A Cloud Architecture for Teleradiology-as-a-Service...services are used every day as an ordinary commodity. Cloud computing is a recent paradigm that consists of delivering com- puting

© Schattauer 2016 Methods Inf Med 3/2016

203Original Articles

1. IntroductionMedical imaging laboratories are complex environments using very specialized equip-ment, from image acquisition to visualiza-

tion. The IT infrastructure, namely storage units and communication layers, needs to be prepared to deal with a large amount of produced data that can reach terabytes per year in a central hospital [1–5]. Three

decades ago, the first Picture Archiving and Communication System (PACS) [6] emerged in healthcare institutions, aiming to store and manage the information pro-duced inside a radiology department. PACS and the DICOM standard (Digital Imaging and Communications in Medi-cine) [7] are now mandatory in medical imaging, by defining a normalized way of acquiring, storing and transmitting medi-cal images.

A PACS also gives healthcare practi-tioners the ability to remotely access multi-media patient information and to set up telemedicine, telework and collaborative work environments [8], even in mobile platforms [9]. Medical image repositories are often accessible only through the DICOM query and retrieve service [10].

Computing devices and Internet access are now available anywhere and anytime, creating new opportunities to share and use online resources. A tremendous amount of computational power (e.g. Google Compute Engine and Amazon Elastic Compute Cloud) and an unprece-dented number of Internet resources and services are used every day as an ordinary commodity. Cloud computing is a recent paradigm that consists of delivering com-puting as a service rather than a physical product (e.g. physical servers) [11]. Ser-vices are provided according to a pay-per-use business model and with important features such as security, reliability, instant scalability and global connectivity. With the rise of cloud computing, many services are being outsourced – data are stored in the providers’ infrastructure and the infor-mation accessed anywhere through a Web browser, even on mobile clients. Without a doubt, these features are also an opportuni -ty for telemedicine. Furthermore, the social media paradigm [12], which consists of on-line social interaction where individual users become part of an online community,

A Cloud Architecture for Teleradiology-as-a-Service*Eriksson J. Melício Monteiro; Carlos Costa; José L. OliveiraUniversity of Aveiro, DETI / IEETA, Aveiro, Portugal

KeywordsTelemedicine, teleradiology, cloud comput-ing, medical imaging, PACS, DICOM, XMPP

SummaryBackground: Telemedicine has been pro-moted by healthcare professionals as an effi-cient way to obtain remote assistance from specialised centres, to get a second opinion about complex diagnosis or even to share knowledge among practitioners. The current economic restrictions in many countries are increasing the demand for these solutions even more, in order to optimize processes and reduce costs. However, despite some technological solutions already in place, their adoption has been hindered by the lack of usability, especially in the set-up process.Objectives: In this article we propose a tele-medicine platform that relies on a cloud com-puting infrastructure and social media prin-ciples to simplify the creation of dynamic user-based groups, opening up opportunities for the establishment of teleradiology trust domains.Methods: The collaborative platform is pro-vided as a Software-as-a-Service solution, supporting real time and asynchronous col-laboration between users. To evaluate the

solution, we have deployed the platform in a private cloud infrastructure. The system is made up of three main components – the collaborative framework, the Medical Man-agement Information System (MMIS) and the HTML5 (Hyper Text Markup Language) Web client application – connected by a message-oriented middleware.Results: The solution allows physicians to create easily dynamic network groups for synchronous or asynchronous cooperation. The network created improves dataflow be-tween colleagues and also knowledge shar-ing and cooperation through social media tools. The platform was implemented and it has already been used in two distinct scena -rios: teaching of radiology and tele-reporting.Conclusions: Collaborative systems can sim-plify the establishment of telemedicine ex-pert groups with tools that enable physicians to improve their clinical practice. Stream -lining the usage of this kind of systems through the adoption of Web technologies that are common in social media will in-crease the quality of current solutions, facili-tating the sharing of clinical information, medical imaging studies and patient diag-nostics among collaborators.

Correspondence to:Eriksson J. Melício Monteiro University of Aveiro, DETI / IEETA 3810-193 Aveiro Portugal E-mail: [email protected]

Methods Inf Med 2016; 55: 203–214http://dx.doi.org/10.3414/ME14-01-0052received: June 2, 2014accepted: January 19, 2016epub ahead of print: March 4, 2016

* Supplementary material published on our web-site http://dx.doi.org/10.3414/ME14-01-0052

For personal or educational use only. No other uses without permission. All rights reserved.Downloaded from www.methods-online.com on 2017-12-30 | IP: 54.70.40.11

Methods Inf Med 3/2016 © Schattauer 2016

204 E. J. Melício Monteiro et al.: A Cloud Architecture for Teleradiology-as-a-Service

generating content and interacting with other users via messaging channels and en-gaging forums, has become increasingly popular among Web users. As a result, it is also being explored by health communi-cation programs aiming to identify new interaction models between patients and physicians [13].

This paper presents a new concept of Teleradiology-as-a-Service based on a cloud service that provides secure and pri-vate spaces where physicians can store exams, establish collaborative groups and share data without complex configurations or set-up problems. The platform facilitates real time and asynchronous collaboration, making use of the social media paradigm. For example, a radiologist can import a study to his personal archive, anonymize the images, append annotations and start a case discussion with a colleague, powered by a set of collaborative tools such as syn-chronized image visualization and video-conferencing.

2. Background2.1 Medical Imaging

Healthcare centres were quick to recognize the benefits of IT in the management of medical information. The first PACS revol-utionized radiology and to some extent medical practice [14]. Due to the large amount of data generated, it was necessary to create major infrastructure and work-flows defining clearly how medical images are stored and accessed in a hospital net-work. PACS systems use the DICOM stand ard [7, 14], which defines data formats, storage organization and com-munication protocols of digital medical imaging.

Typically, a PACS is composed of four components: a) gateway for medical image acquisition, b) PACS archive server, c) workstations for data visualization and d) application servers (▶ Figure 1). In sum-mary, it can be said that a PACS en-compasses technologies used for the ac-quisition, archiving, distribution and visu-alization of a set of digital images using a computer network for diagnosis and re-view at dedicated stations.

2.2 Cloud Computing

The cloud computing concept allows enter-prises to deploy services based on the pro-viders’ infrastructure without having to manage and maintain it. Cloud providers offer a diversified set of services, such as elastic computing, storage, databases and notification systems, based on three dis-tinct kinds of service [15]: a) Software-as-a-Service (SaaS), b) Platform-as-a- Service (PaaS) and c) Infrastructure-as-a-Service (IaaS).

Many services on the Internet, such as Google applications and social network ap-plications (Facebook, Twitter, etc.) already use the SaaS model, usually providing Web-based interfaces, Web services [16, 17], and recent Web technologies to allow richer interaction with their clients.

Currently, the rapidly growing storage and the processing power needed in healthcare IT have positioned cloud tech-nologies as a good solution to provide highly scalable infrastructures. Outsourc-ing PACS to a cloud-based infrastructure can have a notable impact on small centres as they may be able to reduce the costs of hardware, network, equipment, energy and maintenance, etc. [18]. The financial ad-vantage of using the cloud for IT outsourc-ing is an important argument in many sce-narios, including the healthcare area [19, 20]. De spite the advantages, some ques-tions related to trust, data privacy, avail-ability, ownership and protection have to be considered before migrating to cloud solutions. These problems have been the focus of much discussion and are current barriers to wider adoption of cloud-based infrastructures in a healthcare environ-ment [21].

2.3 Telemedicine and Teleradiology

Telemedicine technologies allow physicians to examine, monitor, investigate and treat patients who are physically distant, or are not able to travel to the health centre [22]. In many countries where the population is highly dispersed or isolated, telemedicine emerges as a vital practice that can save lives [23]. Physicians can use telemedicine to supply qualified health services, anytime

and anywhere. Telemedicine solutions con-tribute to reducing the patient travel and the occupation of physical space in the in-stitutions. Telemedicine can also be used by specialized centres to assist remote physi -cians in particular diagnoses.

With the advancement of IT, telemedi-cine has become more mature and very im-portant for patients and healthcare institu-tions [24, 25].

Teleradiology is one of the telemedicine branches where technology has been most widely used to improve processes. This area is focused on operations with medical images in digital format, such as CT (com-puted tomography), US (ultrasound), and MR (magnetic resonance). The first steps in teleradiology date back to 1929 when a medical image was transmitted via tele-graph to a distant location [26]. Kumar [27] reported that the most frequent users of teleradiology are typically radiologists on call, rural primary-care physicians, hospital physicians in inter-departmental collaborative processes, and subspecialist physicians in remote consultations. He also presented some potential applications of teleradiology in scenarios like training of new radiologists, assisting radiologists in developing countries, and providing medi-cal assistance to isolated regions.

Many traditional tools are being ported to the Web aiming to improve radiology processes, namely inter-institutional work-flows. For example, several researchers im-plemented DICOM Web viewers based on Java applets in order to be platform-inde-pendent [28, 29], which cannot be con-sidered pure Web applications due to their dependency on external components. Mahmoudi [30] explored the implemen-tation of 3D volume rendering in a DICOM Web viewer. They used VRML [31] (Virtual Reality Modelling Language) to provide 3D features over the Web. Meanwhile, VRML became obsolete and 3D representation is now possible using the WebGL API (Application Programming Interface) introduced in the HTML5 speci-fications. Following this line of research, ArguinÞarena [32] proposed the use of flash technology to implement a DICOM Web viewer. In summary, the Web technol-ogy panorama is changing very quickly and nowadays the market encourages the



205E. J. Melício Monteiro et al.: A Cloud Architecture for Teleradiology-as-a-Service

use of HTML5, Canvas and JavaScript. Fol-lowing these trends, we previously ex-plored the use of new HTML5 features to build a zero-footprint Web viewer for medical imaging, which allows visualizing DICOM studies in any HTML5 Web browser without the need for extra compo-nents [33].

In the past, researchers have not con-sidered the ubiquitous computing as a core feature of their solutions because the focus was mainly on business environments. Ubiquitous computing consists of design-ing software that is accessible from every-where and anywhere, in a seamless manner using underlying technologies, such as the Internet, mobile computing and sensors [34]. Actually, PACS and all the appli-cations that support radiological workflows

were frequently located in the same local area networks and remote access was typi-cally supported by VPN connections. Now-adays, the local access constraint seems very restrictive when ubiquitous access to data is becoming increasingly important [35, 36].

The flow of data and its availability are very important in telemedicine to build a solid and useful system [27]. Yang imple-mented a cloud-based system for medical images called MIFAS [37] (Medical Image File Accessing System). Cloud-based sol-utions have been used to solve the problem of medical information exchange and com-pute power sharing between hospitals [38]. Also, Silva analysed the advantages of using the new cloud computing paradigm to im-prove healthcare systems [39]. They imple-

mented a PACS archive in the cloud, taking advantage of the cloud’s elasticity and scal-ability.

Some studies outline other features, such as allowing physicians ubiquitous ac-cess to patient records, medical data stor-age, database queries and medical data re-trieval. For instance, systems can be used by physicians to manage the patient health records and medical images using the emerging mobile environments [40, 41]. Viana discussed the limitations of current mobile phones and tablets, such as reduced computational power, limited storage and memory to implement a three-tier archi-tecture with a relay in the cloud [41].

Porumb created a system [42] for synchronous collaboration mechanisms among medical staff, improving telemedi-

Figure 1 Basic PACS network architecture




cine services and real time collaboration. They provided synchronous and asyn-chronous collaborative capabilities but the system was limited to a single institution.

Several web-based platforms such as Radiopaedia [43], GoldMiner [44], myPACS [45], and AuntMinnie [46] pro-vide rich collaborative repositories of radi-ology cases, with descriptive information such as patient demographics, image mo-dalities, related articles, and clinical find-ings. Although these systems are good for-ums for teaching and knowledge dissemi-nation they are not targeted for synchron-ous online collaborative work.

In the following section, we describe a cloud-based collaborative platform that makes use of social media principles and tools to facilitate the ad-hoc establishment of trust domains in teleradiology.

3. System Architecture and Implementation

The proposed collaborative architecture is made up of three main components: the collaborative platform, the Medical Man-

agement Information System (MMIS – a gateway for mediating communication with a PACS inside a healthcare institu-tion) and the Web client application (▶ Figure 2).

3.1 Collaborative Framework

The collaborative framework is the system’s core component. It is hosted in the cloud and has four main components (▶ Fig- ure 3): the Openfire server, the cloud archive, the cloud database and the XMPP protocol middleware layer.

The Openfire real time collaboration (RTC) server allows registering, managing and establishing communication between users [47]. The Openfire server uses an open-standard communication protocol, the eXtensible Messaging and Presence Protocol (XMPP) [48], for message ex-change between client and server. It sup-ports instant messaging, collaborative in-teraction and collaborative workflow, and also works perfectly in cloud-based en-vironments [49]. Openfire server supports plugins, which allows us to develop plugins for enhancing the communication middle-

ware, namely with the capacity to under-stand the new functionalities built on top of the XMPP protocol, thus enhancing the protocol for a medical scenario with fea-tures like encapsulation of DICOM query and retrieve. A user must be registered in the Openfire collaboration server to access the functionalities available. The server also provides features to manage friends and groups. Users can add other users to their roster or create affinity groups, as-sociating users with each group.

The cloud archive is responsible for managing the system storage area. The platform uses this component to store the users’ DICOM images in the cloud. This component can be seen as a pointer to a given cloud storage, or as a stream used by the system to write and read files from and to the cloud. The cloud archive and the cloud database were implemented as plugin-based components and they can be developed to use various cloud providers. The plugin-based approach also gives us more flexibility in the deployment of our system.

3.1.1 Personal Remote Archive

The cloud archive and the cloud database gave users the abstraction of a Personal Remote Archive (PRA) to store private studies, which were accessible anytime and from anywhere. These files could be shared with other users or groups. Each resource has an access control list stored in the plat-form database (i.e. the cloud database). This feature was important to promote col-laboration between users.

An instance of the Openstack Swift blobstore [50] was deployed in our private cloud infrastructure to handle the PRA, but it could also be hosted on any cloud storage provider supporting a blobstore mechan-ism.

3.1.2 XDMCP – eXtensible DICOM Communication Protocol

Focusing on the communication layer, we implemented an extension of the XMPP protocol that supports custom XMPP messages with features addressing the medical environment. XDMCP enriches the messaging middleware to satisfy the

Figure 2 Architecture overview: the cloud infrastructure (cloud database, cloud blobstore) supports the collaborative system. There is an MMIS gateway for mediating communication with a PACS. A message-oriented middleware supports communication between all entities in the platform.

Healthcare Institutions




system requirements. Particularly, Info/Query (IQ) messages were created to allow XMPP clients to manage their PRA. This includes features like sharing, obtaining and putting medical images into the PRA. Also, using specific pay-loaded messages, any XMPP client can send a DICOM c-find and c-move command to a given MMIS gateway. This allows clients to access re-mote DICOM repositories. All system en-tities have to implement specific IQ handlers and IQ providers in order to understand the new features added to the XMPP protocol.

3.2 Medical Management Information System

The Medical Management Information System (MMIS) ensures the platform’s in-teroperability with the institutional PACS. The MMIS gateway allows remote access to standard PACS server archives (▶ Fig- ure 4). This module is extremely important in the framework because it ensures the solution’s interoperability with institutional data sources. It may be used to query and import studies into the user’s private area (PRA).

The gateway was developed using the Java programming language with useful

libraries such as the XMPP smack client [10] and the dcm4che DICOM library [6]. Basically, the gateway consists of an XMPP client, developed using the Jive Smack library, which connects to the Openfire Server and receives messages from users,

controlling their access to the PACS archive server. These messages can be custom XMPP stanzas [51] that are translated to DICOM commands (e.g. c-find, c-move).

Access to the PACS server is restricted to the users on the gateway access list man-

Figure 3 Core components of the collaborative framework. The Personal Remote Archive (PRA) and the eXtensible DICOM Communication Protocol (XDMCP) are the components responsible for providing the data persistence and the communication middleware of our system.

Figure 4 MMIS – PACS Inter-operability: The MMIS gateway plays an im-portant role in estab-lishing interoperabil-ity between PACS in-frastructures and the Web platform. It en-ables physicians to access data stored in their PACS through our platform.




aged by the gateway administrator. This component is installed on a computer in-side the healthcare institution. Users with granted access to an MMIS may execute DICOM queries to retrieve medical im-aging studies. These images can also be stored in the cloud, thus enabling users to share them with others. An automatic mechanism is provided to allow anony-mous sharing of examinations. This an-onymization process changes the patient identification, demographic information and institutional information stored in DICOM metadata aiming to remove data that can reveal the person’s identity [52]. However, our system does not provide pixel data anonymization as required by HIPAA [53].

3.3 Collaborative Web Client Application

After deploying the infrastructure in the cloud, i.e. the collaborative framework oriented to teleradiology services, we also

developed a Webclient to permit remote access to the telemedicine platform. This application allows users to manage their private repository, manage their contact list, visualize, diagnose and share medical images.

The emergent HTML5 standard was used to build a rich Web application. The Strophe XMPP JavaScript client [31] was used to support communications with the server application (in the cloud). The clients use bidirectional-streams over syn-chronous HTTP (BOSH) to transport XMPP stanzas. User interaction with the Web application is very similar to their in-teraction with a desktop viewer. For in-stance, they are able to simply drag and drop DICOM files into the Web appli-cation in order to make the upload to their private repository.

Traditional collaboration in radiology is somewhat limited by the use of store-and-forward technologies, which may have dis-advantages compared to real-time technol-ogies. Therefore, we chose to implement

real-time interactive tools for collabora -tion, using the XMPP and the WebRTC technologies (▶ Figure 5).

XMPP supports the main interactive tools, which are messaging and presence exchange. WebRTC enables collaboration through video and audio in the browser without the need to install any plugin. For instance, in the collaborative mode, it is possible to have synchronized visualization of medical images supported by interactive pointers and videoconferencing.

For medical images visualization, we implemented a DICOM image viewer using JavaScript. The viewer allows direct manipulation of DICOM files within any HTML5 compliant Web browser. Perform-ance optimization was important to pro-vide a system able to deal with medical im-ages that tend to be heavy for Web-based systems. The Web-based DICOM viewer was the main target of this optimization.

Initially, image processing required some tuning in order to improve the per-formance. The image processing consists of direct transformations of DICOM image original pixel data to be able to present it in the Web browser with the original quality. The DICOM standard defines a sequence of transformations composed by modality LUT, VOI LUT, polarity and presentation LUT transformations. These must be ap-plied to the original pixel data to generate the image for display [54]. For image ren-dering we used the library KineticJS, an HTML5 Canvas JavaScript framework that enables high performance animations, layering, caching and event handling for desktop and mobile applications. Further -more, JavaScript modules were imple-mented to optimize DICOM image pro-cessing before rendering. An example of such optimization is how large images are incrementally rendered. For example, we do not process the entire matrix of large images such as a 4740 ´ 3540 CR. We start by showing an image with lower resolution fitted to the monitor resolution. Then, as the user zooms in the canvas the image is incrementally replaced by higher resol-ution ones. With this feature, processing of the entire image matrix is avoided, and it is possible to process large images in approxi-mately 10% of the original time.

Figure 5 Middleware to support interactive collaboration between Web applications. The system uses the WebRTC for audio and video call and the XMPP for messaging and presence exchange.




In addition, CPU intensive tools, such as filters and segmentation tools, were imple-mented on the server-side and provided as services accessible via REST APIs.

The use of cache mechanisms was fun-damental to increase the platform’s per-formance. On the client-side, recently ac-cessed medical images are cached locally in the devices using the HTML5 FileSystem APIs. As a result, we avoided multiple downloads of recently accessed data lead-ing to reduced data access latency. On the server-side, the search and retrieval of medical imaging studies (from the archive) may be particularly computationally inten-sive and may have a high impact on system scalability. In a concurrent environment, these tasks will increase the server CPU load dramatically and reduce its response capacity. In order to minimize this issue, we used a distributed memory object cach-ing system, named Memcached, which re-duces the number of accesses to the Blob storage.

There were other considerations focus-ing mainly on mobile devices performance in the pursuit of a ubiquitous solution. For instance, the Web application allowed adaptive image quality, due to the device resource limitations.

3.4 Privacy and Security

Privacy and security are main concerns re-lated to medical data access, and many re-searchers are targeting these problems [55, 56]. It becomes more critical when medical data are transmitted over the Internet or stored outside institutions’ physical do-mains. In these environments, confiden-tiality, integrity and even the authenticity of the data source are important. In many cases, access control mechanisms and strict security policies are mandatory [57]. Al-though they were not the main target of this article, security and privacy were of major importance in the proposed archi-tecture. We implemented several important mechanisms, for instance, event logging, data encryption, image anonymization, and a secure communication channel through HTTPS. An access control list (ACL) was implemented and only users be-longing to the MMIS gateway ACL are able to communicate with it. A role-based

access control (RBAC) module is used to define roles (e.g. administrator, healthcare professional, student, teacher, etc.), re-sources (gateway and uploaded study) and permissions (read, write, remove). A set of permissions is associated with each role, defining the kind of interactions the user could have access to. Then, resource owners are able to identify the roles required by each user to access the re-sources. The ACL is maintained on the server side so that any XMPP message is analysed before being forwarded to its des-tination. All the access control mechanisms were implemented on the XMPP middle-ware layer. The data source, i.e. the archive, can also be configured to provide access just to a data sub-set, e.g. to ultrasound studies performed in the last week.

Using a public cloud to store and man-age clinical data is a sensitive issue and must follow national legislation. The public infrastructure may bring security concerns associated with the loss of full control over stored data. Hence, it is difficult to concili-ate and ensure both the confidentiality and the search ability in cloud environments. A key component in our secure architecture

is a searchable encryption scheme – Pos -terior Playfair Searchable Encryption (PPSE) [58] – which, besides keeping the confidentiality levels of the stored data, hides the search patterns.

4. Results and Discussion

In the proposed architecture, the teleradi -ology service uses the cloud to relay all messages exchanged between entities. Thus, clients behind corporation firewalls may establish (or receive) connections with (from) external resources without prob-lems. This approach is fundamental to en-sure communication with MMIS gateways located inside healthcare institutions.

The message-oriented middleware pro-vides a set of features that allows us to manage the users’ contact list, subscribe or unsubscribe to a user’s presence, access medical images through the MMIS gate-ways and store them in the cloud, obtain stored images for visualization, and share resources.

The platform allows physicians to cre-ate easily dynamic network groups (▶ Fig-

Figure 6 Physician social media network supporting data sharing between users and work group management

data source sharing




ure 6). Thus, cooperative groups can be created to assist social relationships among physicians. This cooperation can be syn-chronous or asynchronous. The facility of creating ad-hoc networks (or trust do-mains) has the potential to improve profes-sional workflows where radiologists can cooperate and share knowledge through interactive tools.

Besides the files stored in the PRA, radi-ologists may access studies from a remote DICOM repository. The MMIS gateway ensures interoperability between the Web platform and the DICOM universe. Through this module, radiologists can query their institutional PACS and push the examinations to the PRA. Next, the examination can be visualized or shared with third-party experts. In fact, the user can retrieve and view the studies from a re-mote archive without being obliged to store them in the PRA. The platform provides a DICOM viewer (▶ Figure 7) based on HTML5 and JavaScript technology. It is a 100% Web solution, platform independent, and with support for mobile environments. It provides a set of common functionalities,

including surface and angle measurement tools, graphical annotation, window/level manipulation and image processing. It is an application that permits us to access multiple PACS archives with a few mouse clicks.

4.1 System Evaluation

In order to evaluate the feasibility, usability and performance of the proposed platform, we deployed two use cases:

4.1.1 Teaching of Radiology

One major problem of teaching radiology in health schools is that traditional PACS solutions are very limited in supporting the classroom’s workflow. Besides training the radiology trainees with the workflows they will face in their professional future, the technological solution should also support e-Learning. Namely, by providing individ-ual user areas, i.e. private DICOM archives, sharing of resources and supporting collab-orative work environments.

The proposed framework was associ-ated with the Dicoogle PACS archive to support the Radiology classrooms of ESSUA (School of Health, University of Aveiro). On the one hand, lecturers need to upload case studies (i.e. DICOM exams) to be analyzed by a student group. Thus, a public repository owned by the lecturer is required. This repository is shared with class students (read-only permissions). On the other hand, there are acquisitions per-formed by the students, for instance CT or X-Ray in phantoms. Only the student (or group member) should be able to see the acquired examinations. Next, students an-notate and apply transformations in these studies, which can be performed collabo -ratively. The final work needs to be sub-mitted for evaluation, i.e. shared with the lecturer.

During one semester, a class of 21 stu-dents used this platform for having access to medical imaging studies shared by the lecturer. In some assignments, they also used the platform for adding CT acquisi-tions that were later shared with the lec-turer. All actions were logged for evalu-

Figure 7 Web client application: This figure depicts the use of our DICOM Web viewer for analysis of a clinical case.




ation purposes and a summary can be found in ▶Annex 1. We have registered 1921 DICOM actions between the tele-radiology platform and the Dicoogle PACS archive. Several operating systems (Windows, MAC and Linux) and Web browsers (Mozilla Firefox, Safari and Google Chrome) were used, in the univer-sity campus and at home.

At the end of the semester, a question-naire was given to the students to evaluate the impact of the platform during their classes. The responses are presented in ▶Annex 2. Overall, the users’ interest con-firmed that the proposed platform is ex-tremely useful in an academic tele-teaching scenario. The direct contact with students during the classes also provided an impor-tant feedback for improving some func-tional aspects.

4.1.2 Telereporting Service

Existent statistical data and productivity studies showed that medical imaging data can be generated in practically any health-care institution, even one with limited human or financial resources [59]. Never-theless, highly skilled physicians are usually concentrated in a reduced number of specialized medical centers. The asymmet-ric distribution of equipment and service providers across countries typically leads to the need to hire third party reporting ser-vices outside the institutions where the exams were made. In the past, our work-group developed a tele-ECG platform that is being used by cardiologists to support examination reporting [35]. The tele-reporting service presented in this use case has a similar aim, but the technological ap-proach is very different. In this case, the in-formation is not pushed to a central reposi-tory but remains in the remote institution PACS. The technician makes the examin-ation in the modality and the exam is stored in the institution’s PACS archive, connected to the review platform through an MMIS gateway. The physician uses the platform to query, retrieve and visualize the examinations produced in the institutions. The examinations are temporarily cached in the PRA but the data are deleted after user logout.

We also deployed this use case in a sce-nario connecting two geo-distributed insti-tutions sharing a private regional reposi-tory. They belong to the same owner and the radiologists can practice in both insti-tutions and report examinations remotely. These institutions deal with different mo-dalities, handling an average of 3000 exam-inations monthly, with a combined volume of around 60 GB. Previously, the visualiza-tion workstations used DICOM communi-cations, supported by a routing mechanism with cache, to connect with the central

archive. In mid-2014, our platform was in-stalled in their private cloud infrastructure and they started to use the DICOM Web browser for tele-work reporting. During six months, 23164 DICOM actions (c-move and c-find) were performed between the teleradiology platform and the central archive. One of the most used functionality was the ability to collect interesting clinical cases, which can be anonymized, anno-tated and stored in their personal archive.

4.2 Performance Measurements

The cloud infrastructure provides response capacity in case of high demand for storage and processing power. However, the laten-cy associated with remote retrieval and visualization of medical imaging studies is the most critical issue in the platform. In order to demonstrate the performance of the proposed platform, trials consisted of recording the delays in moving, parsing and displaying several medical imaging studies between the PACS-Cloud archive and the remote Web client application. The studies must be retrieved from a remote PACS archive server using the DICOM standard, stored in the cloud container and parsed to extract the pixel data. Next, the pixel matrix is transferred to the remote Web client application that applies trans-formations to the original pixel data (ac-cording to the DICOM standard specifica-

Modality(rows x columns)

CT (512 x 512)

CT (512 x 512)

CT (512 x 512)

CR (2370 x 1770)

CR (4740 x 3540)

CR (2140 x 1760)

US (1024 x 768)

Number of files

65

195

24

4

4

1

25

Volume (MB)

33.4

100.2

12.3

32

128.4

7.2

7.5

Time(s)

12.4

41.5

5.9

12.8

49.0

4.1

5.2

Table 1 The table presents the access time measurements. The size of the study is the factor that has the most impact on access time. The la-tency introduced by multiple files is residual as can be observed when comparing the access time of one CT containing 25 slices with one CR con-taining 1 slice and all studies with a size of ap-proximately 7 MB.

Figure 8 Average response time in milliseconds for each task, varying the number of concurrent users




tions regarding image presentation [54]) before displaying it.

The teleradiology platform was de-ployed in a private cloud infrastructure and several instances with the following charac-teristics were used: 16 Gb of RAM, 4 vCPU at 2.6 GHz and a network connection with

24 Mb/s of upstream bandwidth. The clients were connected through channels with a downstream of 40 Mb/s. The studies’ characteristics (image matrix, the number of DICOM files and the total data volume) and average access times are presented in ▶ Table 1.

▶ Table 1 results allow us to observe that the proposed platform is able to pro-vide an average and approximate through-put of 2.5 MB/s. We also observe that the access time varies between 4 and 50 sec-onds for the studies available on our plat-form, which may be considered acceptable considering the overheads associated with DICOM communications, disk I/O oper-ations, communication latency and image processing.

We also performed a stress test to assess the impact of multiple clients using the platform simultaneously and intensively to retrieve and visualize studies from their cloud archive. In the trials, each client was programmed to request studies and per-form searches continuously. We used Lo-cust [60] to simulate user behavior in our system. This modern load-testing tool allows us to define and simulate a set of tasks performed by each user when access-ing the system. The tasks defined consist of accessing the home page, performing the login, accessing the profile, accessing static contents, performing a search and down-loading a set of images for visualization. The results obtained are presented in ▶ Figure 8.

As expected, increasing the number of concurrent users in our platform affected the server response time. The most impor-tant observation is that the maximum re-sponse time, considering 250 concurrent users, could top 1.2 seconds, which is very acceptable taking into consideration the limited resources of the virtual machine used in the tests.

As expressed in section 3.3, the search and retrieval of medical imaging studies are CPU intensive tasks with direct impact on platform scalability. The stress tests also provided data that allowed us to analyze the impact of Memcached system during downloads and searches. To simulate a real scenario where it is not possible to have everything in cache, the system was tested with distinct percentages of data (images and query results) in the cache, i.e. using missing percentages that vary from 10% to 50%. ▶ Figure 9 and ▶ Figure 10 present the impact of the caching mechanism on the download and search response times, respectively.

Figure 9 Performance improvement using Memcached: Average response time for image downloads with cache and without cache, varying the number of concurrent users

Figure 10 Performance improvement using Memcached: Average response time for study search with cache and without cache, varying the number of concurrent users




The results obtained clearly demon-strated the importance of the caching mechanism to reduce the download and query response times in the proposed plat-form. The impact is most notable when the number of concurrent users is high and also when the missing percentage is small. For the best scenario, where we had 90% of data cached and considering 250 concur-rent users on the platform, the proposed caching mechanism improved the search response time by 66% and the download time by 36%.

5. Conclusion

The purpose of the current study was to improve medical information availability and simplify the establishment of telemedi-cine expert groups. The research has also shown that it is possible to take advantage of cloud-based services and emerging Web technologies to leverage collaboration among physicians and to offer rapid access to clinical data, leading to improvements in clinical practices.

Our solution proved to be adaptable for different tele-radiology scenarios, like tele-reporting service and teaching of radiology. We have used the platform to connect two geo-distributed institutions promoting tele-work and tele-reporting. In addition, the flexibility of using a cloud-base infra-structure allowed us to deal seamlessly with the increase in storage demand, even though these institutions handle tens of GB of data monthly. The storage growth is a problem that these institutions have always been facing, leading to the need for a hard-ware upgrade and replacement. In this sense, our use case has brought another practical perspective to deal with the prob-lem by outsourcing the infrastructure to public or private cloud vendors, thus being able to increase the storage as needed with-out being concerned with hardware man-agement. Further, we have customized the platform to work as an e-Learning plat-form for a radiology class, where the radi-ology trainees have their individual user areas. They can share resources, do collab-orative work and access visualization tools for supporting the classroom’s workflow. The platform was confirmed to be useful in

a tele-teaching scenario where students can remotely access the clinical cases which the teachers share in the platform. These ad-vantages become more convincing when we analyse the students’ difficulties in for-mer classes where, to perform the same kind of analysis of clinical cases, they had to install visualization software on their computers, or they needed physical access to the workstation in their classrooms. Moreover, the management of assignments become much easier, since teachers and students can now access them directly in the platform.

The results of this research support the idea that efficient dataflow and availability of medical records are key factors for the success of telemedicine. Moreover, to be ef-fectively adopted, technical solutions need to follow current trends in software appli-cations – ubiquitous and easy to use. We believe that solutions like the one we pro-posed in this research will have an impor-tant role in the future of healthcare sys-tems, where knowledge sharing and collab-oration is becoming increasingly relevant.

6. Future Work

As future work, we aim at implementing al-gorithms that use a multitude of clinical data available on our platform to generate recommendations. For example, by ex -tracting several features from the medical image repository and from textual reports, similar studies will be provided while the physician is performing a clinical analysis.

We believe it is possible to improve the capabilities of our DICOM Web viewer ex-ploiting the WebGL features available in HTM5. Using WebGL may improve the processing of larger images, such as high-resolution mammographies.

We also aim at developing new features to improve the anonymization of DICOM images regarding annotations in the pixel data.

Acknowledgment

This work was supported by project Cloud Thinking (CENTRO-07-ST24-FEDER- 002031), co-funded by QREN, “Mais Cen-tro” program, and the EU/EFPIA Inno-

vative Medicines Initiative Joint Under -taking (EMIF grant n° 115372).

References1. Montagnant J, Duque H, Pierson JM, Breton V,

Brunie L, Magnin IE. Medical Image Content-Based Queries using the Grid. HealthGrid; 2003. pp 138–147.

2. Viitanen J, Kiuru A, Heiska K, Kärkkäinen A, Kormano M. Finnish PACS project. Computer Methods and Programs in Biomedicine 1991; 36 (2–3): 99 –102.

3. Beard DV, Parrish DM, Stevenson D. Cost analysis of film image management and four PACS sys-tems. Medical Imaging 1990: 7–10.

4. Meenan C, Daly B, Toland C, Nagy P. Use of a Thin-Section Archive and Enterprise 3D Software for Long-Term Storage ofThin-Slice CT Data Sets. Journal of Digital Imaging 2006; 19 (1): 84 – 88.

5. Yakami M, Ishizu K, Kubo T, Okada T, Togashi K. Development and Evaluation of a Low-Cost and High-Capacity DICOMImage Data Storage Sys-tem for Research. Journal of Digital Imaging 2011; 24 (2): 190 –195.

6. Huang HK. PACS and Imaging Informatics: Basic Principles and Applications. Wiley; 2010.

7. Pianykh OS. Digital Imaging and Communi-cations in Medicine (DICOM): A Practical Intro-duction and Survival Guide. Springer Science & Business Media; 2009. p 417.

8. Costa C, Silva A, Oliveira JL. Current Perspectives on PACS and Cardiology Case Study. Advanced Computational Intelligence Paradigms in Health-care – 2. Berlin: Springer; 2007. pp 79 –108.

9. Valente F, Viana-Ferreira C, Costa C, Oliveira JL. A RESTful Image Gateway for Multiple Medical Image Repositories. Transactions on Information Technology in Biomedicine 2012; 16 (3): 356 –364.

10. Costa C, Ferreira C, Bastião L, Ribeiro L, Silva A, Oliveira JL. Dicoogle – an open source peer-to-peer PACS. Journal of Digital Imaging 2011; 24 (5): 848 – 856.

11. Voorsluys W, Broberg J, Buyya R. Introduction to Cloud Computing. Cloud Computing: Principles and paradigms. John Wiley & Sons, Inc.; 2011. pp 1– 41.

12. Kaplan AM, Haenlein M. Users of the world, unite! The challenges and opportunities of Social Media. Business Horizons 2010; 53 (1): 59 – 68.

13. Chou W-YS, Hunt YM, Beckjord EB, Moser RP, Hesse BW. Social media use in the United States: implications for health communication. Journal of medical Internet research 2009; 11 (4): e48.

14. Huang HK. PACS and imaging informatics: basic principles and applications: John Wiley & Sons; 2004.

15. Buyya R, Broberg J, Goscinski AM. Cloud comput-ing: Principles and paradigms. John Wiley & Sons; 2010.

16. Cerami E. Web services essentials: distributed ap-plications with XML-RPC, SOAP, UDDI & WSDL. O’Reilly Media, Incorporated; 2002. p 288.

17. Allamaraju S. RESTful Web Services Cookbook. O’Reilly Media; 2010. p 316.

18. McCarthy CJ, Gerstenmaier JF, O’ Neill AC, McEvoy SH, Hegarty C, Heffernan EJ. “EcoRadi -




ology” – Pulling the Plug on Wasted Energy in the Radiology Department. Academic Radiology 2014; 21 (12): 1563–1566.

19. Yoo S, Kim S, Kim T, Baek R-M, Suh CS, Chung CY, et al. Economic analysis of cloud-based desk-top virtualization implementation at a hospital. BMC medical informatics and decision making 2012; 12 (1): 1.

20. Philbin J, Prior F, Nagy P. Will the next generation of PACS be sitting on a cloud? Journal of digital imaging 2011; 24 (2): 179 –183.

21. Bastião L, Costa C, Oliveira JL. Strengths and weaknesses regarding support of medical images repositories over the cloud. Imaging Management 2012; 11 (5): 14 –15.

22. Maheu M, Whitten P, Allen A. E-Health, Tele-health, and Telemedicine: A Guide to Startup and Success. John Wiley & Sons; 2002. p 400.

23. Bagchi S. Telemedicine in Rural India. PLoS Medi-cine 2006; 3 (3): e82.

24. Lu Q, Koutsakis P. Fair and Efficient Scheduling for Telemedicine Traffic Transmission over Wire-less Cellular Networks. IEEE 69th Vehicular Tech-nology Conference; 2009. pp 1– 5.

25. Chorbev I, Mihajlov M, Jolevski I. WiMAX sup-ported telemedicine as part of an integrated sys-tem for e-medicine. IEEE 30th International Con-ference on Information Technology Interfaces; 2008. pp 589–594.

26. Sending dental x-rays by telegraph. Dent Radiogr Photogr 1929: 6 –12.

27. Kumar S. Introduction to Teleradiology. Teleradi -ology. Berlin: Springer; 2008. pp 1– 9.

28. Fernàndez-Bayó J, Barbero O, Rubies C, Sentís M, Donoso L. Distributing Medical Images with Inter-net Technologies: ADICOM Web Server and a DICOM Java Viewer 1. Radiographics 2000; 20 (2): 581– 590.

29. Eng J. Improving the interactivity and functional-ity of Web-based radiology teaching files with the Java programming language. Radiographics 1997; 17 (6): 1567–1574.

30. Mahmoudi SE, Akhondi-Asl A, Rahmani R, Faghih-Roohi S, Taimouri V, Sabouri A, et al. Web-based interactive 2D/3D medical image pro-cessing and visualization software. Computer methods and programs in biomedicine 2010; 98 (2): 172–182.

31. Silicon Graphics I. The Virtual Reality Modeling Language (VRML) [16-05-2014]. Available from: http://www.web3d.org/x3d/specifications/vrml/VRML1.0/index.html.

32. Arguiñarena EJC, Macchi JE, Escobar PP, del Fresno M, Massa JM, Santiago MA. Dcm-Ar: A fast Flash-based Web-PACSviewer for displaying large DICOM images. Annual International Con-

ference of the IEEE Engineering in Medicine and Biology Society (EMBC); 2010. pp 3463 – 3466.

33. Monteiro EJ, Costa C, Oliveira JL. A DICOM Viewer based on Web Technology. IEEE 15th In-ternational Conference on e-Health Networking, Applications and Services (IEEE Healthcom); 2013. pp 167–171.

34. Lyytinen K, Yoo Y. Ubiquitous computing. Com-munications of the ACM 2002; 42 (12): 63 – 96.

35. Costa C, Oliveira JL. Telecardiology through ubi-quitous Internet services. International Journal of Medical Informatics 2012; 81 (9): 612– 621.

36. Bardram E. Activity-based computing: support for mobility and collaboration in ubiquitous comput-ing. Personal Ubiquitous Computing 2005; 9 (5): 312–322.

37. Chao-Tung Y, Lung-Teng C, Wei-Li C, Kuan-Chieh W. Implementation of a Medical Image File Accessing System on CloudComputing. IEEE 13th International Conference on Computational Science and Engineering (CSE); 2010. pp 321–326.

38. Chenghao H, Xi J, Zhanxiang Z, Tian X. A cloud computing solution for Hospital Information Sys-tem. IEEE International Conference on Intelligent Computing and Intelligent Systems (ICIS); 2010. pp 517– 520.

39. Bastiao Silva LA, Costa C, Silva A, Oliveira JL. A PACS Gateway to the Cloud. IEEE 6th Iberian Conference on Information Systems and Technol-ogies (CISTI); 2011. pp 1– 6.

40. Doukas C, Pliakas T, Maglogiannis I. Mobile healthcare information management utilizing cloud computing and android OS. Annual Inter-national Conference of the IEEE Engineering in Medicine and Biology Society (EMBC); 2010. pp 1037–1040.

41. Viana-Ferreira C, Ferreira D, Valente F, Monteiro E, Costa C, Oliveira JL. Dicoogle Mobile: a medical imaging platform for Android. Studies in health technology and informatics; 2012. p 502.

42. Porumb C, Porumb S, Orza B, Budura D. Com-puter-Supported Collaborative Work and Its Ap-plication to E-Health. IEEE 3rd International Con-ference on Advances in Mesh Networks (MESH); 2010. pp 75– 80.

43. Gaillard F, Jones J. Collaborative Radiology Re-sources: Radiopaedia.org as an Example of a Web 2.0 Radiology Resource. American Journal of Roentgenology 2009; 192 (5).

44. Kahn Jr CE, Thao C. GoldMiner: a radiology image search engine. American Journal of Roent-genology 2007; 188 (6): 1475 –1478.

45. Weinberger E, Jakobovits R, Halsted M. MyP-ACS.net: a Web-based teaching file authoring tool. American Journal of Roentgenology 2002; 179 (3): 579 –582.

46. Minnie A. AuntMinnie.com Launches New Re-source Focused on Diagnostic Imaging Centers. Tucson, Arizona; 2002.

47. Jive. Openfire. Available from: http://www.igniterealtime.org/projects/openfire/.

48. XMPP. The xmpp standards foundation. Available from: http://xmpp.org/.

49. Hornsby A, Walsh R. From instant messaging to cloud computing, an XMPP review. IEEE 14th In-ternational Symposium on Consumer Electronics (ISCE); 2010. pp 1– 6.

50. Arnold J. Openstack Swift: Using, Administering, and Developing for Swift Object Storage. O’Reilly Media, Inc.; 2014. p 338.

51. Saint-Andre P. Extensible messaging and presence protocol (XMPP). Core. 2011.

52. Digital Imaging and Communications in Medi-cine: Supplement 142 – Clinical Trial De-identifi-cation Profiles. Available from: ftp://medical. nema.org/medical/dicom/final/sup142_ft.pdf.

53. Freymann JB, Kirby JS, Perry JH, Clunie DA, Jaffe CC. Image data sharing for biomedical research – meeting HIPAA requirements for de-identifica-tion. Journal of digital imaging 2012; 25 (1): 14 –24.

54. Committee DS. Working Group 6 Printer Ad Hoc: Digital Imaging and Communications in Medicine (DICOM). Supplement 22: Presentation Look Up Table (LUT). 1998.

55. Vazquez-Naya J, Loureiro J, Calle J, Vidal J, Sierra A. Necessary security mechanisms in a PACS DICOM access system with Web technology. Jour-nal of Digital Imaging 2002; 15: 107–111.

56. Kallepalli VV, Ehikioya S, Camorlinga S, Rueda J. Security Middleware Infrastructure for DICOM Images in Health Information Systems. Journal of Digital Imaging 16 (4): 356 –364.

57. Pereira J, Lamelo A, Vazquez-Naya JM, Fernandez M, Lopez-Gestal JM, Teijero J, et al. Design and implementation of a DICOM PACS with secure access via Internet. IEEE 23rd Annual Inter-national Conference of the Engineering in Medi-cine and Biology Society; 2001. pp 3724 –3727.

58. Ribeiro L, Viana-Ferreira C, Oliveira J, Costa C. XDS-I outsourcing proxy: ensuring confidentiality while preserving interoperability. IEEE Journal of Biomedical and Health Informatics 2013; 18 (4): 1404 –1412.

59. Reiner BI, Salkever D, Siegel EL, Hooper FJ, Siddi-qui KM, Musk A. Multi-institutional Analysis of Computed and Direct Radiography Part II. Economic Analysis. Radiology 2005; 236 (2): 420 – 426.

60. Locust.io. Locust – A modern load testing frame-work 2014. Available from: http://locust.io/.


Documents

A Cloud Architecture for Teleradiology-as-a-Service...services are used every day as an ordinary commodity. Cloud computing is a recent paradigm that consists of delivering com- puting