Professor Carmelina Ruggiero
Dipartimento di Informatica, Universitý di Genova, Italy
Via all'Opera Pia 13
16145 Genova
ITALY
(Fax: +39 10 353 2154; Email: carmel@dist.unige.it)

Editor's Note: Elements of this article were published previously in the Journal of Telemedicine and Telecare (Teleradiology: A Review. Vol. 4, No. 1 pp. 25-35, 1998). The publisher, The Royal Society of Medicine,  has given its kind permission to use this material.

The need for teleradiology

In an ideal world, there would be a radiologist constantly available at every facility capable of taking radiographs. In practice, this is not feasible. The traditional solution was to transport the X-ray film to the radiologist for reporting, or to use the services of a visiting radiologist ('circuit rider'). The availability of digital transmission allows the possibility of teleradiology, which is rapidly becoming commonplace as the costs of bandwidth and hardware drop.

Teleradiology can be defined as the electronic transmission of radiological images from one location to another, for the purposes of interpretation and/or consultation [37]. Teleradiology can be used in many scenarios.

A specialist, if equipped with digital acquisition devices for X-ray images, can provide consultation to a remote location. This can include a review of medical data and images as well as direct interaction with the patient. A teleradiology system can be used to let surgeons review pre - and postoperative radiographs of patients without the need to see their patients. Also, it allows primary-care physicians to assemble patient data, including radiographs, for presentation to specialists via videoconferencing, avoiding unnecessary travel for the patient and speeding therapy. Other applications include second-opinion services and the easy brokering of X-ray services, achieving economics in organizing diagnostic interpretations and reduced costs through competition to provide services.

The most important aspects of a teleradiology system are its costs and clinical effectiveness. Key points are the reliability of the system, the quality of the displayed images, the speed of access to the images, and the ease of use [23].

The system should have adequate storage capacity to retain images for at least one week, unless hard copy is going to be stored. In many countries there is a legal requirement to store images for periods of years. Also, it should be possible to select and display previous images together with current images, with enlargement of selected areas.

A teleradiology system should have the capability of providing a radiological report along with the appropriate images. Alternatively, audible playback of the radiological report may be provided; this is quite feasible today, since many transcription systems now use voice digitization.

The accessibility of images on a computer network, which is an important aspect of teleradiology and critical to PACS (below) will bring about significant changes in the practice of radiology in the future. It will be possible for clinicians to view digital radiographs on a display outside radiology departments, with rapid access to current and previous images. Also, look for computer-aided diagnosis in the future. In many countries, economic developments are more rapid than the development of their health care infrastructures, and demands for high quality radiology may be met more easily using teleradiology.

History of teleradiology

In 1959 in MontrÈal, Quebec, telefluoroscopic examinations were transmitted using coaxial cable by Jutra [18]. Later, in the late 1960s, Bird established a microwave video link between Massachusetts General Hospital and a walk-in clinic at Boston's Logan International Airport [24]. The system included a teleradiology application.

Other teleradiology projects followed in the 1970s and 1980s in the USA, usually part of larger telemedicine programs. Although these were effective at transmitting the information needed and although users were satisfied, the projects stopped when external sources of funding were withdrawn. This suggests that they could not justify themselves on a cost-benefit basis. Limited acceptance by physicians may also have played a role.

A period of rapid growth started in the early 1990s. Two of the most important driving factors for this came from outside the medical environment.

First, in the late 1980s and early 1990s a shift towards digital communication technologies took place, so separate information transmission services, such as telephone calls, telegrams, image and document transfer, and television programming became electronically equivalent after conversion to digital formats. As a result, many telecommunications specialty markets have merged into a single market in which the single product provided is digital bandwidth. Telemedicine offers the opportunity to increase sales in the digital bandwidth market because of its high demands for bandwidth, due to the need for interactive video imaging and for the transmission of high-density still images.

Second, there is increasing demand all over the world for equal access to low cost medical care. Telemedicine enables the provision of medical care in rural and undeserved areas. Strong competition is taking place among providers of telemedicine services for winning health care contracts, for economic and medical risk reduction, and for the provision of low cost specialty services.

The elements of a teleradiology system

A teleradiology system consists of an image acquisition section and an image display/interpretation section, connected by a communications system (i.e., a network). A Picture Archiving and Communications System (PACS) is a sister technology of teleradiology that also allows storage and archiving, as well as transmission, of digital images within an enterprise -- typically a hospital.

Standards for teleradiology

The American College of Radiology, which has more than 30,000 members, is the principal organization of radiologists, radiation oncologists and medical physicists in the United States. The ACR periodically defines new standards for radiological practice to help advance the science of radiology and to improve the quality of service to patients. In 1994, the ACR developed a standard for teleradiology [37]. This standard defines goals, qualification of personnel, and equipment guidelines, as well as licensing, credentialing, liability, communication, quality control and quality improvement issues. The standard was intended to serve as a model for all physician and health care workers using teleradiology.

According to the ACR, the goals of teleradiology include: providing consultative and interpretative radiological services in areas of demonstrated need; making services of radiologists available in medical facilities without on-site radiologist support; providing timely availability of radiological images and radiological image interpretation in emergency and non-emergency clinical care areas; facilitating radiological interpretation in on-call situations; providing subspecialty radiological support as needed; enhancing educational opportunities for practicing radiologists; promoting efficiency and quality improvement; and sending interpreted images to referring providers.

The personnel involved include physicians, technologists, physicists, engineers and/or communication or image systems specialists. The equipment guidelines cover two basic categories of teleradiology systems: small and large matrix sizes. Small matrix (lower resolution) systems include computed tomography (CT), magnetic resonance imaging (MRI), ultrasound, nuclear medicine and digital fluorography. According to the ACR standard the small matrix digitization (acquisition) systems should produce 500 pixel x 500 pixel x 8 bit images or better and the small matrix display systems should produce a 500 x 480 x 8 bit display or better. Large matrix systems include digitized radiographic films and computed radiography. For these, the digitization systems should produce 2000 x 2000 x 12 bit images or better and the display systems should produce a 2000 x 2000 x 8 bit display or better.

The ACR standard requires that both small and large matrix systems include a capability for image sequence selection for transmission and display, annotation capabilities at the transmitting station including patient data and brief patient history, provision for interactive windowing at the transmitting site and provision for compression (for improved transmission rates and reduced archiving/storage requirements). For the transmission of images and patient data, the ACR standard requires that new technology systems should include the ACR/NEMA data format standard and the DICOM network standard (see below) [1,3]. For the display, the ACR requires that the luminance of the grey scale monitors should be at least 50 foot-lamberts (170 cd/m²). For large-matrix displays, the ACR requires interactive windowing, magnification, inversion and rotation functions, and the capability of making accurate linear measurements. For small matrix systems, the ACR requires accurate reproduction of the original study. The availability of a patient database and of software security protocols is also required.

Image acquisition

Image acquisition is commonly performed by a film digitizer, converting conventional radiographs to digital form for transmission over a network. Two different techniques are used in film digitization, which employ either lasers or charged coupled devices (CCD). Laser digitizers offer very good contrast and spatial resolution, but are more expensive than CCD digitizers. The latter offer comparable spatial resolution, but their contrast or grey-scale resolution is lower. However, since their performance is improving, it is foreseeable that they will become the technology of choice, because they are less expensive, smaller and easier to maintain.

An alternative to image capture onto conventional film, followed by digitization, is computed radiography (CR). CR uses phosphor storage plates to directly obtain digital images and offers broader dynamic range, which is particularly useful for applications such as portable radiographs. Because CR is becoming widespread, and its use is likely to increase, film digitization will eventually become obsolete.

Other common devices for image acquisition are CT and MR scanners. Digitization is still sometimes necessary because, although CT and MR systems generate digital data, the image formats are often not made available by the manufacturers, so that data conversion into standard formats for transmission is required.

Image acquisition can also be carried out by frame grabbing, where the analog output from a digital display, such as CT, MR or ultrasound is converted into a digital image, normally 512 x 512 x 8 bits. Video frame grabbing suffers from several limitations: data can be lost in the analog-to-digital conversion, and various window settings must be used to obtain suitable representations for bone, soft tissue, lung etc. This results in an increase in the data transmitted and in the number of images reviewed by the radiologist. Nevertheless, video frame grabbing is used in many teleradiology systems [10] owing to economic considerations and to the absence of established standards in some medical imaging environments.

Another important aspect of image acquisition is that in many cases a sequence of images is necessary to evaluate a specific clinical problem. This is the case for echocardiograms, which are normally videotaped and then reviewed subsequently by a radiologist or cardiologist, and for coronary arteriography and ventriculography imaging, which are recorded as cinÈ films normally reviewed at a later stage. Unless a brief but representative sequence of images can be singled out this results in a great quantity of data, whose cost-effective storage and transmission creates serious problems. Serious storage and transmission problems arise also in mammography, because of the extremely high spatial resolution requirements. A minimum resolution of about 4000 x 5000 x 12 bits is regarded as indispensable for an 8 x 10 inch (20 x 25 cm) field of view.

After acquisition, images are transmitted to an interpretation site using local area networks (LANs) or wide area networks (WANs). Transmission can take place either directly to a workstation or to an image server that can distribute the images to one of several workstations. In this case the technologies of PACS at the transmission place are used, including storage facilities.

Image display

Display monitors are crucial to teleradiology, which depends on the ability to display images that are perceived to be identical to those available on conventional or laser-printed film. An important issue is image fidelity, which is measured by parameters which can be physically measured (luminance, dynamic range, distortion, resolution and noise) and by parameters which can be measured with psycho-physical techniques, such as receiver operator characteristic (ROC) analysis and tests for threshold contrast with contrast detail patterns [7].

Grey scale monitors are used for primary diagnosis for CT, magnetic resonance imaging, digital fluoroscopy, ultrasound and scintigraphy, and more recently for thoracic and musculoskeletal radiology [7]. Teleradiology of mammography images has been performed experimentally [8], but the requirements of very high image quality make current commercial teleradiology systems inappropriate at present.

The displayed image is not necessarily identical to the stored image. Digital imaging systems, unlike conventional film-screen systems, physically separate the image receptor and the image display. The stored image may be very rich in contrast or detail, so that the information in it may exceed the capacity of the display terminal. The data read from the stored image must then be processed selectively before being displayed. Also, it is necessary to match the displayed image to the human visual system and provide user-friendly tools for exploration of the stored image. From the observer's point of view, the displayed image has three important attributes: fidelity, informativeness and attractiveness [20]. Image fidelity from the observer’s point of view can be expressed in term of spatial resolution, grey scale resolution, grey scale linearity and noise. Image informativeness can be expressed in terms of the visibility of diagnostically important features or as the detectability of some specific abnormality. Image attractiveness relates to the aesthetic properties of the displayed picture [20].

In terms of image fidelity, the need for equivalence of displayed spatial resolution and of the resolution of film-screen systems has been questioned. The issue is not settled [20]. Apart from the need to display details in the stored image, the pixel size of the displayed image can also be considered from a visual perspective. The visibility of pixel boundaries interferes with contrast perception and global picture perception. Taking into account the threshold contrast of the human eye at different spatial frequencies and at different displayed luminances, it can be concluded that once the pixel size is set for a given viewing distance, moving the observer closer does not increase the visibility of detail because the pixel boundaries may become visible and pixel clutter will reduce contrast sensitivity. Moreover, the sensitivity to contrast and detail depends not only on pixel size but also on the display luminance.

In terms of grey scale rendition and contrast enhancement, variations in the intensity of each pixel in the stored picture can be represented using 8 bits, or preferably 10 bits resulting in 1024 intensity levels. However, a typical television monitor can linearly display only 6 bits, or 64 discrete grey levels. Intensity transformation tables are used, which take into account the fact that the human visual system does not respond to light intensity in a linear way, so the relation between luminance (a physical variable) and brightness (a perceptual variable) is not linear. Moreover, vision is not equally sensitive to contrast at all levels of display intensity, but small differences in intensity are easier to discern at high intensity levels. Therefore, display systems should be designed so that displayed images can be modified so that maximum contrast is achieved in all portions of the image. Further aspects related to the perception of contrast, detail and form, and of factors that influence observer performance may be found in [20].

Image transmission

The choice of the telecommunication medium for a teleradiology system requires finding a cost-benefit trade-off between expense and bandwidth. The higher the bandwidth, the more rapid the transmission and the greater the capacity of the network -- and the higher the cost. The main aspects to be taken into account for the telecommunications solution for a teleradiology system are the number of cases to be sent, the average size of the files, the required turnaround time and the peak activity.

Local area networks (LAN)

A LAN is an information transmission medium equally shared by all connected stations, limited to a local area without crossing any public areas. The transmission speed in LANs is typically between 4 Mbps and 100 Mbps. LANs generally have a service diameter of not more than a few kilometers and are completely owned by a single organization. The medium performance LANs are based on Ethernet, whereas for high performance LANs there are several standards, such as FDDI, 100 Mbps Ethernet and ATM. Further details may be found in [35] and [5].

Wide area networks (WAN)

WANs typically span entire regions or countries, have data rates below 1 Mbps and are owned by multiple organizations. The carrier owns the communication subnet and numerous clients own the hosts. The available telecommunications solutions depend on the existing infrastructure. Often a discrepancy exists between need and availability of communications services for telemedicine. For example, places which greatly need teleradiology and telemedicine services are often located in remote areas in which the latest advances in telecommunications technology are not available.

The data rate that can be achieved commonly with a fast modem on the standard public telephone network (the PSTN) is 28.8 Kbps, although faster rates, to about 44 Kbps, are possible.

ISDN networks offer a bandwidth of up to 2 Mbps for users in Europe (1.5 Mbps in the USA). ISDN services are quite widespread, but in some locations transmitting medical images over ordinary telephone lines may be the only possibility.

Image transmission over the Internet is now possible. However, image transmission speed over the Internet is often low in practice since the efficiency of the transfer depends on the global throughput at the time at which the transmission takes place. At present, the Internet can be useful in teleradiology for education and training, but its potential use would increase if dedicated medical networks become available so that the global traffic on the network is not heavily influenced by commercial information.

The highest performance transmission protocol for image communication is Asynchronous Transfer Mode (ATM). However, this service is not widely available at present and normally requires fiber optic cabling between the transmitting and the receiving station. ATM is a packet switching technology, in which the data to be transmitted is divided into cells (packets) which may arrive from various sources in a random and discontinuous fashion. ATM provides integrated support for a variety of communication services, due to its ability to manage asynchronous and synchronous traffic, to scale according to demand-oriented growth, to integrate different communication systems and to support virtual networking [26]. Although ATM technology is very recent, ATM products and services are now becoming available, and it is felt that ATM technology is emerging as a leading candidate for medical image transmission in both LAN and WAN applications [16]. More information on ATM may be found in [15, 35].

PACS and Teleradiology

Many teleradiology systems require PACS at transmitting and receiving sites. However, whereas PACS use LANs, teleradiology requires WAN technology. Due to the differences between LAN and WAN technology, integration problems arise when both teleradiology and PACS facilities are required. ATM technology satisfies the requirement that no physical or logical boundaries should exist between LANs and WANs [15].

Recently, some teleradiology systems with PACS have been set up using ATM. Some examples follow.

A WAN and LAN tested network was set up in 1994, connecting the University of California at San Francisco, Mount Zion Hospital, and the San Francisco VA Medical Center [15]. Subsequently, a large scale ATM OC-3 LAN and WAN connecting the locations above has been designed and implemented. The results demonstrate that the transmission rate between two workstations can reach 5-6 Mbps from a redundant array of inexpensive disk (RAID) to memory, and 8-10 Mbps from memory to memory. When the server sends images to all four workstations of the system simultaneously, the transmission rate to each workstation is about 4 Mbps. Both situations are adequate for PACS and teleradiology applications.

An experiment begun in 1996-1997 aims to test the medical usability of the European ATM network in medical image transmission. The Department of Radiology of the University of Pisa (Italy) and St. Luc University Hospital in Brussels (Belgium) established several connection sessions over the European ATM network to assess the usability of DICOM image transmission and interactive telediagnosis tools in daily radiological practice [25]. The connection between the two sites was available for a period of two weeks, at 2-Mbps bandwidth, and allowed the transmission of MR images (256x256x12 bit) and the simultaneous interactive discussion of the cases. All performance aspects of the system were successfully tested.

Image compression

Many teleradiology systems include image compression facilities, in order to obtain transmission rates compatible with an efficient teleconsulting service and to reduce storage requirements.

Image compression may be lossless (reversible) or lossy (irreversible). The advantage of lossless compression is that the original image can be recovered - there can therefore be no subsequent claim that important information was lost as a result of the compression process, which could be crucial in the event of legal action. The advantage of lossy compression is that higher degrees of compression can be achieved and therefore transmission times reduced. The effects of image compression on transmission times are illustrated in Table 1.

The main stages that may be present in radiological image compression are image transformation, quantization (which is present only for irreversible compression), and entropy encoding [38], [19], [29].

Image transformation is often performed in order to eliminate redundant information, to reduce the dynamic range and, in general, to obtain a representation that can be coded more efficiently.

Quantization achieves compression by representing transform coefficients with the minimum precision necessary to achieve the desired image quality. To simplify the quantization process, the information in the transformed image is compacted into a minimum number of coefficients, which are quantized according to a table specified as an "input" to an encoder. Quantization is inherently a lossy technique, and the type and degree of quantization has a great impact on the quality of a lossy compression.

Entropy encoding is a lossless compression process based on the non-random statistical characteristics of the transform coefficients. Entropy encoding consists of a conversion of coefficients into a sequence of symbols by a statistical model, followed by conversion of the symbols into a data stream in which the symbols have no externally identifiable boundaries. Code tables - predefined or adaptive - are used. The same tables used for compressing an image are needed to decompress it.

Direct coding of medical images by entropy encoding does not achieve a very high degree of compression, and a prior decorrelation is needed. All coding techniques for medical imaging use a predictive model or a multiresolution model, or both, to reduce the statistical redundancy during image transformation, and then encode the residuals. The most commonly used encoding schemes for medical image compression are Huffman coding and run-length encoding (RLE)[29].

Huffman coding is based on the idea of assigning short code words to the most probable messages and the longer code words to the least likely messages. The digital picture is regarded as a sequence of source messages, which may be the grey level of individual elements or, alternatively, other information such as pairs of neighboring pixels or arrays of elements of the original array. Huffman coding produces a code with a low average length. It guarantees that a uniquely decodable code can be obtained with the minimum average number of bits per message. However, its variable length makes it more difficult to implement than a fixed length code. Moreover, a change in the digital image requires a new code mapping to ensure minimum length.

Run-length encoding uses picture element-to-element correlation by a simple reversible technique. This technique is based on the definition of a "run" in the digital image as a sequence of consecutive pixels of identical values along one direction. If long runs occur, transmitting the start and length of the run rather than the individual pixels results in a reduction in the average bit rate. The efficiency of run length coding increases if the number of grey level transitions or edges is lower, so this method is most suitable for images with little edge and texture content. The method is also quite sensitive to errors. An extension of run length coding to two dimensions is area coding, in which an area is characterized by a continuous group of picture elements with identical values.

Lossless compression

A representation of a digital radiograph by a list of pixel values always contains redundant information, because the statistical behavior of the pixel values is not taken into account. Lossless coding methods exploit this with mathematical techniques that do not cause any information loss. The major advanced lossless techniques are differential pulse code modulation (DPCM), hierarchical interpolation (HINT), difference pyramid (DP), bit-plane encoding (BPE) and multiplicative autoregression (MAR).

DPCM is a simple and often used predictive coding method, which exploits the property that the values of adjacent pixels in an image are often similar and highly correlated. In DPCM an image is encoded one pixel at a time across a raster scan line and the value of a pixel is predicted as a linear combination of a few neighboring pixel values, which have been previously reconstructed. DPCM has been carefully studied, especially for lossless compression of medical images, for which it generally achieves average compression ratios ranging from 1.5 to 3.

The HINT method is a variable-resolution pyramid coding scheme based on subsampling. It starts with a low-resolution version of the original image and successively generates the higher resolutions using interpolation. The image data at the lowest resolution method is entropy coded and transmitted. Subsequently, an interpolation scheme is used to generate estimates of the unknown pixel values at a higher resolution by calculating the average of its four nearest neighbors at the immediate lower level. The estimates are rounded to their nearest integers and subtracted from the true values and the difference signals are also coded and transmitted. It has been shown that 2-D HINT gives compression ratios from 1.4 for 256 x 256 x 12 bit MR images to 3.4 for 512 x 512 x 9 bit angiographic images.

DP is another kind of compression method based on the variable-resolution model. It is based on the construction of a mean pyramid and on the subsequent calculation of a difference pyramid containing the differences between successive levels of the mean pyramid. Difference pyramids at several levels are entropy-coded and transmitted.

With BPE the reconstructed image is of the same size as the source. Using single bits from the same position in the binary representation of each pixel value, an n x n image called a "bit plane" can be formed. Repeating the process for the other bit positions of each pixel of the original image produces a set of p n x n bit planes, which can be transmitted in sequence, with the most significant bit plane first and the least significant bit plane last. The reconstructed image is binary and additional grey levels are added as more bit planes are received. Bit plane encoding takes advantage of the existence of large uniform areas in each bit plane, which allow to it to achieve useful compression. Successful medical image applications of BPE are images with areas with low variation in pixel values, such as soft issue regions in CT images.

MAR has two versions: 2-D MAR and 2-D multi-resolution MAR (MMAR). In 2-D MAR the image is subdivided into smaller blocks over each of which the data are assumed to be locally stationary and representable by a 2-D linear stochastic model. A MAR encoder consists of a parameter estimator, a 2-D MAR predictor, a rounding operator and a lossless encoder for the residuals. In MMAR the MAR structure is adapted to multi-resolution image representations by filtering and subsampling the original image into three resolutions and coding them in an interpolative manner. For a limited set of radiographs and MRI image MAR has been shown to perform better than HINT, DPCM, DP and RDP, but more extensive sets of image should be considered. MAR coding techniques are more difficult to implement and slower than other lossless techniques.

Lossless techniques achieve maximum compression ratios in the range between 1.5:1 and 3:1. However, for a substantial practical and economic impact, compression ratios closer to 10:1 or 20:1 are required.

Lossy compression

There is growing evidence that lossy compression can be implemented without compromising the diagnostic content of images. The volume of data to be transmitted depends on the type of radiological examination and on the number of images per study. An average conventional radiographic study may require four radiographs, digitized at 2048 x 2048 pixels with 12 bits precision. This corresponds to a data file size of about 32 MByte (MB).

Mammography images produce larger file sizes. An average study requires four images, and the required size of a digitized mammogram is 4096 x 5120 pixels with 12-bit precision. This corresponds to a data file size of about 160 MB.

Computed tomography magnetic resonance, ultrasound, and nuclear medicine images produce smaller data files than digitized radiographs. Computed tomography examinations consist of 512 x 512 pixels display. Magnetic resonance images are 256 x 256 pixel x 12 bits and a study may contain about 50 images, so the resulting size of the data file is about 6.3 MB.

Ultrasound and nuclear medicine images only require 8 bit precision. The average number of images is 256 x 256 pixels for ultrasound images, and 128 x 128 for nuclear medicine images, so the resulting file sizes are 1.5 MB and 0.4 MB respectively.

Lossy image compression techniques allow much higher compression ratios than lossless compression techniques. For this reason, recent research activity has focused on lossy image compression. The most widely used algorithm today is the standard of the Joint Photographic Experts Group (JPEG), which was not originally created for medical applications [17]. The JPEG method is available on many types of computer and is inexpensive, but it suffers from artifacts that create artificial edges to which the human visual system is quite sensitive. However, recent extensions of JPEG have improved its performance, achieving satisfactory results in some cases of radiograph compression.

An alternative algorithm, the wavelet transform has been recently used with success for compression of high resolution requirement images, such as those from mammography and other X-ray techniques, [21] and [22].

The lossy methods which are presently under study for medical imaging include techniques based on linear transforms, that represent the pixel data compactly in a spatial-frequency-like domain, such as 2- D discrete cosine transform (DCT), full frame DCT, lapped orthogonal transform (LOT), subband coding. Other techniques include vector quantization, quad trees and adaptive predictive coding.

Both JPEG standard and wavelets derive from techniques based on linear discrete transforms: DCT for the JPEG standard and subband coding for wavelets. The DCT is a transform that constructs a set close to the image-specific set of basis functions that correspond to the normalized eigenvalues of the covariance matrix of the image, which provides maximum decorrelation and entropy reduction. In many applications of the DCT for image compression, the original image is divided into adjacent blocks. The JPEG compression standard is based on DCT with division into 8 x 8 sub-matrices. The DCT is computed for each block and a regionally adapted quantizer is applied to the transform coefficients.

The subband coding algorithms are based on a set of filtering operations which divide the image into spectral components or bands. This image decomposition can be accomplished with a wide range of orthogonal and non-orthogonal transform schemes, of which wavelet transforms are a category.

Some examples of compression on a mammography are shown in figures 1-5.

Interconnection of radiology systems

In the late 1970s, and with the increasing use of computers in clinical applications, the need was felt for a standard method for transferring images and associated information between devices manufactured by various vendors, some of which maintain the image information in proprietary format. This happens, for example, for CT and MR imaging scanners, so it may be necessary to convert the data into a standard format for transmission over a communications network.

The ACR-NEMA and DICOM standards

In 1982 the American College of Radiology (ACR) and the National Electrical Manufactures association (NEMA) formed the ACR-NEMA committee to develop a standard to promote communication of digital image information regardless of device manufacturer, in order to facilitate the development and expansion of PACS, to allow the creation of diagnostic information data bases for remote access, and help ensure the usability of new equipment with existing systems.

The first version, ACR-NEMA Standards publication No.300-1985, published in 1985, was designated Version 1.0. It was followed by two revisions, No. 1 dated October 1986 and No.2 dated January 1988. In 1988 the ACR-NEMA Standards publication No. 300-1988, designated Version 2.0, was published, which included version 1.0, the published revisions, and additional revisions. Moreover, Version 2.0 also addressed point-to-point image transmission providing command support for displays, a hierarchy scheme to identify an image, and the possibility of adding data elements for increased specificity when describing an image. It also provided semantic rules by which messages (streams of bits in transit from one device to another) were organized.

Version 3.0, also referred to as Digital Imaging and Communications in Medicine (DICOM), was released in 1993.

The DICOM standard provides enhancements to the previous ACR-NEMA versions 1.0 and 2.0 in several respects. The most important innovation is the use of information modeling for the design basis of the standard and for the development of the data structures. Moreover, DICOM is applicable to networks whereas the previous versions were only applicable to point-to-point connections. The DICOM standard encourages open systems interconnection of imaging equipment over standard networks while maintaining compatibility with earlier point-to-point connection standards. Finally, DICOM specifies levels of conformance in detail and specifies how devices claiming conformance to the standard must react to commands and data being exchanged.

Information modeling in DICOM

The ACR-NEMA standards, Version 1.0 and 2.0, relied on an implicit model of the information used in radiology departments, in which the data elements were grouped on the basis of the experience of the designers. In contrast, the DICOM standard uses object-oriented analysis, and an information model in which the information is organized into a formal structure. In this model essential relationships are identified, classified and abstracted. DICOM uses explicit and detailed models (entity - relationship models) of how the patients images, reports etc. (identified as "objects") are described and of how they are related.

This information modeling is the basis for developing the data structures used in DICOM. The advantages are the reduction of redundancy and ambiguity.

Information objects are defined only in terms of their fundamental qualities or values; related types of information on objects are grouped into information object classes in which each individual member is an instance of the class type. Properties of inheritance and hierarchy determine the attributes of objects and allow the definition of levels (superclasses, classes, and subclasses) avoiding overlap or duplication of attributer and sharing of attributes.

The DICOM standard defines two types of Information Object Classes: normalized and composite. Normalized Information Object Classes include only those Attributes inherent in the real-world entity represented. Composite Information Object Classes may also include Attributes which are related to but not inherent in the real world entity. An example of the difference between normalized and composite Information Object Classes is the patient name attribute. The study Information object class, which is defined as normalized, contains study date and study type attributes because they are inherent in a study but not patient name attribute because this attribute is inherent in the patient and not in the study. On the contrary the Computed Tomography Image Information Object Class, which is defined as composite, contains both attributes which are inherent in the image and attributes which are related to but not inherent in the image such as patient name. Normalized and composite object classes have been defined to facilitate future growth of DICOM and to maintain compatibility with systems still using ACR-NEMA version 1.0 or 2.0.

A further noteworthy feature is the specification of an established technique for uniquely identifying any information object, which makes it easy to define relationships between information objects unambiguously as they are acted upon across the network. This is achieved by assigning each class of information objects a unique identifier value which consists of a prefix assigned by an ISO member organization and of a suffix assigned by the local organization. Each local organization is responsible for developing its own unique coding system for suffixes. The prefix uniquely identifies a specific organization, and the suffix is unique for a particular individual belonging to that organization. The combination of the prefix and suffix forms a unique identification number.

It is also worth noting that the DICOM standard, unlike ACR-NEMA versions l.0 and 2.0, introduces explicit information objects not only for images and graphics but also for reports, studies, etc.

Conformance with the DICOM standard

The DICOM standard provides a rigorous treatment of the issue of conformance, specifying levels of conformance in terms of specifically-defined service classes in which the functional units are precisely described. Moreover, the structure of a manufacturer's conformance claim is explicitly defined. This does not prevent manufacturers from implementing any function in their software. However if they claim to conform to the DICOM standard the conformance claim precisely states which services are supported by their product and which are not. Further details on DICOM may be found in [1], [3], [9], and [30].

Clinical acceptance and utilization of teleradiology

Since 1994 more than 7,000 teleradiology systems have been sold by two of the largest manufacturers. In the first six months of 1994 approximately 15 teleradiology programs were operational in North America, providing services to about 90 remote sites. Approximately 22,000 studies were interpreted. In 1993, approximately 2,250 patients were seen through non-radiology teleconsultations in the United States and Canada, and in 1994 there were approximately twice as many. Most telemedicine programs in the USA are financed (at least partially) by state funds. Military applications are being carried out as well as applications in the civilian sector [28].

The efficacy of commercially available digital teleradiology systems has been assessed for several scenarios. Examples include teleradiology workstations versus radiograph readings in emergency medicine, in subtle orthopedic fractures, and high-resolution teleradiology. These assessments concluded that the teleradiology equipment was reliable and effective [2], [6], [11], [13], [18], [23], [24], [33], [34].

Technology trends suggest that in the near future health care providers will be able to see patients at remote sites using desktop workstations or laptop computers. Simple software shells will be available for access to multimedia patient records, radiographs, pathology images etc. Also, on line libraries on medical information and on decision support systems will be accessible.

References

[References are found at our website, www.telemedtoday.com, under the "Articles" section of our Home Page.]

1. ACR-NEMA Digital Imaging and Communications in Medicine (DICOM) - 1992 National Electrical Manufactures Association, 1993.
2. Batnitzky S, Rosenthal SJ, Siegel E et al. Teleradiology: an assessment. Radiology 177:11-17, 1990.
3. Bidgood W, Horii S. PACS mini refresher course - introduction to the ACR-NEMA DICOM standard. Radiographics 12:345-355, 1992.
4. Cannavo MJ. PACS prices, performance showing improvement: Health Management Technology, pp 22, 24, 44, Feb 1996.
5. Comer DE. Internetworking with TCP/IP. Vol I: Principles, Protocols, and Architecture - Second Edition. Prentice Hall, Engelwood Cliffs NJ 1991.
6. De Corato DR, Kagetsu NJ, Ablow RC. Off-hours interpretation of radiologic images of patients admitted to the emergency department: efficacy of teleradiology. Am J of Roentgenology, 165:1293-1296, 1995.
7. Dwyer SJ, Stewart BK, Saire JW et al. PACS mini refresher course - performance characteristics and image fidelity of grey scale monitors. RadioGraphics 12:765-772, 1992.
8. CAR Proceedings paper on teleradiology for mammograms.
9. Fortner B. Scientific data formats, Part VIII. Scitech Journal 5(1):333, 1995.
10. Golberg MA. Teleradiology and telemedicine. Radiologic Clinics of North America, 34 (3):647-665, 1996.
11. Goldberg MA, Rosenthal DI, Chew FS, Blickman JG, Miller SW, Mueller PR. New high resolution teleradiology system: prospective study of diagnostic accuracy in 685 transmitted clinical cases. Radiology 186:429-434, 1993
12. Greinacher CFC, Bach EF, Herforth M, Luetke B, Senfert G. Computer-assisted radiology - requirements and solutions for digital diagnostic imaging. Medical Informatics 15(1):21-29, 1990.
13. Halvorsen PA, Kristiansen IS: Radiology services for remote communities: cost minimisation study of telemedicine. BMJ: Biomedical Journal 312:1333-1336, 1996.
14. Hall EL. Computer image processing and recognition. Academic Press, New York, NY 1979.
15. Huang HK, Arenson RL, Dillon WP et al. Asynchronous transfer mode technology for radiologic image communication. Am J Roentgen 164:1533-1536, 1995.
16. Huang HK, Wong AWK, Zhu X: Performance of asynchronous transfer mode (ATM) local area and wide area networks for medical image transmission in clinical environment. Computerized Medical Imaging and Graphics 21(3):165-173, 1997.
17. JPEG - Still image data compression standard. New York: Van Nostran Reinhold, 1993.
18. Jutra A: Teleroentgen diagnosis by means of videotape recording. Am J Roentgenology 82:1099-1102, 1959.
19. Kuduvalli GR, Rangayyan RM. Performance analysis of reversible compression techniques for high-resolution digital teleradiology. Institute of Electrical and Electronics Engineering Transactions on Medical Imaging, 11(3):430-445, 1992.
20. Kundel HL. Visual perception and display terminals. Radiol Clin of N America 24(1):69-78, 1986.
21. Laine A, Fan J, Yang W. Wavelets for contrast enhancement of digital mammography. Institute of Electrical and Electronics Engineers in Medicine and Biology 14(5):536-549, 1995.
22. Lucier BJ, Kallergi M, Qian W et al. Wavelet compression and segmentation of digital mammograms. J of Digital Imaging 6(1):1-13, 1994.
23. MacMahon H, Giger M. Portable chest radiography techniques and teleradiology. Radiol Clinics of N America 34 (1):1-20, 1996.
24. Murphy RL, Bird KT. Telediagnosis: A new community health resource. Observations on the feasibility of telediagnosis based on 1000 patient transactions. Am J of Pub Health, 64:113-119, 1974.
25. Neri E, Thiran JP, Caramella D et al. Interactive DICOM image transmission and telediagnosis over the European ATM network. IEEE Trans. On Information and Technology in Biomedicine 2(1):35-38, 1998.
26. Park B, Bashshur R. Some implications of telemedicine. J of Communication 161-166, 1975
27. Pelikan E, Ganser A, Kotter E, Schrader V and Timmurmann U. Experience with PACS in an ATM/Ethernet switched network environment. IEEE Trans. On Information Technology in Biomedicine 2(1):26-29, 1998.
28. Perednia DA, Allen A. Telemedicine technology and clinical applications. JAMA 273(6):483-488, 1995.
29. Rabbani M, Jones PW. Digital image compression techniques, SPIE Optical Engineering Press, Bellingham, Washington, 1991.
30. Ratib O, Hoehn H, Girard C, Parisot C. PAPYRUS 3.0: DICOM - compatible file format. Medical Informatics 19(2):171-178, 1994.
31. Richardson WB. Applying wavelets to mammograms. Institute of Electrical and Electronics Engineers Engineering in Medicine and Biology 14(5):551-560, 1995.
32. Russo G, Talone P, Caramella D: Lossy techniques for radiogram coding: application environment and constraints. Mediterranean Conference on Telemedicine, Capri, 1995.
33. Scott WW, Bluemke DA, Mysko WK et al. Interpretation of emergency department radiographs by radiologists and emergency medicine physicians: teleradiology workstations versus radiograph readings. Radiology 195:223-229, 1995.
34. Scott WW, Rosenbaum JE, Ackerman SJ et al. Subtle orthopedics fractures. Teleradiology workstation versus film interpretation. Radiology 187:811-815, 1993.
35. Tanenbaum AS. Computer networks. Prentice Hall, Engelwood Cliffs NJ 1989.
36. Thrall J. Telemedicine exports medical expertise. Diagnostic Imaging 37, 1995.
37. The ACR Standard for Teleradiology: www.acr.org
38. Wong S, Zaremba L, Gooden DS, Huang HK. Radiology image compression. Proceedings of the Institute of Electrical and Electronics Engineering 83(2):194-219, 1995.

Table 1. File size and transmission times at various speeds (according to data from [32])

*Compression Ratio

Figure legends

Fig.1: Mammography with visible retroareolar microcalcifications. Dimension: 740000 bytes (740x1000 pixels, 8bits/pixel).

Fig.2: Same image of fig.1 with 4 bits/pixel (370.000 bytes). Much relevant information is lost.

Fig. 3: Same image of fig. 1 with a conservative JPEG compression (90.112 bytes). Differences among fig. 1 and fig. 3 are visible (on a high quality screen) only if a 5:1 zoom is applied

Fig. 4: Same image of fig.1 with JPEG compression optimized for dimension (23.552 bytes). Differences among fig. 1 and fig. 4 are visible (on a high quality screen) only if a 3:1 zoom is applied

Fig. 5: 3:1 zoom of the region with microcalcifications from fig. 1 (left) and from fig. 4 (right)