Author: Logan Young

_{Published August 9, 2021}

From 1950 to 1967, CBS ran an American game show called What’s My Line? As part of the show, celebrity contestants tried to guess a person’s occupation for a $50 prize that was often donated to charity. In the spirit of learning from each other as radiologists and continuing to explore how our services can best serve our patients, colleagues, and other stakeholders, I introduce to you the medical edition of What’s My Line?

Today, you will be asked to guess my medical subspecialty. (Hint: I’m not referring to abdominal radiology; it’s even more specific than that!)

Here are your clues:

• This field of medicine acquires and manages medical imaging information. That’s our primary business. We acquire data, optimize it into viewable images, retrieve them, and then interpret what they reveal. Based on the information we receive, we generate reports, some of which contain recommendations, and we communicate the results. Some results require urgent notification or follow-up measures to ensure all safety loops are closed.

• We examine high-resolution images to pinpoint abnormalities, propose diagnoses, stage diseases, characterize the effects of therapies or underlying conditions, and guide follow-ups.  We also enjoy and thrive at developing diagnostic differentials; we observe the abnormality and derive appropriate and rational shortlists of the most likely etiology for the findings we observe.

• We are dedicated to providing top-level, timely service to our many stakeholders, including patients and their families, our referring providers, administrators, and payors. As part of this commitment, we are exploring informatics solutions to improve our diagnostic sensitivity and specificity.

• Ensuring staff and patient safety is paramount. We are proud of the quality and safety initiatives that our field has introduced.
• Our field is rapidly transforming from primarily gross anatomic interpretations to today’s cellular and even molecular imaging applications. We now use site or physiological targeting agents to improve our diagnostic sensitivity and specificity. We are excited about the emergence of functional and metabolic imaging applications.

• As you might expect, the costs of our imaging equipment and storage continue to escalate. We are constantly seeking solutions for maintaining and upgrading our fleet of imaging equipment. Not surprisingly, the costs of image management software and its associated licensing are escalating. Our cybersecurity risk management efforts are also ever-vigilant to safeguard all patient health information.

• We are vitally dependent on our superb technical staff on many levels, not only in the pre-imaging stages, but also in helping us to acquire, label, store, and retrieve digital images. We seek to continuously improve our clinical performance through peer feedback and learning, and often rely on the opinions of colleagues to broaden our differentials and confirm our impressions.

• We devote many resources to upholding regulatory requirements. We must meet all HIPAA requirements, be compliant with the National Patient Safety Goals, and ensure that all of our physicians are fully credentialed, licensed to practice, and achieving required CME credits.

• We’re excited and optimistic about the opportunities that artificial intelligence (AI) and machine learning will bring to our field. On one hand, AI will become an essential diagnostic tool to manage the growing demand and sheer volume of images and sequences. On the other, informatics is rapidly becoming a necessary tool for managing and optimizing our operations and ability to positively impact patient outcomes. Our expanding digital image storage needs are resulting in new discussions about data ownership, costs, security, access, and responsibilities.

• We use operational metrics—capturing report content and turnaround time, standardized recommendations, and diagnostic accuracy—to manage and continuously improve our clinical workflows. We strive to add value through our wide array of clinical contributions and are sensitive to stakeholder feedback and experience, which we also actively measure.

Any more clues would simply give it away! What is my line? You’re quite correct if you guessed anatomic pathologist.

The point here really is to show the many and ever-increasing overlaps between pathology and diagnostic radiology. These two subspecialties are rapidly converging from an operational perspective, based in large part on the increasing complexity of effectively and safely managing medical hordes of imaging information and our escalating reliance on sophisticated machines that can learn and facilitate our tasks. At the end of the day, there are so many operational parallels for us to examine and glean best practices for the benefit of those in our care.

_{Updated October 25, 2021}

Practical pediatric imaging education, especially virtually, benefits most from a combination of didactic lectures and interactive case-based reviews that are relevant for a broad audience: dedicated pediatric radiologists, general radiologists who interpret pediatric imaging examinations, and radiology trainees—residents and fellows.

On September 8 and 9, the ARRS Virtual Symposium, Practical Pediatric Imaging, provided a head-to-toe review of both standard and novel, state-of-the-art methods, delivered by an esteemed and diverse faculty. If you were not able to attend this live symposium, the following offers a perfect primer.

During the neonatal session, Judy H. Squires of UPMC Children’s Hospital in Pittsburgh addressed neonatal head ultrasound, including the application of newer imaging methods, such as ultrasound elastography and contrast-enhanced ultrasound (Fig. 1).

• 

Recently, AJR published Squires’ “Contrast-Enhanced Ultrasound in Children: Implementation and Key Diagnostic Applications” manuscript. According to her review, “contrast-enhanced ultrasound utilization is rapidly expanding, particularly in children, for whom there is a growing range of FDA-approved and off-label diagnostic indications throughout the body.” Noting that ultrasound contrast agents lack renal excretion and can be administered to neonates, Squires concluded that the research thus far points to an “excellent pediatric safety profile of ultrasound contrast agents.” Squires’ facility with contrast-enhanced ultrasound implementation, as well as her knowledge of the supporting evidence in children, proved to be prized assets. Likewise, ARRS Symposia course codirector Rama S. Ayyala’s update on evaluating the vomiting infant was a boon for the curriculum, as was the comprehensive review of neonatal chest disorders from Peter J. Strouse of C. S. Mott Children’s Hospital in Ann Arbor, MI.

After Nancy A. Chauvin of Penn State Health Children’s Hospital in Hershey expertly examined pediatric arthritis, bone tumors, and bone marrow disorders, Strouse, senior author of the recent AJR Expert Panel Narrative Review on “Debunking Fringe Beliefs in Child Abuse Imaging,” explained how misattributing injuries leaves children at risk for future abuse (Fig. 2).

• 

“Careful review of the scientific evidence and professional society consensus statements is important in differentiating findings attributable to child abuse from fringe beliefs used to discount the possibility that a child’s constellation of injuries is consistent with abuse,” Strouse maintained. He then listed the three categories that these fringe beliefs most often fall into:

1. legitimate alternative diagnoses that should be considered;
2. real disorders with actual findings that do not mimic child abuse;
3. fabricated pathologies.

Meanwhile, 2018 ARRS Scholar Rupa Radhakrishnan of the Riley Hospital for Children at Indiana University Health in Indianapolis led an enlightening session on contemporary imaging and pediatric stroke, later helping ARRS Symposia course codirector Jonathan R. Dillman and Andrew T. Trout, also of Cincinnati Children’s Hospital Medical Center, to steer the interactive case-based review session. Additional neuroimaging lectures were given by Children’s Healthcare of Atlanta researcher Neil Lall, who skillfully detailed congenital brain malformations and pediatric brain tumors.

Following the thorough analyses of pediatric renal and liver tumors by Ayyala and Squires, respectively, Trout shared his multimodality assessment of adrenal tumors—staging, treatment response, the role of nuclear medicine—with Dillman joining to discuss peritoneal, mesenteric, and omental disorders, as well as congestive hepatopathy and Fontan-associated liver disease. (Fig. 3). For more Trout and Dillman research, turn to page 16 for a summary of their study on reduced-dose CT for lung nodules in children and young adults with cancer.

• 

Trout was the primary author, alongside Ayyala and Squires, on yet another AJR Expert Panel Narrative Review, “Current State of Imaging of Pediatric Pancreatitis”. Imaging timing, secretin-enhanced MRCP, urgent MRI, severity prediction, autoimmune pancreatitis, serial imaging best practices—these are all distinct concerns for imaging pediatric acute, acute recurrent, and chronic pancreatitis. Trout and colleagues’ appraisal concluded with a 12-point consensus statement and structured reporting template; we are still studying both. Apropos, AJR editor in chief Andrew B. Rosenkrantz handpicked this Expert Panel Narrative Review to lead ARRS’ newest newsletter for in-training radiologists, The Resident Roentgen File.

_{Published July 28, 2021}

During this year’s virtual and highly successful American Roentgen Ray Society meeting, it became apparent that we are living in a time of accelerated development and deployment of existing and emerging digital technologies. Individuals and teams are using innovative solutions to care for patients, teach trainees, collaborate with colleagues, and connect within an expanding digital universe.

I for one never imagined that my weekly mobile COVID-19 prediction report would include hourly population densities in nearby airports, supermarkets, restaurants, and bars. With geographically traceable devices, what data could possibly be next?

In the same way that NASA’s Apollo program sparked the development of new technologies (many of which were largely realized and appreciated years later) that landed the first humans on the moon, we are witnessing a fundamental transformation in health care operations that will be captured in future history books. Few could have predicted, for example, that CT scans would become an indispensable screening, diagnostic, staging, and management tool during a global pandemic. Providers have harnessed such a wide swath of tools—from laptops, mobile and wearable devices, and video conferencing to artificial intelligence, thermal sensors, and robots—to better serve patients and their loved ones, sustain remote reading and teaching environments, and uphold compliance and safety protocol. We now achieve efficiencies through rapid scanning, recruit new faculty through social media, teach our trainees in cloud-based classrooms, and attend national conferences with just a click—all without ever boarding a plane or even crossing clinical campuses.

The Future Is Now

The evidence shows that embracing digital technologies results in improved patient outcomes, cost savings and efficiency, increased productivity, heightened compliance and safety, transformed teaching methods, stakeholder satisfaction through digital connections, sustained remote teams, and accessible employee communications and wellness initiatives.

Previously, such innovation resided primarily within the hospital and physician domains, with the gradual integration of patients as they began accessing their personal electronic health records. Now, our digital stakeholders include not only patients, but referring providers, remote teams, educators and learners, researchers, public health authorities, policymakers, schedulers, transporters, the public, commuters, and travelers.

And as the digital stakeholder pool expands, so does its impact: Such technologies now routinely support telehealth, data analysis, access, scheduling, and follow-ups, management decision-making, bidirectional communication, safety compliance and practices, PACS enhancements, teaching and readouts, patient monitoring, diagnostics, consulting, screening, training, forecasting, reporting, and, of course, socializing.

Examining Digital Disparities

We must remember that our digital environment is far from globally universal. At-risk, vulnerable, underserved, and marginalized populations, such as those living more than 7,600 miles away in India today, are grappling to secure simple access and connect effectively with providers and health care delivery services through traditional means, let alone digital ones. They desperately need hospital beds, oxygen and plasma, life-saving vaccine doses, and medical workers. Resources that hospitals, such as ours, are so fortunate to have readily on hand. However challenging these issues are to address, such disparities in access, care, and connections must be studied and included in the many national efforts aimed at eliminating them. What a terrific opportunity for us to make a meaningful difference that matters.

To a large extent, this digital divide is driven by equality, equity, and justice, or the lack thereof. With equality, we assume that here in Massachusetts, for example, all of our patients benefit from the same supports. All are treated equally, irrespective of any differences. But this isn’t necessarily true yet. Having a laptop certainly doesn’t mean a patient can easily access and understand one’s medical records. Additionally, not all laptops have video cameras, and not all hardware supports the ability to participate in video conferencing or telehealth solutions. And then there are those patients who don’t have access to a laptop to begin with. Where does that leave them? It is our responsibility to find out.

From the perspective of equity, everybody receives the specific and different supports they need and, therefore, receive equitable treatment. This is closely tied to justice (some view this as liberation); our underserved patients receive access to appropriate care without requiring specific accommodations because the fundamental causes of inequities have been addressed. In other words, the preexisting systemic barriers have been effectively identified and removed. Consider the impacts and barriers that may exist due to language, poverty, mobility, cognition, geography, access to water, electricity, food, transport, comorbidities, and employment status. By working to eliminate or flatten these barriers, care becomes more equitable and just. There are innumerable opportunities for making a difference that matters here, starting locally.

Bridging Local Gaps

Consider your own imaging team: When you hold video meetings, do all members have equal access to the necessary hardware and software to participate effectively? Are all members afforded the same privacy and time to participate in these meetings? This lesson was brought home to us when we recently convened a video meeting of our wellness council and noticed that several of our technical and nursing staff did not have access to video equipment in their workplace.

Consider your patients, as well: While a health care system might deploy sophisticated software to support their telehealth endeavors, this does not mean that all patients have the necessary hardware or software to participate. Additionally, solutions to barriers such as vision, language, and hearing must be readily available. One additional effort I applaud is to make our digital reports more comprehensible; not every patient understands what is meant by the phrase “the hepatic parenchyma demonstrates a normal echotexture,” nor should they. We should support software solutions to simplify the communication and accuracy of our recommendations.

And in keeping with our educational mission, think about the brisk implementation of so many solutions to support ongoing academic efforts. Will we ever return to our traditional morning resident teaching conference? I’d imagine not; if anything, the pandemic will finally allow us to move away from the prolonged didactic and synchronous teaching methods to ones that are more appropriate, personalized, and contemporary.

Another essential pillar of academic radiology is teaching and developing the next generation of radiology leaders during readouts. We seem to be mired in surveys and comparisons about what processes work best for our traditional readouts. Let’s instead open our eyes to completely new and asynchronous approaches. What an opportunity! And last within this category is lifelong learning. The necessary transformation to virtual national academic meetings this past year has demonstrated the many advantages that our digital environment offers for such forums. Be it cost savings for participants and practices, wider availability of CME credits and on-demand content, less time away from the workplace, or and the ability to directly connect with speakers, the benefits are plentiful.  

Keeping Our Imaginative Focus

Where the opportunities lie here are in fostering participant connections and rethinking how we should transform the content, styles, and media of our traditional talks to take full advantage of individual learner needs and preferences. Again, what terrific opportunities exist in this domain!

So, where do we go from here?

While tremendous and necessary strides have and continue to take place in our abilities to communicate, manage, and connect remotely, I only ask that we continue to be mindful and considerate that not all stakeholders are currently able to participate equally and effectively. The phrase “you’re only as fast as the slowest member of your relay team” is so apt nowadays. In our digital environment, the concept of “precision medicine” should now expand to embrace the specific needs and preferences of our many stakeholders.

As we continue to build and expand our digital frontend, it is equally necessary to focus on supporting the backend, so that all of our team members and stakeholders can participate and benefit from the systems and solutions that are being deployed. The opportunities here are endless, and we need to develop, implement, and share solutions that will ultimately meet the needs and improve the outcomes for our patients. Let’s please keep our imaginative focus on why we entered this wonderful, exciting, and ever-expanding field of radiology in the first place.

^{_{Updated October 25, 2021}}

As MRI technology continues to improve, radiologists must maintain a mastery of complex sequences, evolving protocols, and advanced techniques. Likewise, disease-specific guidelines, protocols, and reporting continue to evolve. Radiologists in all practice types are often tasked with managing a busy practice at this intersection of advanced technology and state-of-the-art clinical care.  

From September 9–10, the ARRS Virtual Symposium, Abdominal MRI: Practical Applications and Advanced Imaging Techniques, delivered trusted perspectives on the most pressing issues in body MRI—ranging from protocol and acquisition optimization, incorporation of advanced techniques in routine clinical practice, and interpretation/reporting for the most important trends in liver, prostate, emergency, and gynecological MR imaging. What follows is our primer for those unable to attend the live sessions.

Emergency MR Imaging

Although CT is the workhorse imaging modality in the emergency department (ED), there are many concerns related to ionizing radiation exposure in populations that are vulnerable due to young age, pregnancy, or exposure to repeated imaging examinations. In patients with nontraumatic acute abdominal symptoms, non-contrast MRI offers similar diagnostic performance to CT. For instance, in a prospectively enrolled cohort of 48 patients with head-to-head comparison of MRI to CT, there was no significant difference in performance for diagnosing acute appendicitis in young adults and adolescents (Fig. 1).

• 

A key to harnessing the efficacy of MRI in the ED setting is through protocol optimization, focusing on efficiency and speed. Abbreviated protocols, comprising fewer and faster sequences, allow for improved adoption, decreased cost, and maintained sensitivity for answering directed clinical questions. For example, in the ED setting, compared to conventional MR cholangiopancreatography (MRCP) protocols, abbreviated MRCP provides significant time savings, while maintaining similar diagnostic accuracy for the detection of choledocholithiasis. ARRS Symposia course director Victoria Chernyak, ARRS Instructional Courses Committee chair Courtney Coursey Moreno, and Elena Korngold highlighted ED MR protocols and imaging findings for common genitourinary and gastrointestinal emergencies.  

Advanced MR Techniques

Beyond the ED setting, advanced MR imaging techniques have opened the door for quantitative imaging, and multiparametric assessment has become standard practice for many disease processes. Sequences for assessment of liver steatosis, iron deposition, and fibrosis are now available on all major MR vendor platforms, allowing accurate diagnosis and monitoring of patients with chronic liver diseases. A working knowledge of how to extract and report the quantitative data derived from MRI proton density fat fraction, R2* maps, and elastography is required to build a state-of-the-art radiology liver practice. Diffusion weighted imaging (DWI) is no longer an ancillary or optional sequence but is required for accurate multiparametric assessment of prostate cancer. DWI can be used as a biomarker of tumor response; for instance, DWI adds value in assessing response to neoadjuvant therapy in patients with rectal cancer. While important and useful, DWI can be challenging to optimize and interpret in practice. 2002 ARRS Scholar Claude B. Sirlin imparted his applied wisdom regarding DWI acquisition optimization and interpretation, Mustafa Rifaat Bashir offered insights into some of the new sequences and future directions, and Antonio Carlos A. Westphalen emphasized the utility of DWI in the current prostate imaging reporting and data system (PI-RADS) v2.1. 

Protocol Optimization

In practice, MRI provides both great potential and great challenges. Optimizing sequences across multiple scanners, technologists, and protocols can be laborious and frustrating for radiologists. Optimized MRI protocols achieve a delicate balance between acquisition times, image quality, and sequence comprehensiveness, and they are—to paraphrase Albert Einstein—made as simple as possible, but no simpler. Richard Kinh Gian Do, Steven S. Raman, Elizabeth A. Sadowski, and Korngold shared their expert insights into optimal protocols for hepatobiliary, prostate, female pelvis, and bowel imaging. Bashir presented real-world methods for recognizing and, most importantly, mitigating artifacts commonly encountered in abdominal MR imaging.

State-of-the-Art Reporting

Optimizing images is just one hurdle for delivering the best imaging care to patients. Over the last decade, there has been a major movement toward standardized reporting for many disease processes in the abdomen and pelvis. Standardized reporting allows for more precise communication of results and improves compliance with diagnostic criteria. Notably, the liver imaging reporting and data system (LI-RADS), PI-RADS, ovarian reporting and data system (O-RADS), and updated Bosniak criteria are increasingly recognized as the standard of care for interpretation and reporting. ARRS Symposia course codirector Kathryn J. Fowler, Westphalen, Raman, and Sadowski provided an overview of these important systems, a framework for applying them, as well as insights into creating templates to improve reporting efficiency (Fig. 2).

• 

Delivering state-of-the-art care with MRI requires comfort with the technical aspects of image acquisition, reporting standards and approaches to common disease processes, and advanced sequences as integrated into practice. Taught by world-renowned MR imaging experts, Abdominal MRI: Practical Applications and Advanced Imaging Techniques provided practicable tools to harness the power of these expanded procedures for improving patient care.

^{Published June 28, 2021}

In the past decade, interest in artificial intelligence (AI) research in radiology has increased dramatically, as reflected by the tenfold increase in papers published in the subject over this period. This development is accompanied by fears from radiologists that AI will eventually replace human readers for the diagnostic interpretation of imaging studies. This article will review the current methods AI uses for diagnostic imaging tasks, the challenges AI algorithms have to overcome for their more widespread implementation, and how radiologist roles may shift to better accommodate the strengths of machine learning.

Deep Learning: An Introduction

The term “artificial intelligence,” in the most general sense, refers to the ability of algorithms to mimic human cognitive abilities. AI algorithms used for image analysis are typically designed using neural networks, which involves layers of processing nodes organized into an input, output, and multiple hidden layers. The architecture is feedforward; outputs of the previous layer are used as the inputs of the next. Each layer is composed of nodes, each of which has a weighted function that associates the inputs to outputs. The function weights are determined by backpropagation during training, an algorithm which updates function weights by minimizing a loss function.

Convolutional neural networks (CNNs) are neural networks specifically designed to handle image data and are the most common, though by no means only, deep learning architecture used for medical image applications. In diagnostic imaging studies, CNNs are usually trained with supervised learning by radiologist annotated data. The layers of a CNN process input images into an image classification output decided by weighted probabilities. For example, one hypothetical CNN’s classification output of a chest radiograph may be an 80% probability of pneumonia diagnosis, 15% pleural effusion, and 5% pneumothorax. The architecture of a CNN can be summarized as follows: first, convolutional layers extract features from an input image. The resulting feature map then passes through pooling layers, which decrease the number of trainable parameters by reducing the dimensions of the convolutional layer input [2]. Finally, a fully connected layer uses the inputs from the feature map and applies weights to parameters to classify the image, typically using rectified linear unit (ReLU) as its activation function.

CNNs have been developed for a wide range of diagnostic tasks based on modality image input data. They have shown comparable results to radiologists for identification of lung nodules and quantification of coronary artery calcium volume during low-dose CT screening, comparable performance to radiologists on the diagnosis of multiple thoracic pathologies from chest radiographs, and better diagnostic performance than radiologists for diagnosis of pneumonia from chest radiographs. Aside from applications for chest imaging, CNNs have achieved high accuracy (AUC) for liver fibrosis staging, accurate segmentation of clinical target volume (CTV) and organs at risk (OARs) in colorectal cancer films, and improved prediction of overall survival in glioblastoma patients from MRI data. However, while the results of CNNs are often comparable to diagnostic radiologists, it is important to note their detection tasks are narrow in scope, and like other AI methods, they are dependent on high quantities of quality training and validation data.

Challenges

There are several challenges to AI implementation in diagnostic radiology not readily solved with mere improvements in technology. The opaque nature of an AI algorithm’s functionality is one such challenge. The nature of backpropagation is such that algorithm developers are inherently unable to explain why parameters end up at their trained values, only that they were the values the algorithm discovered as optimal. Likewise, the precise features of an image that a neural network used to arrive at its classification output are impossible to know. In this regard, a neural network resembles the brain it was loosely inspired by—scientists understand that they work, but not how or why. This “black box” quality of machine learning has not, however, prevented the FDA from approving commercial AI-based medical devices and algorithms, and the history of medicine shows the field has accepted other black boxes for patient care. For example, the exact mechanisms by which many drugs work or produce side effects are poorly understood, but such drugs are used and trusted by physicians. Computer scientists are also attempting to make AI more transparent by developing processes to delineate the methods by which neural networks reach conclusions. This new field, called explainable artificial intelligence (XAI), is an active area of research.

Nevertheless, it is easy to imagine a future where AI becomes part of standard of care for diagnostic radiology, wherein health systems, and specifically the radiologist, would still be responsible for algorithm errors, even if an AI’s reasoning for the error is unexplainable. It is still unknown who would bear the liability for an algorithm misdiagnosis, nor whether the public and their primary physicians would accept machine diagnosis.

The training and validation of medical image algorithms is yet another challenge to their implementation. Deep learning algorithms such as CNNs often require huge amounts of data because of the high number of parameters that must be optimized during training. Large high-quality datasets, which must be annotated by radiologists, are expensive to procure and often several orders of magnitude smaller than the training datasets used for other deep learning applications. They usually number from the hundreds to the tens of thousands, compared to the 4 million labeled images Facebook’s DeepFace facial recognition neural network used as its training dataset. There has been recent progress on creating publicly available image datasets for algorithm development and on medical image analysis challenges that provide datasets for developers. Stanford’s CheXpert dataset is the largest of these, with 224,216 labeled chest radiographs available for public use. However, training datasets is still often proprietary due to privacy concerns or intended commercial use of the algorithms utilized. This dearth of training data and lack of transparency can create problems of algorithmic biases; homogenous training data can cause health care AI algorithms to weigh certain diagnoses unequally based on socioeconomic status, race, or gender, and properly diverse datasets are difficult to assemble.

AI and the Future of Radiology

However, the greatest challenge for AI in diagnostic radiology may be whether it replaces a diagnostic radiologist’s function entirely, and the resulting health care implications for patients. Medical experts have even raised the question if diagnostic radiologists should continue to be trained, given the fast-approaching integration of AI technologies as part of diagnostic algorithms, which will almost certainly include some form of initial automated machine-driven image analysis. The advantages of AI over human radiologists seem overwhelming at first. Computers can work 24/7 without signs of fatigue. While diagnostic radiologists typically read between 10,000–20,000 films annually, the number of images a neural network could read in the same time period is multiple orders of magnitudes greater, in the tens or hundreds of millions. As such, the economic reasoning for institutions to replace diagnostic radiologists seems obvious. However, currently developed algorithms only accomplish tasks narrow in scope (e.g., the binary classification of a lung nodule or the density analysis of a renal stone), whereas radiologists read studies holistically, looking for all possible abnormalities displayed in an imaging study. For a computer to read a diagnostic imaging data set in the same way a radiologist typically does would require the development, training, validation, and integration of thousands of discrete AI algorithms into one workflow. This is a difficult task, and though theoretically possible, a single generalized AI algorithm that can holistically analyze all features of an image for all possible diagnoses is even harder. For the foreseeable future, radiologists will still be required for accurately reading film, with AI packages offering decision support algorithms, and boosting reader efficiency by both providing second opinions on the primary detection tasks and simplifying workflow (e.g., prepopulating reports, automatically creating ROIs, etc.)

Even if AI technologies improve such that diagnostic AI application packages become capable of holistically reading modality studies, one must also remember that radiologists are not just image analysts, but fundamentally clinicians, trained to interpret and communicate imaging findings in the clinical context of the patient. Radiomics that uses AI generates complex data that must be interpreted by radiologists and linked to clinical uses; this is a need that will only grow in the future. AI-driven increased efficiency of image analysis may cause radiologists to pivot to more patient interaction and to become even more integrated in the clinical decision processes, in collaboration with a patient’s care team.

Given the uncertainties of the black box nature of AI, radiologists appear to be the best positioned medical professionals to elucidate a neural network’s probability outputs to a patient, along with the image features the machine is using to arrive at its diagnoses and their meaning. To best synergize with AI, radiologists of the future will have to improve care through knowledge of and empathy for their patients, the ability to identify which AI-produced data is relevant to meet diagnostic demands, and the interpretation of scans to elucidate the next steps following image findings to better advocate for their patients.

AI, and in particular convolutional neural networks, have seen success in narrow detection tasks for medical imaging diagnoses. While various challenges exist, such as their opaque nature and the cost and small scale of training datasets, they will likely be surmountable in time. AI will inevitably enter the diagnostic radiologists’ workflow. Although AI will not replace radiologists in the future, it is likely that radiologists who use AI will eventually replace radiologists who do not. The use of AI will lead to more efficient image diagnosis, in combination with optimized decision support algorithms, which will benefit patient care. In the not too distant future, AI could usher a potential realignment of radiologist duties towards a more interactive and patient-focused paradigm, in addition to traditional models centered around the reporting of imaging findings.

^{Published June 28, 2021}

For valiant service selflessly rendered on the frontlines of the fight against COVID-19, the American Roentgen Ray Society symbolically awarded each and every one of our members the 2021 ARRS Gold Medal. The ARRS Gold Medal Story Series shares perspectives of imaging professionals who conquered the day-to-day challenges of battling COVID-19.

The declaration by the World Health Organization of the coronavirus disease (COVID-19) pandemic in December 2019 led to an avalanche of information which did not manifest clearly, especially here in Nigeria. The signs, symptoms, presentation, and management of patients with COVID-19 slowly evolved for us.

The breast imaging sections of our hospitals went into an almost complete lull, as radiologists were exceedingly careful about exposing themselves to patients whose infection status was unknown. Testing patients for COVID-19 was not easy, due to lack of resources and consumables. More so, there was a dearth of personnel protective equipment (PPE), which was in extremely short supply from hospital management. Initially, breast clinics were shut down, and all activities were suspended. We had to set up an Infection, Prevention, and Control (IPC) Committee, which I chaired, while developing core guidelines for operations in the department.

All of the underlisted were considered crucial standard operating procedures:

• Undergo IPC training and tutorials
• Audit staff for protection
• Inventory, maintain, and decontaminate equipment
• Aerate all facilities
• Identify isolation opportunities for at-risk patients

Contamination Prevention

With temperature checks and masks mandatory for entry, our front desk served as patient registration only, not for triaging, so that foot traffic did not build up. To maintain social distancing of at least 2 meters apart in all waiting areas, the floors were marked in red to depict this distance. All persons entering the triage zone had to wash their hands with soap and running water. Posters showing proper modes of mask-wearing and hand-washing—as well as for gloves, goggles, and faces shields—were posted around the hospital. Only one relative was allowed to accompany a patient who could not stand on their own. To further reduce foot traffic, payment areas were augmented and fast-tracked.

Examination Rooms

Only one patient at a time was allowed in the examination room, and the radiographer/radiologist was well-kitted, depending upon that patient’s risk assessment. For COVID-19-postive patients, we had a dedicated mobile radiography unit, though all rooms were decontaminated after every procedure.

Decontamination

With hand sanitizers at each interactive point and zone, hands were properly washed before and after every procedure. All surfaces were decontaminated after each procedure, including cleaning and decontamination of reusable PPE and proper disposal of disposable ones. Our hospital’s standardized protocols for decontaminating an imaging room and equipment—especially CT after attending to a patient with COVID-19—featured downtime of about one hour in between procedures. Result retrieval points were set up in a cubicle outside the facility, and to reduce human occupancy, results were dispatched by e-mail. Remote reporting was encouraged, too.

Communication

Bold signage was displayed all over the hospital, including clear instructions and visible explanation notices posted on our doors. Relevant phone numbers (e.g., Central Preparedness Team, Hospital Infection Control Committee, Departmental IPC, National Center for Disease Control, etc.) were displayed on the front desk, triage zone, and notice boards.

Capacity Building

In addition to restricting their local and international travel, our entire staff received regular training (and retraining) in IPC protocols, especially how to don and doff PPE.

Some of our radiologists also contracted COVID-19. It got to a stage where all the staff in the pathology department tested positive, so ultrasound-guided biopsies were not possible. Even patients had restricted themselves from showing up for treatment in the hospitals, for fear of the unknown and restriction of movement. Breast cancer patients preferred to receive treatment through phone calls, since there were no teleradiology options.

However, musculoskeletal services were made available for patients who needed to be seen as emergency cases.

With the gradual improvement of the knowledge of the disease, availability of testing facilities, PPE, and now, vaccines, there has been much easing of restrictions. A gradual return to normal is envisaged. In our hospitals, there has been a return of influx for breast imaging patients in the last two months, especially for ultrasound and mammography, since these are the readily available investigations.  

_{Published March 16, 2021}

With our April 2020 AJR paper, “DICOM Images Have Been Hacked! Now What?,” well received by the ARRS membership, we were asked to record both an accompanying podcast and live webinar. We also presented our results at the Cybersecurity Refresher Course during the Radiological Society of North America 2020 Scientific Assembly, as well as at many other venues. Due to so many new developments in the worlds of cybersecurity and health care since the publication of our original AJR review, I have been invited to provide an overview summarizing a few of those latest developments.

Recent Hacks

The number of cyberattacks targeting medical institutions continue to increase. In 2020, 79% of all cyberbreaches affected the health care sector, so our sector has become a prime target. The breach portal of the U.S. Department of Health and Human Services Office for Civil Rights has reported a total of 620 new breaches of medical records in 2020. Trinity Health in Livonia, MI took the largest hit, when more than 3.3 million records with patient information were compromised as a result of a ransomware attack on Blackbaud, a cloud computing provider selling a fundraising database software.

Other recent breaches and attacks have received lots of media coverage, for example:

• In September 2020, Universal Health Services (UHS) was forced to shut down all computer systems at its facilities around the U.S. after a cyberattack by the Ryuk ransomware. This attack was likely triggered by a phishing email. UHS operates more than 400 health care facilities across the U.S. and U.K. After the shutdown, many of their facilities were left without access to computer and phone systems. Access to anything computer-based—from old labs and ECGs to radiology studies—was lost. UHS was forced to redirect ambulances and relocate patients in need of surgery to other hospitals nearby. Following the incident, four deaths were reportedly caused by delays in lab results arriving via courier. However, it is unclear whether or not these deaths were directly related to the cyberattack.
• In Düsseldorf, Germany, also in September 2020, a ransomware attack against Heinrich Heine University gravely affected its University Hospital. Cybercriminals encrypted about 30 hospital servers, preventing access to important medical information for patient care. Doctors had no access to this information, so patients had to be redirected to other hospitals. As a result, a female patient in transit to the emergency department in critical condition was rerouted to a hospital 20 miles away. With the detour causing a one-hour delay in her care, she died in transit—known to be one of the first cases of proven death from ransomware.
• A study by Greenbone Networks in September 2019 revealed that, worldwide, billions of confidential medical images on DICOM servers were freely accessible on the internet. This study headlined the news and even caught the attention of Congress, where Senator Mark Warner of Virginia became a strong supporter of improving the security of medical servers and images. In October 2020, New Net Technologies directed a follow-up study and discovered that, in the U.S. alone, millions of unprotected medical images were still exposed on the internet.
• In March 2020, Russian hackers (APT29), who were also responsible for the Democratic National Committee hack in 2016, inserted malicious code within an update of SolarWinds’ Orion software, which monitors the computer networks of governments and businesses to detect problems. Once the hacked update was downloaded by users, the perpetrators were secretly granted remote access to all the networks monitored by Orion, allowing for complete control and the ability to easily steal information. Hundreds of government institutions and private companies have been affected, including the Departments of Homeland Security, Treasury, Commerce, and Justice, as well as the Pentagon, Postal Service, and National Institutes of Health. As this article is being written, investigation continues to reveal the full extent of the damages. So far, at least 250 federal agencies and businesses have been compromised by the hack.

COVID-19 Vulnerabilities

The outbreak of coronavirus disease (COVID-19) created a new set of problems for cybersecurity, producing a triple threat for health care systems:

1. rapid expansion of networked devices and services, creating an expanded attack surface
2. increase in the different types of cyberattacks
3. fewer available resources to defend against cyberattacks

The use of telehealth has surged during the COVID-19 pandemic. At my institution, the Hospital of the University of Pennsylvania, consultations by telehealth skyrocketed from 50 to 7,000 per day. Many radiologists continue to work remotely from home, creating additional vulnerabilities in security.

A home radiology workstation connects to a home router, which connects via the internet to the hospital virtual private network device, which itself connects to the hospital servers. Each of these connection points are vulnerable. Radiologists reading remotely often forget to change the default administrator password on their home router. Hackers have performed domain name system hijacking on hundreds of thousands of home routers, redirecting links from legitimate institutions to hackers’ websites and intercepting data.

• 

As soon as COVID-19 became a pandemic, there was a massive surge in different kinds of cyberattacks. Whenever there is a social crisis, cybercriminals exploit the situation in full force, as people are more stressed out and, therefore, more prone to make mistakes. Phishing attacks pretending to originate from entities such as the World Health Organization spread across the globe like wildfire. These emails included fake links and malicious file attachments.

Real websites, such as the Johns Hopkins Coronavirus Resource Center’s COVID-19 map, were quickly duplicated, becoming major sources of worldwide malware distribution. Upon visiting the fake websites, computers became infected by Trojans stealing crucial information. Information about those fake websites was spread via phishing emails, malicious online advertisements, social engineering, and search engines. Since the beginning of the pandemic, over 60,000 COVID-19-related fake websites have been created.

Meanwhile, nation states, like Russia and China, have been attacking pharmaceutical companies and vaccine developers to steal intellectual property. They use phishing emails to obtain login credentials, and then use exploits to transmit files or execute code remotely. Two well-known groups of cybercriminals, APT29 from Russia and APT41 from China, have stolen COVID-19 research data by exploiting weaknesses in servers and routers.

Some Solutions

Late last year, the National Cybersecurity Center of Excellence, part of the National Institute of Standards and Technology (NIST), released its final practice guide on how to protect PACS technology and DICOM servers. NIST recommended a combination of strategies, including defense in depth, access control mechanisms, a holistic risk management approach, and the use of cloud storage. A defense in depth strategy involves multiple layers of defense at the level of the perimeter, the network, the workstation, the application, and the data; if one fails, data are still protected. NIST also recommended network segmentation into groups of devices sharing similar functionalities. This can be accomplished through virtual local area networks or by finer segmentation using software-defined networking—often used to secure legacy devices that lack inherent security features.

Also in December 2020, the Health Sector Cybersecurity Coordination Center issued a sector alert regarding vulnerabilities in DICOM image servers, while adding a series of recommendations on how to better protect them. These suggestions included general ones (use secure passwords, close unused computer ports, apply the most recent patches), as well as using encryption of data at-rest and in-motion, restrictions on network access, and network segmentation.

• 

The U.S. government enacted into law the Internet of Things (IoT) Cybersecurity Improvement Act of 2020 in December, too. Responding to the significant increase in control of IoT devices by cybercriminals during the year, this law establishes new mandatory minimum-security standards for IoT devices purchased using government funds, including supply chain security.

Early this year, the U.S. government enacted into law the HIPAA Safe Harbor Bill to amend HITECH (Health Information Technology for Economic and Clinical Health) to incentivize the use of cybersecurity best practices for meeting HIPAA requirements, especially those recommended by the NIST. The HITECH Act from 2009 was responsible for expanding the adoption of electronic health records (EHR) by health care providers— keeping one million U.S. physicians busy every evening, entering clinical data into the EHR.

What can we expect in 2021 for cybersecurity in health care? The new administration will start by repairing the massive damages to the cybersecurity infrastructure caused by the previous administration. Nation states will continue cyberattacks on COVID-19 vaccine developers to steal trade secrets and gain a competitive advantage. Smaller health care institutions, barely surviving this pandemic economically, will face increased attacks by cybercriminals. They do not have the same cyberdefense budget and manpower as larger institutions, although they face the same cyberthreats. As health care organizations transition more and more of their data to the cloud, we should see increasing numbers of data breaches from patient data on cloud infrastructures. Phishing with a health care theme will be more prevalent, given the focus on all health issues during COVID-19. And IoT and wearable medical devices will remain targets of cyberattacks for the foreseeable future, until the industry fully implements the new IoT security standards imposed by the latest legislation.