As most of you in the industry know, the Radiological Association of North America (RNSA) held it's annual conference at the end of November... Countless vendors showed off their latest and greatest developments pertaining to medical imaging, and professionals and industry intellectuals informed us about what we can expect in the next few years.
We know that with so much information coming from the show, it's hard for important information to get lost in the shuffle. Well, the gracious folks at Aunt Minnie have compiled their Top 5 Trends again for us this year so that even if you didn't attend the conference, you now know some of the highlights. Read below for Trends #1 and #2, and come back later in the week for the second installment in this 2-part series...
Stop us if you've heard this before, but there was a dominant theme at this year's RSNA 2018 meeting in Chicago. And that theme, believe it or not, was artificial intelligence (AI).
As at RSNA 2017, sessions on AI and the related topics of machine learning and deep learning dominated the halls of McCormick Place. However, although the topics were similar from year to year, the debate around artificial intelligence has begun to shift in subtle ways.
Gone is the paranoia about AI taking radiologists' jobs -- but, to be honest, that disappeared a couple of years ago. It's been replaced by a frank discussion of the best way to put AI to work for radiologists, such as whether AI should be launched from the PACS or the reporting system. Also, which of the AI "marketplaces" under development will achieve dominance? And how will reimbursement be handled? The coming year should be an exciting one as more AI algorithms receive regulatory clearance and begin to move into clinical use.
In addition to AI, gadolinium deposition was another hot topic at RSNA 2018, as researchers struggle to figure out what's causing the phenomenon. Other important trends included contrast-enhanced ultrasound and the maturation of 3D printing.
1. AI Once Again Dominates RSNA, But Caution Emerges
Artificial intelligence was omnipresent at RSNA 2018, beginning with RSNA President Dr. Vijay Rao's call in her opening address for radiologists to embrace AI in order to situate the specialty once again at the center of patient care.
Commercial activity around AI continues to surge as start-ups and established companies alike vie to secure a niche in the burgeoning new era of AI-enabled radiology. The Machine Learning Showcase grew dramatically in size, with more than 70 small companies joining the fray to display their latest algorithms.
Elsewhere on the show floor, traditional imaging and IT firms spotlighted their AI initiatives throughout their booths. A number of vendors highlighted efforts to enable radiologists to utilize AI as part of their normal PACS workflow, addressing a key problem that could impede clinical adoption if unsolved.
The scientific program at RSNA 2018 was also permeated by presentations on AI. Researchers traveled to Chicago to share their progress in developing deep-learning algorithms for a variety of clinical applications, such as detecting all types of intracranial hemorrhage, classifying liver lesions on MRI scans and explaining the findings, characterizing pulmonary nodules on CT, acquiring diagnostic-quality MR images with just a fraction of the typical dose of a gadolinium-based contrast agent, and helping to avoid unnecessary thyroid nodule biopsies.
Other hot areas included triaging studies that need urgent review by radiologists, lowering radiation dose and decreasing scanning time, improving the accuracy of digital mammography and digital breast tomosynthesis, facilitating the communication of urgent results, and analyzing radiomics data for assessing risk and predicting treatment response. There was also a flurry of scientific papers showing high performance for detecting a variety of conditions on chest x-rays.
While excitement over the potential of AI in radiology remains high, there was also a growing awareness at RSNA 2018 of the thorny challenges that still need to be addressed before machine-learning tools can achieve widespread adoption. For example, AI algorithms can certainly achieve impressive results on carefully selected datasets, but there's no guarantee they will deliver the same performance on other patient populations or with other models of imaging equipment. Indeed, one presentation at RSNA 2018 highlighted the need for deep-learning algorithms to be trained with diverse datasets and to be tested on real-world cases prior to being deployed in clinical practice.
Speaking of data, there is still an urgent need for large, well-annotated training datasets; algorithm development to date has mostly focused on imaging applications for which there are readily available sources of training data. For example, the significant number of scientific papers presented at RSNA 2018 on chest x-ray applications can be credited in large part to the release in late 2017 by the U.S. National Institutes of Health (NIH) of a massive database of more than 100,000 chest x-rays. A new database of over 10,600 CT scans released in July by the NIH will also undoubtedly spur algorithm development. In addition, researchers continue to investigate the use of techniques such as natural language processing that could make it easier for institutions to assemble their own training datasets.
While vendors showed progress at RSNA 2018 in tackling the crucial issue of integrating AI algorithms into PACS software, much work remains before radiologists can easily use AI as part of their regular workflow and not have to transfer to another software application.
In addition, only a small percentage of firms that were showing AI software for clinical use at RSNA 2018 have received U.S. Food and Drug Administration (FDA) clearance for their products. Importantly, though, regulatory policy changes at the FDA toward imaging AI software have led to a smattering of recent approvals, with many other companies expecting to receive FDA clearance in 2019.
Financial matters also must be worked out for both vendors and users. While many start-ups focused initially on developing high-performing algorithms, the quest is now on to find sustainable business models. On the end-user side, cost justification will be a big hurdle for many healthcare institutions; any new technology released without reimbursement is always a difficult sell to the C-suite unless it can yield quantifiable benefits. Clinical validation studies would be a big help here.
For now, the biggest opportunities for AI appear to be in use cases that improve the workflow and efficiency of radiologists, as well as enhance the value and efficiency of radiology. That's how AI can help fulfill Rao's vision of truly patient-centered radiology and solidify the role of radiologists in the future.
2. Gadolinium Retention Still Baffles The Experts
One area that could certainly benefit from AI is the ongoing effort to understand more about why trace elements of gadolinium remain in the tissue of patients who have received MRI contrast agents. The gadolinium retention issue continues to baffle experts, and it was a major point of discussion at RSNA 2018.
One example of how AI could help came from U.S. and Swiss researchers, who developed software for mining data in PACS/RIS databases to drastically reduce the time it normally takes to analyze brain MRI scans from patients who were given macrocyclic or linear gadolinium-based contrast agents (GBCAs). Tools like this should go a long way toward accelerating research into gadolinium retention.
But even with all the studies that have been conducted so far, just as many questions remain about the short- and long-term presence of gadolinium and to what extent traces of the element are detrimental to patient health. Many patients who have experienced serious conditions after their MRI scans naturally attribute them to gadolinium exposure, but can a connection be definitively proved or disavowed?
"That is the single most important piece of information we still don't have an answer to," Dr. Emanuel Kanal, director of MR services at the University of Pittsburgh, told AuntMinnie.com. "We are not really sure whether or not having residual gadolinium in whatever form or forms it may be found in the brains of these patients [has] any significant clinical consequences."
Perhaps some answers will come from the recently formed worldwide collaboration of researchers who will investigate gadolinium deposition from two perspectives. One strategy will target animal and basic science research, while the other will focus on clinical research. At some point -- perhaps even by RSNA 2019 -- clinicians could have a better foundation for deciding whether to administer a GBCA and for knowing the minimal dose to achieve their diagnostic goals.
Beyond the gadolinium debate, researchers presenting at RSNA 2018 continued to broaden MRI's horizons by experimenting with new imaging techniques, such as black bone MRI for detecting skull fractures. The reconstructed 3D MR images appear to be comparable to 3D reconstructions of CT scans for detecting fractures. In addition, current tried-and-true MR protocols such as diffusion-tensor imaging could help predict which patients will progress to Alzheimer's disease or dementia.