parameters help characterize the development of a watermarking scheme, whereas the evaluation parameters help determine the performance of a developing/existing scheme. Typical parameters for watermark generation and embedding include visibility, blindness, embedding capacity, imperceptibility/perceptual similarity, etc. Similarly, blindness, invertibility, robustness, error probability, etc., are the parameters for the detection [66].

Moreover, deploying a watermarking system in medical image applications broadly includes two phases, namely a development phase and a validation phase as illustrated in Fig. 3. In the development phase, optimum criteria for the necessary design parameters of the system are to be defined properly according to the medical image requirements, since all the design parameters of watermarking frequently influence one another (directly or indirectly) [67]. Similarly, it is also necessary to have a careful consideration on the evaluation parameters, their suitable measures, and the requirements of the medical images in the validation phase, in order to justify the suitability of existing/developing watermarking schemes for medical image application.

Fig. 3.

Interlinking of digital watermarking with medical image applications

The system design and evaluation parameters for image watermarking are mainly associated with its core components: watermark generation, embedding, and detection [66]. Various design and evaluation parameters play an important role in achieving a particular objective in an application scenario, which has been reviewed for teleradiology application and is discussed in the following.

Visibility

Visible watermarking are important in recognition and support of possessing a digital image, where watermarking objectives is mainly to show some necessary information such as logo, icon, courtesy, etc., through the watermarked image. Contrariwise, invisible watermarks are used in digital image applications, where watermarking objectives are to addressing security issues of the images. Like various digital image applications [68–76], invisibility of the watermark appears to be the main interest in the research of medical imaging [8–10, 12, 13, 41, 77–79].

Robustness

Robustness is an important parameter for the watermarking detector defined in different ways [80]. Robustness, basically, is defined as the degree of resistance of a watermarking scheme to modifications of the host signal due to either common signal processing, or operations devised specifically in order to render the watermark undetectable [81]. This parameter categorizes watermarking schemes to be robust, fragile, and semi-fragile. In a robust watermarking, a watermark usually carries information regarding the owner in order to validate who the image belongs to (e.g., which person, which institute or organization, etc.). Thus, these watermarking schemes are being used for content authentication purposes in various digital image applications (e.g., copyright protection) [68, 73, 79, 82–86]. Semi-fragile and fragile watermarks are being used to carry much information about itself, its owner’s metadata, its distribution, etc., and are thus used for annotation (e.g., hiding ETR or EPR, etc.) [15, 55, 57, 61, 65, 87] and integrity control (e.g., tamper detection and recovery) [8, 59, 88–90].

Blindness

Blindness in watermarking refers to the ability of a component function (e.g., watermark generation, detection) to work without any original version of input (e.g., image or watermark, etc.). Non-blindness in watermark generation is important while an original image dependent watermark is required. An original image-dependent watermark is helpful in addressing ambiguity and forgery attacks (e.g., copy attacks) [67]. Here, if the watermark is not dependent on the original image, it can be easily copied to another image or forged to output an invalid watermarked image [91]. Besides, blindness in detection is important, where availability of the original image or watermark at the detector can thwart watermarking objectives. Non-blindness in detection is used sometimes in developing tamper-recovery watermarking schemes, where the recovery of tampered regions is often difficult to achieve from the watermarked image itself.

Embedding Capacity

Embedding capacity is generally measured by number of embedding bits. High embedding capacity is a key issue in developing annotation or integrity control watermarking schemes [92], which are generally of fragile or semi-fragile nature to some common image processing. Achieving high embedding capacity often introduces more distortions to a watermarked image and thereby often makes it difficult to preserve high imperceptibility. A robust watermarking used for content authentication purpose requires comparatively lower embedding capacity than that required for annotations purposes of a fragile/semi-fragile watermark [93, 94]. Research shows that LSB embedding techniques offer comparatively higher embedding capacity [13, 95].

Invertibility

Invertibility of a watermarking system indicates the detection function to be the inverse of the embedding function. Invertible (or sometimes referred to as reversible or lossless) watermarking is of special interest in digital image applications where no distortions are allowed in the original image. Therefore, an original image is required to be restored from respective watermarked images by the detector. Invertibility seldom gets interest for non-blind detector since detection itself requires the original image, although developing a blind detector for invertible watermarking is more challenging, especially when a high embedding capacity is desired. Developing this type of watermarking received much attention in medical image applications to avoid any misdiagnosis from distortions in a watermarked image [8, 72, 78].

Perceptual Similarity

Perceptual similarity determines the degree of imperceptibility between the original image and its watermarked version, especially in developing an invisible watermarking scheme [67]. Different similarity metrics are used for this parameter such as correlation quality; signal-to-noise ratio (SNR), peak SNR (PSNR), weighted PSNR, mean square error; structural similarity (SSIM), mean SSIM; and normalized cross-correlation (NCC). In medical image watermarking applications, perceptual similarity must be very high to avoid any risk of misdiagnosis.

Security

Security requirements of watermarking include the legitimate access, use, disclosure, disruption, modification, or perusal of the watermarking system. Determining these security requirements in a target application is crucial for the system design, and that can be determined through comprehensive risk management (e.g., examining security policy, access control, physical and environmental security, operation managements, etc.) [96, 97].

Error Probability

Error probability is another important parameter for assessing detection performance of a watermarking scheme. Irrespective of application scenarios, zero error probability is always desirable, although achieving this is practically difficult considering higher degrees of robustness to any distortions [17]. However, like in other digital image applications, keeping the error probabilities lower as much as possible is very important in a medical image application scenario in order to ensure reliable detection. Some of the important and commonly used error probability metrics are bit error rate, false-positive rate, false-negative rate, etc.

Digital Watermarking Versus Other Security Measures/Tools

Digital watermarking has some unique advantages for teleradiology, although few existing security measures/tools may serve its other objectives together, for example, encryption, cryptographic hash function (e.g., MAC, DS, etc.), perceptual hashing, etc. Following our previous discussion on watermarking and other comparable security measures/tools, an extensive comparison among them based on various key properties and requirements of medical image applications is made and presented in Table 4.

Table 4.

Watermarking versus other security measures/tools

Properties and requirements	Digital watermarking	Hash function		Encryption
		Perceptual	Cryptographic
Objective	Data and copyright protection	Data protection	Data protection	Secure communication
Host-signal/cover-object	Mostly image/audio data	Mostly image data	Plaintext messagea	Plaintext messagea
Secret data	Watermark	–	–	Plaintext
Key	Optional	Optional	Optional	Necessary
Input	Generally the watermark and the cover-object/host-signal	Arbitrary block of host-signal	Arbitrary block of host-signal	Arbitrary block of host-signal
Output	Watermarked data	Hash-values/message-digest	Hash-values/message-digest	Ciphertext
Detection type	Blind, semi-blind, non-blind	Non-blind	Non-blind	Blind
Failure	If an invalid watermarked image is detected as valid, or vice versa (e.g., from unauthorized removal or embedding of watermark)	If the message is generated from the hashes, or if another message or perceptual changes in the original gives the same hashes.	If the message is generated from the hashes, or if another message or bit changes in the original gives the same hashes.	If a ciphertext is illicitly de-ciphered
Impact on quality/content of the image	Yes, but can be acceptably reduced/resolved by considering non-region of interest (RONI) or reversible watermarking [14, 98, 99]	No	No	No
Sensitivity to bit error	Low	Low	High	High
Robustness	Can be designed as robust, semi-fragile, fragile	Robust only	Robust only	Robust only
Authentication/integrity check	Yes	Yes	Yes	Yes, but as long as data are encrypted
Tamper localization	Yes (also can suggest for recovery to a certain extent [14])	No	No	No
Annotating metadata (e.g., EPR, ETR, etc.)	Yes, but to a limited capacity	No	No	No
Confidentiality of metadata	Yes (also, for higher confidentiality, encrypted information can be used in generating watermark [57, 100, 101])	No	No	Yes
Database requirement	No, it can operate in stand-alone environment [8, 31]	Yes, for storing pre-computed perceptual hashes [46]	No	No
File-format independent	Yes	–	–	–

^aImage and audio data can be used, if they are represented as plaintext message

As Table 4 illustrates, cryptographic/perceptual hashing has no impact on quality of the host-signal, and is suitable for legacy content, but they are either bit-sensitive (for cryptographic hash functions) or need access to a central database to search for a match with a pre-computed hash (for perceptual hashing), whereas research suggests that a carefully designed watermarking scheme does not alter medical diagnosis [102]. Although watermarking has an impact, more or less, on perceptual quality and difficulties with legacy content, Guo and Zhuang [103] suggested three ways to overcome the distortion induced in images by watermark embedding. They are: (1) defining acceptable range of distortion for watermarking; (2) separating an image into protection zone and insertion zone such as ROI (region of interest) and RONI (region of non-interest); and (3) considering watermarking as an invertible manner to recover the original image at the watermark decoder site. Hence, ROI indicates the region significant for diagnosis and other clinical uses, and RONI indicates the complementary region of ROI, which has lesser or almost no significance in diagnosis.

Defining acceptable range of distortion for watermark embedding through clinical validation is expensive, which is applied by Zain et al. [102]. In contrast, separation of ROI and RONI in medical images is not straightforward and may require the interaction/approval of doctor/radiologist. In addition, making such separation is sometimes very difficult, although it is applied in several watermarking schemes [9, 12, 77]. Besides, developing reversible watermarking is promising for medical image application with taking no risk for sacrificing the diagnostic accuracy, although computational properties may incur additional complexity in different processing domains. Additionally, Coatrieux et al. [104] discussed two limitations of reversible watermarking: (1) It imposes the watermark removal before the diagnosis, and (2) it assumes a secured environment because, once the watermark is removed, the image is not protected anymore like in cryptography. All these suggest that a combination of suitable type of watermarking schemes, where the concept of multiple watermarking stems from, can be developed in order to address the rising security problems of medical images in teleradiology [77, 105–107]. Studies also show that incorporation of asymmetric encryption and lossless compression can help attain additional confidentiality, non-repudiation property, high embedding capacity [15, 103, 108].

Watermarking allows using DS or perceptual hashing for appropriate applications [78, 105, 109, 110]. Watermarking systems have room for employing encryption for the additional confidentiality of metadata (e.g., in generating watermark). Memon et al. [13] proposed a digital watermarking scheme, in which watermark is comprised of patient information, hospital logo, and message authentication code, computed using hash function. To ensure inaccessibility of embedded data to the adversaries, BCH encryption of watermark is performed there. For the same purpose, Li-Qun et al. introduced DSA [111] and digital signature technology based on RSA public cryptosystem [110], integrating reversible digital watermarking with digital signature to form an authentication system.

Furthermore, a few of recent studies show the use of a compression technique for attaining the embedding capacity requirements of watermarking. Nambakhsh et al. [112] presented a watermarking method on several computed tomography (CT) and magnetic resonance (MRI) images, where the original image is compressed using the zero-tree wavelet (EZW) algorithm. Raul et al. [101] used Huffman compression and RC4 method that respectively compress and encrypt the metadata in a blind watermarking scheme. Kundu et al. [15] presented a watermarking scheme that combines lossless data compression and advanced encryption standard for encryption of medical images. In addition, Sung-Jin et al. [52] proposed an algorithm that utilizes both JPEG 2000 and robust watermarking for protection and compression of the medical image. Thus, depending on the application, a choice for the appropriate mixture of various technologies can be made to devise a suitable watermarking system for teleradiology.

Objectives and Applications of Watermarking for Medical Images

Popularity of Internet has become a boon to patients and low-capital hospitals to utilize the facility to communicate with the clinicians for clinical diagnosis purposes [54], where the security of medical images can presumably be addressed to a considerable extent by inserting a properly selected additional data into medical images through digital watermarking. A digital medical image application is therefore one of the prospective target areas of using digital watermarking. Studies show that various watermarking schemes can be used in teleradiology for (1) origin/content authentication [9, 13, 41, 107, 113–119], (2) EPR annotation [57, 65, 88, 120–122], and (3) tamper detection and recovery of medical images [8, 14, 59, 123, 124]. Some important aspects of medical image watermarking schemes for their different objectives are summarized below.

Origin/Content Authentication

Watermarking has received much interest in the research for origin authentication of the medical images. The important details can be stored in images imperceptibly, causing no harm to the ROI of the images. This kind of brief descriptions can be hidden in images immediately after the production of the images in the radiology departments. This can be done by incorporating the watermarking in the different modality machines namely, CT or MRI scanners. The database systems use the mechanisms of granting and revoking privileges and of authorization control to ensure the security of data with the permanent association of the watermark. Our observation suggests the following requirements for this type of watermarking in teleradiology: (1) The watermark should be invisible, blind, and robust; (2) watermark should incorporate the minimum information required for the origin authentication; (3) embedding process must consider the RONI; and (4) proper validation of a watermarking scheme such that the permanent association of the watermark is reliable and safe for diagnosis.

Regarding the validation of a watermarking scheme, although it is required for any scheme to be applied in any application scenario, extra care needs to be taken when the effect of watermark embedding is not recoverable. Hence, the permanent association of such robust watermarking requires compromising few bits, which further warrants determining the acceptable range of distortion. Moreover, this type of watermarking should incorporate the RONI embedding for the reliable clinical uses of medical images, particularly, when used along with a reversible watermarking (to form multiple/sequential/hierarchical watermarking scheme) that assumes a secure environment as mentioned in previous section.

EPR Annotation

EPR and other useful medical information annotation are other key objectives for medical image watermarking. Navas et al. [125] suggested three key requirements for EPR data hiding and transmission: (1) The recovery of the EPR should be blind due to the unavailability of the original image; (2) zero bit-error rate (BER) is essential for EPR data; and (3) imperceptibility should not be compromised for any reason. These requirements suggest necessary criteria of a watermarking scheme for medical images to be invisible, blind, and reversible. Such a watermarking scheme can be either robust or semi-fragile. For higher capacity, the watermarking scheme can be semi-fragile, although it requires defining appropriately the set of necessary operations/processing, to which the scheme needs to be robust or not to be. A bit-error correction technique can be used for attaining zero BER and improving watermarking performance [126, 127]. For additional confidentiality, encryption of the EPR can also be used in watermark generation [127, 128].

Tamper Detection and Recovery (Integrity Control)

Medical images in different radiological modalities such as X-rays, ultrasounds, and MRI contain vital medical information that can be tampered with easily available image processing tools. Thus, their protection and authentication seems of great importance, and this need will rise along with the future standardization of exchange of data between hospitals, or between patients and doctors [118]. Integrity of a medical image can be achieved in three levels [129]: (1) tamper detection, (2) tamper localization, and (3) possible recovery by approximating the tampered region. In order to achieve this along with the requirements of medical image needs a watermark to be (1) fragile and blind and (2) reversible or RONI-embedding-based. Hence, fragile watermarking help locate the tampered region with its fundamental property that a watermark becomes invalid for any malicious or unintentional modifications in the watermarked image.

If the origin authentication of a medical image is achieved by the robust watermarking, fragile reversible watermarking (in the form of multiple watermarking) can further locate and possibly recover any tampered region of the watermarked image. This will allow the system to control the integrity as well as authentication. In that case, if the watermarking is RONI-based instead of reversible, then the limit of additional distortion must be taken care of. Furthermore, as in EPR annotative watermarking, LSB-embedding-based watermarking schemes for tamper detection and recovery received much interest in the research, since consideration on the embedding capacity is equally important for the both watermarking objectives.

Discussion and Conclusions

Study of security and privacy problems is a continuous process and is mainly influenced by the technological advances in the field. It has been more than a decade since the study of digital watermarking (finding relevance and suitability, and developing of new schemes and their evaluation) has found its way to medical image applications. However, watermarking for medical images is not practically well accepted yet. This reluctance is instigated from the incomplete justification of watermarking applicability for the strict requirements of medical images. In this paper, an extensive investigation is conducted and described in three parts, namely: (1) the security and privacy requirements; (2) conventional security measures and their limitations; and (3) justification of using watermarking for medical images, in teleradiology.

The need for sharing of medical images and information is growing rapidly for improved healthcare access, delivery, and standards. Web services technology has recently been widely proposed and gradually adopted as a platform for supporting systems’ integration [130]. The DICOM standard as well as ISO27799 and other government regulations such as HIPAA, CFR 45, Directive 95/46/EC, etc., impose rules as national/international standards to protect individuals’ health information, highlighting security and privacy protection requirements. Our study suggested three mandatory characteristics: confidentiality, reliability, and availability that need to be achieved for medical images in teleradiology.

However, a complete solution for various security problems discussed so far is still lacking. Although conventional security measures have their limitations, they cannot be replaced with any individual measure. For example, authentication based on watermarking cannot replace classical cryptographic authentication protocols that protect communication channels [131]. However, well-known cryptographic algorithms can be used to guarantee the privacy, authenticity, and integrity of messages embedded in multimedia content, where there is no cryptographic solution for the threat of unauthorized watermark removal [132]. To this, other conventional security measures may still be required, while watermarking complements the security of multimedia data. Especially, watermarking provides a great prospect for teleradiology because it functions as a communication tool with the authenticity of the origin/sender, nonrepudiation, detection of data tampering, memory and bandwidth saving, integrity of the image, and so on.

We observed that the general requirement for any medical image watermarking implies that watermarking needs to be invisible and blind, whereas, robustness, reversibility, and RONI embedding as well as other design parameters must be taken into consideration according to the objectives dictated by the application scenario. Although it is not identified as a general requirement, a prior clinical validation of a watermarking scheme may always be subjected to its medical image application irrespective of the type and properties of watermark(s). As a result, any permanent or temporary modification due to watermark embedding may remain reliable and safe for diagnosis.

In teleradiology, the primary objective of a watermarking scheme for medical images should be authentication (e.g., origin or content). EPR annotation and integrity control (or, tamper detection and recovery) can be a further goal(s) to form a multiple watermarking scheme. Thereby, a properly designed multiple watermarking scheme may have the potential intelligence to address the rising problem in teleradiology. Although the concept of multiple watermarking scheme is not new, their applicability in teleradiology naturally requires more explicit consideration on the performance evaluations and security analysis, including overall computational complexity, speed, and cost–benefit analysis.

Finally, without considering and characterizing, the required design and evaluation parameters systematically may pose serious flaws and may render a watermarking scheme ultimately useless for the application. Therefore, this study recommends a systematic development of multiple watermarking schemes and their complete assessment through defining the parameters properly such that they can offer a better complementary solution for achieving improved security in teleradiology. Hence, a suitable generic watermarking model and a point of reference for benchmarking is recommended as another milestone to be addressed in future research.