What: 2017 C-MIMI
When: Sept. 26-27, 2017
Where: Johns Hopkins Medicine in Baltimore
The Society of Imaging Informatics in Medicine (SIIM) wrapped up a successful SIIM 2017 in Pittsburgh earlier this summer, and now its members are preparing for the second annual Conference on Machine Intelligence in Medical Imaging (C-MIMI) in September.
Paul G. Nagy, PhD, CIIP, currently serves as the chair of SIIM and is also an associate professor at Johns Hopkins University School of Medicine in Baltimore. Nagy spoke with Radiology Business Journal about C-MIMI and how he thinks deep learning will affect the future of radiology.
What are some the biggest concerns you hear about when speaking to SIIM members and others from within the industry?
Both in imaging and in health IT, security has become a big topic. We really hadn’t hardened our systems like other industries and we have become vulnerable to hackers trying to steal data. So people are putting a lot more energy into security from all levels; it’s a hot topic we all have to deal with.
There are some other things I see people grappling with, including the need to share and exchange medical images between institutions. Even within large institutions where there are multiple hospitals that may not be on the same PACS, image exchange has become a really important part of patient care. And a third topic I see people beginning to tackle is enterprise imaging governance. In radiology and cardiology, we were ahead of the ball with how we managed images in terms of system integration and data management practices. Now, however, we’re finding that there are images everywhere in every department. Whether they are taken on a phone or are managed in ophthalmology or neurology, we’re finding that the techniques we’ve learned in radiology and cardiology need to be adapted throughout the enterprise. It means working across clinical specialties, working with the electronic medical record (EMR) and working with the clinical leadership.
Radiology, like all other specialties within healthcare, is experiencing a lot of change right now as it shifts toward value-based care. What role do SIIM and its members play in this transformative time?
First and foremost, for radiologists, the PACS workstation is their surgical suite. It is the instrument through which they deliver patient care. Ensuring that technology is an optimal environment for them to deliver care to their patients is paramount. Radiologists ultimately
depend on PACS workflow, protocols, data management—if these things are not properly configured, it can hamper their ability to deliver care and cause patient safety issues. That’s the first primary role, to ensure radiologists understand how to use this technology for delivering patient care.
I think there’s a second role as well, and that’s training physicians to be leaders in both imaging informatics and the use of IT in adopting new technology. This entails communicating these needs to technical folks and the vendor community. That training helps create a bridge between radiologists and the technical side.
This is all true for radiologists as well as for technologists and other specialty physicians leading change in healthcare.
SIIM joined forces with HIMSS in 2014 and established the HIMSS-SIIM Enterprise Imaging Workgroup. Can you speak a bit about that group? Why was it so important to collaborate with HIMSS and focus on enterprise imaging?
It has truly been a fantastic collaboration, bringing two perspectives together. For us, it was important to understand a non-specialty perspective of image management workflow. We’re used to a referring physician placing an order for a CT scan and then tracking everything stemming from that order. We create a worklist, we ensure images are associated with the worklist, we create a report—we have it all down tight in terms of workflow.
But we had to go back to the drawing board and learn about encounter-based imaging, which is a large part of enterprise imaging. This is when a physician is with a patient and wants to image the patient to document and track clinical changes. We didn’t have a workflow worked out for that, and the HIMSS-SIIM Enterprise Imaging Workgroup helped us identify the various methods of clinical documentation that can benefit from good data integrity around imaging.
Another great benefit of the collaboration was gaining the HIMSS perspective on security and governance. I’m happy with how much we’ve learned and shared by joining forces with HIMSS.
Speaking of enterprise imaging, what does a sound, efficient enterprise imaging strategy look like?
Governance and integration are the two fundamental keys to having a successful enterprise imaging strategy. It needs to be closely tied to the organization and it needs to be heavily integrated with the EMR. The HIMSS-SIIM Enterprise Imaging Workgroup has put out white papers that lay this all out. They’ve been downloaded more than 15,000 times in the last year and they cover workflow, data integration strategies, governance, how to gain different perspectives, and so on. This is redefining the field of enterprise imaging.
SIIM is hosting its second annual C-MIMI in Baltimore in September. What led SIIM to developing this two-day conference?
We’ve seen a disruptive growth in what we call Deep Convolutional Neural Network imaging machine learning algorithms and their ability to extract promising features from images in an analogous way to pattern recognition functions of the visual cortex. It takes considerable computing power, but it has been able to improve the performance of image detection compared with other algorithms. You see this being used by Google and Facebook in facial recognition, and researchers are now applying it to medical imaging. This meeting is our way of bringing different groups that are using this research together. There is a large need for curated data sets to do these types of tests, so there’s also a lot of discussion about how to build data sets.
At SIIM, our role is identify exciting technology and help translate them into clinical value. We’re looking at how to evaluate these algorithms and how to find the right clinical problems they can be applied to. It’s an exciting new tool in healthcare research.
What does SIIM have planned for C-MIMI this year?
The meeting is going to have two components. First, there’s going to be a deep learning institute where we have training on machine learning algorithms —it is geared for physicians, medical students and researchers. There’s a lot of ways to use these algorithms and several ways to get in trouble.
The second part of C-MIMI is going to be where researchers present what they’ve been able to accomplish and the barriers they’ve faced. It will help us see the state of machine learning in medical imaging today.
Radiologists appear to be divided on AI and deep learning—some think these advances are exciting, but others fear they could soon be replaced. What are your thoughts on the future of these technologies and how they will impact radiologists?
There’s this great book about the history of computer technology by Walter Isaacson, The Innovators. The moral of the story is that human and computers work better together. What we see today is that most research groups are working to get machine learning to assist the radiologist with documentation and as a potential screening tool. I see machine learning as a complementary tool for radiologists that makes them faster and helps with documentation tasks. I see it as an assistant for quite some time. We’re already having problems today where radiologists are overwhelmed by information, and they need tools to help them search and find context so they can be effective with their diagnostic approach. We need to be able to help radiologists consume more and more information, and I think these technologies are tools that hold a lot of promise to help.