Introduction
Economic pressures combined with accelerating technological advances have led to the reorganization of health care delivery. These changes have been dramatically witnessed in the field of radiology, due at least in part to its intimate connection with technological advances. Hospitals (and eventually all patient care centers) are beginning to move toward the elusive goal of an electronic medical record (EMR) system. Radiologic images will be part of this EMR, and the modeling architecture used in image management will serve as a model and major component of information technology systems.
The radiology practice of the future involves similar processes used in today's practice of radiology, but is arrived at through different means with the use of information technology and computers. Imagine, for instance, the following scenario:
- A new patient arrives in the radiology department for a magnetic resonance (MR) scan of her brain.
- Upon arrival, her electronic medical record is accessed.
- The record is automatically scanned to access the clinical guidelines and appropriateness of the examination regarding the patient's tentative diagnosis.
- The request for the examination has also arrived electronically and has been screened appropriately–defining the problem and pertinent clinical information.
- Guidelines regarding the appropriate image sequences to be used and appropriate post–processing have been standardized and forwarded to the technologist.
- The images are interpreted by the radiologists with the use of decision aides and educational resources.
- Images from prior studies performed on the patient are automatically retrieved for comparison from the image database.
- Comparison is made to other cases of other patients who had similar findings or diagnoses.
- A diagnostic assessment integrating these findings, the patient's specific information, and the experience of the radiologist is compiled.
- These results, along with pertinent images, are then forwarded through a central network to the referring physician's office, where both images and the consultation can be reviewed.
- While all of the above is taking place, the methods and pathways by which information is processed is automatically and continually monitored and improved by the network.
- Outcomes analysis and cost effectiveness data from each evaluation is fed back into a database, which will be applied to future analyses.
Although this hypothetical scenario is somewhat futuristic, radiology information systems are evolving in this direction. Different institutions are incorporating such changes with variable expediency due to a number of pressures ranging from cost savings, efficiency, accuracy, efficacy, and the ability to accurately and continually assess outcomes.
This paper attempts to provide an overview of the present state and future directions of information and image management in radiology, with emphasis on network architecture and computer–aided diagnostic tools.
1. Background
Cost reduction and cost effectiveness are the two primary beacons that have guided the health care reform we see today. Health Maintenance Organizations (HMO) and other managed care plans were established in order to decrease the costs paid by a business to fund their employee healthcare plans. The market leverage assumed by HMO's motivated health care providers to respond: Healthcare workers began to consolidate into informal networks or actually merge; radiology groups, physicians, and hospital–based organizations combined. These mergers provided the expanded ability to negotiate larger service contracts with HMO's. Additionally, such mergers provided cost savings resulting from the consolidation of administration and other expenses. These regional mergers reflect the direction of medical information and imaging technology towards a unified patient care system without boundaries [1–3].
In motion now is a shift in the focus of radiology computing and information systems from the department level to a multi–hospital based enterprise level. (The integration of these systems will be discussed later in this paper.) The data accumulated within such a system provides information regarding the cost effectiveness and efficacy of specific radiologic studies. Such information is useful for both health management organizations and healthcare providers. In addition, such information may be used for educational and training purposes and may, in the future, be applied in the remodeling and design of work within the radiology system.
The speed at which information derived from radiologic studies is communicated, and the impact such information will have on patient decision–making, health and economic outcomes will be of the utmost importance in the future. As these networks evolve, there will be anticipated improvement in quality. The integration of image technology will be utilized not only in diagnosis but also in treatment. For example, visualization of anatomy by surgeons prior to operations will become commonplace. As cost effectiveness and quality become governing principles for deciding an appropriate course of action there will be more intimate interaction between image–oriented specialties and the therapeutic arm of interventionists and surgeons. Radiologists will therefore need to identify themselves in terms of quality service and expertise in consultation. The expectations of patients and referring physicians will also be raised to a new standard [1,3].
2. Networks for Electronic Radiology
These trends are stimulated in part by economic pressures on healthcare, but are also enabled by advances in technology — e.g., cheaper, faster, and better workstations hardware, software, and communication technology. It is, in fact, sometimes difficult to identify whether technology leads to organizational change or vice versa. New organizational forms that deliver regional services through networking have empowered this evolution. Teleradiology systems providing service to locations all over the country, and in some instances worldwide, exist and operate today. Similarly, the data set provided through the information base regarding cost effectiveness and efficacy allow managed care enterprises to better implement and continually update decision making processes.
Communication and integration of services are key to the growth and implementation of such information technology. Radiology departments have long held the edge in such technology and are natural participants or leaders in the future integration of information technology into both radiology and clinical systems. Data collected about the effectiveness of radiological procedure will be able to guide the selection of optimal work–up and treatment methods. Reports will be structured to force the collection of clinical databases capable of being automatically monitored for guidelines and outcomes research.
Network–based architectures are essential to the evolution of the new radiology department. Specification of servers for specific purposes (e.g., images, medical history, Medline, medical knowledge, discussion aids, etc.) has led to the need for a wide connectivity environment in which a user workstation may access functions by a variety of servers which are integrated through applications and workstation software [4].
Image managers, storage systems, and workstations supporting high–performance image presentation provide opportunities to reengineer the practice of medicine and radiology. The distribution of information and the storage and presentation of images necessitate the availability of a communication network. A seamless view of complete electronic medical records and images to both the primary care physician and the radiologist through a distributed medical enterprise is essential to the eventual future success of such endeavors.
The imaging portion of the electronic medical record is the most demanding component, due to the volume of digital data associated with an image or series of images composing the examination. Therefore, if the network can accommodate the images, it will be able to easily accommodate the other information transfers. Network transfer of the image information is referred to as a picture archiving and communication system (PACS). This system involves acquisition, communication, storage, and display of images. Although the PACS model appears simple, implementation is exacting. The devil is in the details as will be addressed in Part 6 of this article, "The Process of Implementation."
It is important to understand the basic principles and products relevant to building a network for the future: digital networks, optical fibers, integrated services, and internetworking. The introduction of digital systems (replacing their analog predecessors) has allowed the for the transition to digital networks. These have been shaped and enhanced by optical fibers, integrated services and internetworking. Optical fibers transmit media (images, voice, video, etc.), as they have among the highest (and continually increasing) information–carrying capacities available. Integrated services means the merging of the three existing networks (voice, data, and video) into a single network capable of carrying all these services and more. Internetworking is a technology that allows multiple different networks to operate together; it is a "network of networks" [4].
Network size can be measured in terms of the number of end points, i.e., points at which users can attach devices to the network. Network design is fairly simple until network size becomes large. Moderate–sized networks should be able to grow to many millions of end points. Networks can be interconnected to form a more complex and far–reaching network. Complex interconnected networks are likely to the be the topologies chosen for future public networks that cover large metropolitan areas. These topologies are known as WAN (wide area networks), LAN (local area networks), and DAN (desk area networks). The network needs to allow the availability for increased connections. These services need to be reproducible and secure. There needs to be uniformity between the protocols used so that future integration will also be reliable and relatively simple [4].
There are available today sundry products and services relevant to radiology information systems and PACS. The transmission substrates include, but are not limited to, the following: POTS (plain old telephone service); N–ISDN (narrowband integrated services digital network), which is very similar to POTS but is digital and standardized; Ethernet, which is a shared access LAN technology with data transmission rates of 10 Mb/s to 100 Mb/s; SONET (synchronous optical network); ATM (asynchronous transfer mode), which are broadband networks (B–ISDN) that allow voice, data, images, and full resolution video to be carried over the same network; IP (internet protocol), which is the successful demonstration of interconnecting (internetworking) heterogeneous networks. IP is tightly linked with the transmission control protocol (TCP) and is often referred to as TCP/IP.
The many dimensions of network technology and services make the design and implementation of networking and structural systems quite complex. Network requirements are related to the scale of the hospital–wide radiology operation or network operation. The requirements for hospital– and other large enterprise–scale PACS are subject to the specifics of each installation. Network traffic due to image data acquisition, transfer to storage, and quarry/retrieve operations supporting image presentation are heavily influenced by design decisions made by the system integration or system component supplier [4]. The implication of the choice of data flow model does not have significant impact on the network until one reaches or exceeds the available bandwidth and capacity of the system. Important features involved in the organization of network infrastructure involve the quantity of information to be transmitted and the rate at which it needs to be transmitted. The system must be built to allow peak data acquisition traffic to be more than one order of magnitude greater than the average acquisition rate. This implies bandwidths allowing for data flow rates greater than 40 Mb/s and up to 100 Mb/s [4]. Response times for image display and quarries must similarly be integrated into the infrastructure of the network [3,4].
Investments in a network infrastructure will be determined increasingly by hospitals, rather than radiology departments, and the investments are likely to be substantial, because they will be seen as a critical tool to reduce operating expenses. This will be fueled by consolidation of the health care industry and the creation of large health care enterprises, the components of which will often be widely distributed via an electronic infrastructure linking the components of the medical enterprise. This will reduce cost as well as improve quality [2,4].
3. Computer–Aided Diagnosis
Computed–aided diagnostic (CAD) methods that direct the radiologist's attention to suspicious regions in an attempt to overcome problems of simple oversight and to refine detection of abnormalities are in the works today [5]. Digital imaging allows the extraction of features, which may or may not be visible to a human observer. Once the radiological image is in a digital format, the data is accessible for computer manipulation. Various methods may be used to extract information from images. Artificial intelligence techniques used in image analysis include such methods as discriminate analysis techniques, rule–based expert systems, and artificial neural networks [6,7]. Such techniques have been used to merge computer–extracted features and radiologist–extracted features in diagnostic decision aides. Neural networks address problems by learning from examples that are presented repeatedly, instead of by means by pre–specified algorithms; the results, therefore, stand alone, with the path or logic "invisible."
Mammography, chest radiology, skeletal radiology, and angiography have utilized CAD methods. In mammography, for example, the primary goal of CAD research is to develop a computer method that aids the radiologists (but does not replace him or her) in order to increase diagnostic accuracy in screening mammography. CAD would serve as a second reader (like an imperfect spell checker), leaving the final decision regarding the likelihood or presence of cancer and patient management to the radiologist. Programs have been utilized which increase the detection of both microcalcifications and masses; furthermore they assist in characterizing both the microcalcifications and/or masses as probably benign or malignant [5,8,9,10].
Analogous use of CAD has been employed in chest radiology in the detection and characterization of lung nodules and interstitial lung disease. These methods of CAD have been shown to improve radiologists' performance because CAD is used as a second opinion and not alone [5,6]. It need not be perfect to be useful. It should be noted that a CAD scheme can be useful even with an overall accuracy less than that of the radiologist, because lesions detected by the computer may not coincide with those detected by the radiologist — i.e., they may be complimentary rather than redundant. In the case of mammography, it has been shown that the use of CAD achieved an average increase in sensitivity of 9.7% with no decrease in specificity [5,8,9].
Intelligent workstations can incorporate various computer versions of artificial intelligence. Although the intelligent workstation is still slightly futuristic, it will eventually become a diagnostic tool used by radiologists in the assessment and valuation of images as well as education. Once radiologists become familiar with the strengths and weaknesses of each CAD program, it will improve and optimize diagnostic performance. The appropriate role of each CAD program will be determined by each radiologist according to his or her individual training and observational skills, thereby reducing intra–observer variations and improving diagnostic performance [6,7].
The development of a useful computer–based decision aid requires appropriate design, operation and evaluation. The design of decision aids requires choosing the most appropriate technique for a particular application — case–based reasoning or neural networks. System designers must be sensitive to many issues such as the availability of necessary information, the availability of expert opinions and its nature, the required speed of the response, and whether the reasoning must be explained by the system [7].
A successful decision support system requires attention to its organizational and operational aspects. The system should be readily available and imbedded within the physician's usual clinical functions [6,7]. Validation and evaluation of decision aides are important. Validation of decision aids requires attention to appropriate measures of performance and sound testing methodologies [6,9]. Computer–based expert systems may work best in highly focused domains. Although decision support systems have great potential to improve medical care, relatively few have come into routine clinical use as of yet.
The future of decision aids in radiology relies in integration. Integration must take into account other information sources, with the clinical workflow, and with radiology education [6]. Decision support systems will be incorporated into radiology results reporting systems so that their use becomes part of the routine clinical milieu. Because we learn as we practice medicine, decision aids can play an important role in education and can derive their knowledge from prior cases and clinical records. As radiologists play a greater role in determining and justifying the use of imaging procedures, decision support systems may help insure compliance with criteria for appropriate use of diagnostic imaging procedures. Decision aids may be used to formulate guidelines about the optimal use of radiological procedures. By integrating procedure selection, diagnosis, result reporting, and education, computed–based decision aids have the potential to broaden the role and responsibilities of radiologists in patient care [7].
4. Evaluation of Radiology Systems
The future evolution of PACS and radiology information systems will depend on justifying their utility. This involves demonstrating better functionality, efficiency, and economy to the medical imaging process. PACS offers increased operational capacity. PACS can perform functions without allocating additional resources. PACS can perform the same functions using a reduced staff. The bottom line is increased patient throughput and decreased cost. The increased efficiency for the radiologists may allow them to perform more studies or allocate the time for productive patient care activities. Additionally, patients and their physicians may experience shorter waiting periods before studies can be performed. Efficient communication would allow for more expeditious decision–making and save money. Lastly, PACS virtually eliminates lost images. The cost savings seen in such a system is therefore easily seen in its potential to decrease the performance of repeat studies, eliminate the loss of time required to search for films, and the ability to compare new studies with prior examinations [11].
Considerable overhead is involved in the original capital expenditure for hardware and software related to PACS. In addition, these systems need to be maintained and updated. Space and overhead for the system is often offset by a cost of savings in the reduction of primary film library space. There would also be a reorganization of personnel, with a reduced number of personnel in the film library and increased number of workers maintaining or operating the PACS hardware and software. Economic evaluation of PACS systems have shown them to be cost saving or cost neutral. Unfortunately, the large up–front costs and disagreements about who should pay and who would benefit have delayed the widespread adoption of PACS technology [11].
