What Lies Ahead II: The Impact of Future Technology on the Practice of Radiology
What Lies Ahead II: The Impact of Future Technology on the Practice of Radiology |
|
Steven Willing, MD, MBA has reported no financial interest, arrangement or affiliation with a commercial organization that may have a direct or indirect influence in the subject matter of this presentation.
Contents
Predicting the future is a dicey business. Throughout the generations many have tried and failed, but some succeeded. The best "prognosticators" are usually those whose predictions have yet to be tested. With humility and great caution, let us take these few moments to imagine where computing technology will take the practice of radiology in the years ahead.
Most readers have heard of Moore's Law,1 named for Gordon Moore, one of the co-founders of Intel. It's not a law, of course, but merely the extrapolation of a trend. In 1965 Moore plotted the growth of semiconductor density on memory chips over time, and noted an impressively constant exponential growth rate. By simple extrapolation, Moore's "Law" anticipated that transistor density on integrated circuits would continue to double approximately every eighteen to twenty-four months. More to the point, it implied that processing power would grow exponentially over time. Remarkably, this prediction has held now for thirty-five years, and is expected to hold for at least a few more. Clearly, this cannot go on forever or the components of a microprocessor would become smaller than atoms, which is impossible.
What is less well appreciated is that while processing power continues to increase at a rapid rate, two other fundamental components of computing, storage capacity and network bandwidth, have been increasing as fast or faster.
For purposes of our discussion, we will make three assumptions regarding the direction of hardware technology. We expect the future to bring us ever nearer to:
Based on these three assumptions, we can make certain predictions of how our practices might be altered through a two-step process. First, we consider our present operations and identify how our practices are constrained by limitations of processing power, storage, and bandwidth. Then we try to imagine how our operations would be altered if those constraints were removed.
Although we concede that Moore's Law will not hold forever, that may be irrelevant. Even as we approach the absolute limits of physics, there are other ways to boost processing capacity (by using multiple processors, for instance, or wholly new forms of computing). Our ability to design useful software may be the ultimate limit. We could envision a time when processing capacity will be more than sufficient to rapidly handle the most complex image processing algorithms humans can write. At that point, automated software generation may take over. Even now, the constraints of processing power are fading from our workplace. Certainly the processing of raw CT and MRI data will become essentially instantaneous, as will 3-D reconstruction and other forms of image manipulation. Processing capacity will become an insignificant component of our operational costs; indeed we are almost there.
As processing becomes ever cheaper and more powerful, we predict it will become ever more ubiquitous. Not only will our computers grow ever smarter. Our equipment will as well. Automatic calibration, built-in diagnostic troubleshooting, and exposure dose reporting are benefits to be realized as developers seek creative ways to extend the utility of cheap microprocessing.
Unlimited processing capacity and unlimited storage will bring us instantaneous results and multitasking. Our systems will permit multiple users to view and manipulate images simultaneously. Simultaneous access will be limited only by the number of keyboards and monitors. (The latter will be inexpensive flat panel displays). Gone will be the days when the technologist cannot film an angiogram (should filming actually be needed) while acquisitions are still being made. Enhanced processing will also allow a single terminal to perform all of the operations of reading, including display and transcription. It will no longer be necessary to have rooms crowded with separate systems to handle display, transcription, and information system access.
The first hard drive2 was manufactured by IBM in 1956 with a whopping capacity of 5 megabytes. The massive device consisted of 50 rotating disks each 2 feet in diameter. The first hard drive for a personal computer, the 5-megabyte Seagate ST5063, was released in 1980. As of this writing, a consumer oriented 160-gigabyte hard drive4 can be found for under $300, a thirty two thousand-fold increase in capacity from 1980. The doubling time of 1.5 years for hard drive capacity is remarkably close to that for microprocessors. This may not continue indefinitely. Physicists believe the limit of magnetic memory is about 12 gigabits per square centimeter. Eventually the hard drive itself will be replaced, perhaps by technology even now under development5 that can achieve far higher bit densities.
On a 160 gigabyte hard drive, one can store over 18,000 digital chest x-rays or approximately 3000 chest CT's. We are near the point where an inexpensive consumer hard drive could store the monthly volume of an average-sized, fully digital radiology department.
In what ways are we constrained by storage capacity? Planning archival for a digital department is a complicated and expensive process. The massive storage required has necessitated solutions such as large tape systems or optical jukeboxes. These tend to be slow and expensive. The desire to maximize storage capacity has necessitated various forms of image compression, some of which result in real information loss.
With no constraints on storage, we can envision a not too distant future where a year's worth of images for a large tertiary center could be stored on any desktop computer. The expense of storage for a fully digital department may shrink to virtually nothing, while vendors will derive their income from providing software and services rather than hardware. The archive could be duplicated many times over, so if one fails another comes right online. Or perhaps there will be multiple "mirrored" archives just to minimize response times. Local storage will follow suit. Patients could plug their PDA into a port, download their images, and take them across town for a second opinion or home to show their family.
To get the data from storage to processor, it first must be read off the media and then transferred to the system board (motherboard) or network. Both are bottlenecks. The fastest consumer desktop hard drives6 now achieve sustained data transfer rates approaching 50 MB/sec, sufficient to retrieve most digital imaging studies in less than one second. In a radiology enterprise, a mass storage solution is required based on tape, optical disk, or hard drive arrays. Commercial disk storage systems using multiple hard drives in an array are the fastest possible storage solution, boasting data transfer rates of up to 2 gigabytes per second7. Tape storage systems are nearly glacial by comparison: the best data transfer rate obtainable with tape systems is 15 megabytes8 per second, and it may take up to two minutes to load and search a tape. The most current optical disk jukeboxes transfer data at 6.1 megabytes9 per second, and the time required to load optical disks from a jukebox may range from as little as 5.5 seconds to 90 seconds or more, eclipsing actual read times.
Examinations can be retrieved faster from a hard drive than any other storage option other than computer memory itself. The practice of "pre-fetching" old comparison studies in a PACS system has been necessitated by the otherwise prohibitive time it takes to retrieve comparison studies from an optical or tape jukebox. With unlimited cheap storage, old studies will be retrieved almost instantaneously at the time of interpretation, and the patient's entire digital "jacket" will be there, not just one or two previous examinations. There will be no need to "pre-fetch" anything. It will (finally) be better than film, because we will no longer waste time searching for one particular page out of a thick, heavy film jacket.
Compression technologies, although useful in their day, will be rendered unnecessary as storage and bandwidth constraints disappear. There would be no reason not to use full uncompressed DICOM images within the working environment. The use of compression has always involved a series of tradeoffs. Our lives and our software applications will be simplified when we no longer have to worry about it.
Bandwidth is the size of the "pipe" that transmits data from one point to another. The large bandwidth in modern networked medical centers enables us to send images freely from scanner to archive and archive to viewing stations. A lack of bandwidth is what keeps us from reading exams at home or our remote interpretation facility in Honolulu. Of course we can do all that now, for a price. The price is paid in image compression, transfer times, monetary costs, and geographical constraints.
Today we have the technological capability to build optical fiber networks with data transfer rates of 1.6 terabits per second per fiber10. Over such a connection it would take only a few seconds to transfer a year's worth of images from a large medical center. At these speeds, the time limiting factor will be processing and retrieving the data from storage. The transmission capacity of optical fiber has doubled every 16 months since 1975.
The last link of the network connecting to the desktop has grown almost as rapidly. For computers on a local area network, as in most hospitals, the standard until recently was 10 Mbps Ethernet. In the space of a few years we have seen an exponential rise in the speed of Ethernet technology, progressing from 10 to 100 to 1000 Mbps. Gigabit Ethernet11 (1000 Mbps) has now become practical for local and wide are networks, and 10 Gigabit Ethernet is seriously regarded12 as the next stage in evolution of the wide area network.
Home connectivity has been slower to develop. My first modem operated at 2.4 kilobits per second. For today's home consumer, the inexpensive options of DSL or cable modem13 can provide transfer rates of around 1.5 megabits per second, comparable to the old "gold standard", a T1 line costing $1000 per month or more. Unfortunately, geographic, economic, and technical barriers still render these choices unavailable to large segments of the North American population. Looking forward, however, we see the availability of cheap, fast access expanding rapidly.
With unlimited cheap bandwidth, we will be able to effortlessly transmit transfer full uncompressed DICOM studies to anywhere, anyplace, anytime, rapidly and at insignificant cost, along with whatever software is needed to view and manipulate them. At that point we would truly be untethered from the workplace. Wherever we find an Internet connection, we could turn on any laptop, unfold our portable display, plug a microphone into the sound jack, and return transcribed reports to the clinician. The images will be downloaded in a work list with all the relevant clinical information. Any clinical information we lack will be easily obtained by tapping into the hospital information system. The software for voice recognition could be downloaded over the Internet connection, and the portable would process the voice recording into text. We would check and approve the report, and upload a final report back to the site.
It is not only the receiving end of the pipe that can move, however. Unlimited bandwidth means the serving end is also free to move, downtown or across the country. Application service providers (ASP's) providing remote PACS service already exist. This nascent business could eventually lead to outsourcing of the entire PACS infrastructure from hospital to third party. Like outsourcing of any other service, not all will choose that route. However, once the technical obstacles are overcome PACS ASP's could potentially save money by consolidating fixed personnel costs for service, maintenance, and security. PACS will be available to any hospital or clinic on a subscription basis, sparing them a large capital outlay.
We are under increasing pressure to provide round the clock interpretations for hospitals and emergency rooms. Providing such a service comes at a cost to our health and family lives. The United States spans six time zones including Alaska and Hawaii. Were we to establish a network of cross-coverage across time zones, we could extend our coverage by five hours without any change in our work schedules, six if we include eastern Canada. Unlimited bandwidth can make that possible. By adding readers on other continents, round-the-clock coverage is possible with all participants working a normal daytime schedule. In fact, this is already being done14 on a limited scale.
Wireless data transmission is a burgeoning area of network development, but for the practice of radiology we would anticipate that in most instances a physical connection will be necessary. Wireless data transmission faces a fixed constraint in the fact that the radiofrequency spectrum is crowded and incapable of growth. Although certainly capacity will increase as we learn to make better use of the RF spectrum, we cannot assume the inevitability of unlimited RF bandwidth for data transmission.
Our emphasis has been on hardware development, believing it is more inevitable. If we extend our consideration to software, the list of possibilities expands mightily. The probability for some of these occurring in our lifetimes, however, is far less certain.
In the worst-case [to us] scenario, we wonder if radiologists could be "replaced" by computers. If we exclude performance of procedures, this would in essence require a system of artificial intelligence using pattern recognition to recognize abnormalities and generate reports from a live database. The digital image need never be displayed, and "image" would be no more than a metaphor for the dataset. Radiologists could still have a role as researchers responsible populating the database as new discoveries unfold. Although such scenario is possible, we believe that few if any are likely to see it in their professional lifespan. Optical or pattern recognition remains quite primitive, and progress has been slow. It would take years of programming from multiple radiologic subspecialists to develop the artificial intelligence, with little incentive at present to do so. It is more likely that some forms of image processing will be used to aid the human reader, such as automated processing of attenuation coefficients to detect microcalcifications in mammograms.
A more attainable objective is some form of live decision support at the viewing console, perhaps tied directly into continuing medical education. Most radiologists would welcome an easy and efficient way to research a case that did not require a library of books or a cumbersome literature search. Live human consultation is already a reality through the various teleradiology services that offer subspecialty backup.
Barriers to our technological vision arise from many fronts. Legal and regulatory obstacles (e.g. licensure, hospital credentialing) stand in the way of unfettered teleradiology. Conversely, we will increasingly rely on technology to help us meet present and future regulatory burdens imposed by government and payers.
Seamless integration of workflow will require acceptance of universal standards, such as those under development under the Integrating Healthcare Enterprises (IHE) initiative sponsored by RSNA and HIMSS. The goal of this initiative is to establish standards for free communication between radiology information systems, PACS systems, and imaging equipment with reliable integration of patient data.
The social concerns voiced by many are genuine causes for apprehension. Practicing radiology without a physical presence cannot but hinder the cultivation of relationships with clinicians and patients. Radiologists who are not physically present cannot serve on hospital committees, forfeiting their voice in credentialing disputes and other critical matters. [Although all such meetings could be replaced with electronic conferencing, this is not likely to happen soon].
Opportunities and constraints arising from fundamental economics cannot be ignored. We are in a service profession. There is a limit to how many exams we can read in a day, creating an upper limit on the productivity we can achieve. The good news is that by streamlining the process most of us have some distance to go before we hit that limit. Increasing our productivity will permit us to maintain or increase income even in the presence of declining reimbursement.
We must work to ensure that systems are designed to serve us, rather than vice versa. The radiologist's primary interaction will be with the viewing terminal. The software should be easy to customize so that when the radiologist logs on all of his viewing preferences are implemented automatically. Routine functions like windowing, subtraction, or re-masking should be automated. Technical parameters germane to the report could be automatically extracted from the DICOM header and converted to coherent text.
The benefits of unlimited processing power will only be fully realized if software development proceeds at a similar pace. Two areas where rapid progress is occurring are voice recognition and natural language processing. Eventually we will see universal deployment of voice recognition for reporting as the accuracy continues to improve. We must pressure the developers to ensure the transcription systems are simple and user friendly. Coupled with natural language processing, improved voice recognition has the potential to make our systems more user friendly and interactive. We will be able to use natural free speech to operate our systems and request instructions and assistance.
The traditional viewing station consisting of four 14"x17" light boxes in one or two rows has served us well. It has been efficient and could display (through traditional film) resolutions exceeding anything yet available electronically. The cathode ray tubes (CRT's) used on current viewing systems have significant deficiencies15. We would expect future viewing systems to use banks of flat panel monitors, which could be arranged to closely replicate the format of a traditional light box.
Workflow management software will bring up a user-specific queue of studies ready for interpretation, monitor the dictation process and recognize when studies are completed. It will automatically display the most recent comparable examinations, and provide on request a list of all other examinations on that patient. The system will be integrated into the hospital information system so that additional clinical information is easily accessible.
View boxes rarely crash! The significance of this is not to be taken lightly. The radiology department of the future can ill afford to experience interruptions in service. Cost savings achieved in some areas (e.g. film library personnel, transcriptionists) will be offset by the expense of new personnel to provide hardware and software support, training, and network security. Backup systems, now a rarity, will hopefully become more common as hardware costs plummet.
Extracting useful information from gigabytes of free text is a sort of Holy Grail for health care information specialists. All the information that could be used for quality control, outcomes analysis, and regulatory compliance resides in millions of free text documents such as discharge summaries, clinic notes, and radiology reports.
One solution proposes the abolition of free text by the compulsory use of structured reporting. The task of creating structure for the myriad of possible radiology reports is daunting, but not impossible. Unfortunately, this solution requires forced behavior modification likely to be resented by many, and orphans the decades of data already acquired.
Others are aggressively pursuing an alternative solution, through the development of natural language processing16 - software that can extract coherent data from free text. Progress so far is encouraging. Although there is still a very long way to go, it is not unrealistic to believe we will see significant achievements in this area in years to come.
The "dark side" of technology
An essay by Bill Joy in the April, 2000 issue of Wired generated quite a stir in the technological community. Joy is Chief Scientist and a co-founder of Sun Microsystems whose technical achievements have earned him international fame and respect. His essay entitled "Why the Future Doesn't Need Us"17 considered the pace of development in genetics, nanotechnology, and computing/robotics and expressed concern that unrestrained development of these technologies poses a threat to the continued existence of our species. There is not sufficient space here to discuss these themes in detail; the reader is encouraged to read the original article as well as responses by Ray Kurzweil18, Michael Dertouzos19, and others. For our immediate discussion what is striking is the extent to which all are agreed. It is now widely predicted that within 30 years "desktop" processing technology will surpass human intelligence, and continue to increase rapidly beyond that point. Joy wrote in part:
The 21st-century technologies - genetics, nanotechnology, and robotics (GNR) - are so powerful that they can spawn whole new classes of accidents and abuses. Most dangerously, for the first time, these accidents and abuses are widely within the reach of individuals or small groups. They will not require large facilities or rare raw materials. Knowledge alone will enable the use of them.
Thus we have the possibility not just of weapons of mass destruction but of knowledge-enabled mass destruction (KMD), this destructiveness hugely amplified by the power of self-replication.
I think it is no exaggeration to say we are on the cusp of the further perfection of extreme evil, an evil whose possibility spreads well beyond that which weapons of mass destruction bequeathed to the nation-states, on to a surprising and terrible empowerment of extreme individuals.
The prescience of that final sentence is chilling. Pundits from all corners of the globe have weighed in on the purported implications and lessons of last year's terrorist attack on the United States; to raise it at this juncture risks sounding hackneyed. Nonetheless, both the airline attack and subsequent anthrax scare surely validate what Joy and others were saying long before September 11, 2001: technology amplifies not only our ability to do good, but the power and reach of those with malicious intent, or even the merely careless. It might take a single knowledgeable person a few keystrokes to destroy a PACS archive; destroying a film library is much more difficult. Consider for a moment the major events in radiology last year as reported by the mass media: reports of excessive radiation to children undergoing CT scanning, and a child killed while undergoing a routine MRI. The "law of unintended consequences" is more real than we would like to admit. If we believe all these dangers to be real, then we must take seriously our roles of promoting patient safety, safeguarding data security within the enterprise, and providing strict ethical oversight of biomedical research (and researchers).
We would conclude this discussion by emphasizing that not everything that can be done either should be done or will be done. It is not inevitable that progress in technology will make us more productive or efficient, improve patient care, or make our lives any easier. As we plan for the future, it is incumbent on us to define our goals and seek out means by which technology can advance them. As clinicians and teachers we share a set of common objectives. We all desire to increase our effectiveness as teachers and clinicians. It is to our advantage to increase our productivity as well in both patient care and teaching.
Technology can improve our teaching effectiveness and productivity if we spend the effort to learn what teaching methods are most successful and seek creative solutions. A single lecture can reach hundreds of residents instead of a handful. One on one teaching at the view box is popular with residents, who appreciate the attention, but inefficient, costly, and of unproven value. Interactive software might perform a similar function with a marginal cost approaching zero.
Technology can enhance our clinical effectiveness with real time decision support and readily accessible CME. It can facilitate communication with referring physicians and patients through a variety of methods.
The aging of baby boomers presents a looming demographic crisis in financing their retirement and health care. If reimbursements continue to decline, which is probable though not inevitable, the only way to maintain or increase our income will be through increased productivity. No one is exempt from the law of economics dictating that real personal income growth can only accrue through productivity enhancement. Otherwise personal income is a zero-sum game.
Even if economics were of no concern, we face increasing per capita workload caused by the triple threats of increased utilization of imaging, aging of the population, and retirement among our ranks. Increasing our productivity may be a necessity, or we cede more and more turf to non-specialists. Technology can improve our productivity by reducing the countless inefficiencies that occupy many radiologists, such as hanging films, rifling through thick film jackets, digging up clinical information, and giving telephone and preliminary reports.
We are not passive participants of our own future. We should use our ability to shape it in ways that benefit our patients, our students, and ourselves. Technology is not our destiny, but a means of getting there. While the potential is great, user-friendly systems will not happen by themselves. Poorly designed technology can diminish our productivity - we've all seen it. We must make our needs known and pressure the vendors at every opportunity. We have an opportunity to use our experience and creativity to make the future practice of radiology better for all.
Copyright © 2004 American Society of Neuroradiology, www.asnr.org