Medical imaging advances have occurred rapidly over recent decades.  These advances typically increase the resolution, image count, and complexity of imaging studies.  Examples include 2D mammography to 3D mammography/tomosynthesis, planar nuclear medicine to SPECT/CT, radiographs to CT, and PET to PET/CT and PET/MRI. Each technological iteration assumes to improve imaging diagnosis and patient care. However, imaging advances frequently also extract additional time and effort from the medical imaging workforce to implement, perform, and/or interpret the new advanced exams compared to prior generation technologies.

In ideal circumstances the sum capacity of the medical imaging workforce will match the labor demands of existing and emerging technologies.  If new technological demands exceed the sum capacity of the radiology workforce a problematic shortage of expert labor ensues. Indeed, it is possible that medical imaging’s most valuable, tenuous, and scarce resource are the human workers who make medical imaging possible. This workforce comprises not only radiologists, interventional radiologists, and nuclear medicine physicians, but also residents and fellows, imaging technologists, other clinical support staff, administrative staff, imaging physicists, information technology staff, schedulers, coders, and numerous additional important individuals. While Moore’s law predicts an accelerating rate of computational advancement, there is, unfortunately, no Moore’s law for the expansion of human capacity. Imbalance between the labor demands of medical imaging technologies and the sum capacity of the medical imaging workforce risks the well-being of medical imaging—and therefore patient care.  

For reasons that can be debated, policy-makers have enacted decreases in reimbursement for many imaging exams for more than a decade.  As a result, the medical imaging workforce is compelled to complete medical imaging tasks, including image interpretation, at a faster rate to accomodate high patient volumes, preserve salaries, meet practice expenses, and maintain staff.  Demand for many medical imaging exams continues to increase, propelled in part by an increased need for medical services from the aging Baby Boomer Generation and an increasing reliance on medical imaging to confirm, monitor, or exclude disease.

If individuals in any profession are required to work faster than efficiency gains allow, a reduction in quality is inevitable, and professional burnout is likely. A system is toxic if it requires work to cycle faster than maintenance of quality permits, while concurrently penalizing workers if quality wanes. In terms of healthcare, the stakes are high for both patients and providers. Indeed, physicians and other clinical practitioners throughout healthcare in the United States frequently encounter workplace dissonance wherein the daily work required by the system approaches or exceeds what is possible to maintain quality and, in healthcare, a lower quality of care cannot be tolerated. In addition to potential harms to patient, this dilemma raises substantial risk of moral injury to the providers, which has been defined as “powerlessness to optimally perform due to lack of support/resources that contradicts one's professional mission”.

As members of the medical imaging profession, we must ask ourselves this imperative question: are we maintaining balance, in the interest of patient care, between the increasing complexity of medical imaging and the available workforce to support this vital work?

To be honest, I don’t know. I believe the current job market demonstrates a shortage of radiologists, especially in certain subspecialties. I have heard from senior radiologists that we are, as a profession, completing increasingly complex interpretive tasks at a faster rate. I do believe, based on my own experience, that this complexity is accelerating and the volume of complex imaging studies is increasing.

Let me present one example to illustrate this point. As a first year radiology resident, PET imaging in clinical practice essentially equated to performing a PET scan with the radiopharmaceutical FDG.  Therefore, to be proficient with clinical PET imaging, radiologists and nuclear medicine physicians needed only to master the uptake and physiologic distribution of FDG on an imaging study. However, by the end of my residency and fellowship training, clinical PET imaging also included additional radiopharmaceuticals such as sodium fluoride, choline, fluciclovine, various forms of somatostatin receptor imaging, and various forms of amyloid imaging, each with unique imaging characteristics. While clinical use of each radiopharmaceutical represented significant clinical advances, these also greatly increased the complexity of being proficient at clinical PET imaging. We have since added more PET radiopharmaceuticals to include various forms of PSMA, each with slightly different uptake patterns, as well as fluoroestradiol. A multitude of additional PET imaging agents are on the horizon of anticipated clinical use.

This renaissance of clinical PET imaging tracers is amazing and will likely provide immense clinical benefit to appropriately selected patients. Coexisting technological advances in PET scanner design now allow more detailed scans to be obtained in a more rapid manner than what I experienced as a resident—thereby allowing a single PET scanner to image many additional patients in each workday compared to scanners from only 5-10 years prior. However, an increase in the number of nuclear medicine or nuclear radiologist physicians to match these advances has largely not occurred. 

There is a connected renaissance in nuclear medicine therapeutics, often termed theranostics. Strong demand already exists and is anticipated for many nuclear medicine therapeutic agents that include PSMA-directed radiotherapies with other emerging therapeutic agents on the horizon for a variety of malignancies. According to current trends, entire treatment clinics will be necessary for nuclear medicine practices throughout the U.S. to meet the demand for these therapies. Substantial professional time from nuclear-trained physicians may need to be allocated to therapies alone, adding to the increasing time- and cognitive-demand of clinical nuclear medicine practice. By adding significantly increased volume and complexity to the clinical practice of nuclear medicine, these remarkable imaging and therapeutic advances risk overburdening the current sum capacity of the nuclear medicine/nuclear radiology workforce.

While current advances in nuclear medicine are, in my opinion, some of the most dramatic in all of medical imaging (and perhaps medicine), similar stories of technical advancement ushering in complexity that is unmatched by an expanded workforce likely exists within all medical imaging subspecialties (and perhaps all medical fields).  It has been said that it is easier to build new hospitals than to staff them.  The same may be true for medical imaging technologies. It may now be easier to develop complex imaging technologies than to assemble the human workforce necessary to utilize those technologies.

important Questions

If the bottleneck in medical imaging advancement is not technological innovation but rather a limited capacity of the human medical imaging workforce, can we replace human effort with computer diagnostics and be better off in the end?

Will artificial intelligence (AI) and machine learning allow medical imaging interpretation to advance to superhuman levels of efficiency while simultaneously maintaining or improving human interpretive performance? 

Will AI and machine learning offset the increasing technological/human workforce mismatch?

I do not believe that anybody truly knows the answers to these questions. One can speculate, but not predict the future.  And that is the leading problem. If our current trajectory demands a future wherein computers must alleviate the clinical pressures increasingly placed on the medical imaging workforce, what happens if our software saviors fail to deliver? 

When I imagine future decades of medical imaging, I anticipate much that is good. I also have two principal fears. The first is that the demands of future imaging technologies—in terms of volume, interpretive complexity, and regulatory burdens—will advance beyond human capabilities and/or the sum capacity of the medical imaging workforce. We are already approaching the limits of human cognitive ability in terms of PET interpretation and other advanced imaging studies when one considers the speed at which the interpretation must occur to match clinical necessity.  In my opinion we, at least in the United States and many other developed countries, are potentially already at or exceeding the maximal workload that can be placed upon the existing radiology workforce.

While many in radiology fear the day when human radiologists such as myself will be replaced by computers, my second and overall greatest fear is that the help we will receive from AI and machine learning will not be enough. What if artificial intelligence and machine learning offer too little assistance during my own career to augment human interpretive capacity and speed to match the volume and interpretive complexity of future imaging technologies? I fear the lack of needed support from technology more than obsolescence of human radiologists that may result from future technological advances.

The crisis on the horizon may be one of a severe shortage of humans trained and available to complete the highly complex interpretive tasks of tomorrow’s medical imaging. I foresee the need for a technological lifeline. What if AI interpretive algorithms do not answer our call?

I might be wrong. Indeed, it is a nefarious task to speculate on the future. But anticipating the future is necessary as the policies of today shape the reality of tomorrow.  If today’s policies rely on the assumption that computers will soon augment human radiologists, substantial risk to the health care system exists if this doesn’t occur.  Until the machines prove themselves worthy of the task, we must prepare for all possibilities, including the potential that humans will continue to be tasked with the heavy lifting in medical imaging. In the movie The Matrix, it is the machines that require human energy to function, not the reverse. In some parallel ways, the complexity of tomorrow’s artificial intelligence and machine learning that may revolutionize medical imaging could require substantial human effort, including that of highly trained radiologists, to implement and sustain.

Potential Solutions

If a mismatch between technological complexity and human capacity is looming, solutions to prevent widespread burnout and maintain quality in medical imaging likely need to incorporate some or all of the following:

1. Ensure an adequately-sized training pipeline of radiologists, interventional radiologists, and nuclear medicine physicians to meet future clinical needs in terms of complexity and volume.

2. Ensure that payments for imaging services match that needed to maintain viable imaging practices. Medical imaging cannot face perpetual reimbursements cuts and still assemble the diverse and robust workforce that will be necessary to meet future demand.

3. Prioritize technology that is easier to interpret, or that allows imaging diagnoses to be made by reviewing fewer images. These technologies are difficult to find as the rule-of-thumb for new technologies seems to be more images, more sequences, and more complexity than prior generation imaging studies. However, one example of an existing technology that potentially simplifies imaging interpretation is contrast-enhanced mammography (CEM). In only 8 images per patient, given high signal-to-noise and removal of dense breast tissue via dual-energy recombined imaging, powerful diagnostic information can be obtained on an exam that is relatively easy to interpret. In our current environment of a shortage of breast imagers, wider adoption of CEM could offer a helping hand in terms of providing high diagnostic yield and shorter imaging interpretation times compared to technologies like automated whole breast ultrasound or contrast-enhanced breast MRI, for appropriately selected patients. Unless more radiologists (or AI interpreters) become available to address current and future demands of medical imaging, we need technologies in all imaging subpsecialties, perhaps augmented by AI, that simplify imaging interpretation by providing high signal-to-noise exams requiring fewer images to interpret.

4. Trim unnecessary regulation, administrative burden, and other non-clinical tasks without compromising, in any manner, patient safety or outcomes. I believe this is possible. As ironic as it sounds, large hospital systems and government entities associated with healthcare should probably have something akin to “efficiency committees” whose task it is to find and remove policies that increase complexity within the healthcare system without providing benefit to patients or healthcare providers.

5. Aggressively develop AI and IT systems that offer a true helping hand to the human workforce. I have seen some progress on this front that is encouraging. However, more progress needs to be made. I dream of the day when I sit down at a radiology workstation and I no longer need to login to a variety of systems with various rotating passwords and phone prompts. I dream of the day when all prior imaging and imaging reports for every patient I see are readily available, regardless of whether imaging was performed at my own institution or elsewhere. I dream of Al that offers even simple helps, such as catching left-right (wrong laterality) dictation errors in reports. After an imaging report is drafted by a radiologist, I dream of an AI system that will automatically cross-check what was reported with actionable findings it has detected on imaging review, and alert the radiologist of any potential discrepancies to review before the report is finalized. I also dream of an IT system that allows radiologists to communicate actionable findings by seamlessly connecting me with the referring provider or patient. Most of all, I dream of AI systems that will improve my diagnostic accuracy and efficiency, and IT systems that simplify work for radiologists.

I am excited to see what new technological innovations will occur during my career as a radiologist.  I anticipate that I, and other radiologists of my generation, will stand in awe of the remarkable technological advances we saw during our careers.  We may also marvel at the years when we interpreted images without the aid of computer interpretive aids and with rudimentary IT systems.. I hope we will say something akin to the following: “Thank goodness that AI and machine learning emerged when it did.  Can you imagine doing the radiology work of today without it?”.

Note that this article contains opinions from the author about general trends in radiology and nuclear medicine in the United States. What is discussed herein is not specific to any individual institution, vendor, or healthcare facility.