Dario Arfelli is the World Wide Healthcare IT Marketing Manager at Carestream and is based in Genoa, Italy. He has a special interest in supporting market model changes including EHRs, VNAs, Enterprise Imaging and Cloud computing. He analyzes emerging approaches to integrating PACS within an enterprise-wide healthcare IT platform, as well as the VNA approach to centralizing all clinical data storage, management and access and how it all ties back to the future of the EMR/EHR. More recently he has turned his attention to artificial intelligence in medical imaging and how it will likely be integrated into radiology and medical imaging. He recently published an article defining the terms surrounding artificial intelligence for Carestream’s blog. PARCA eNews talked to him by phone.
Q. What exactly are we talking about when we're talking about artificial intelligence in radiology?
I decided to write the blog because every buzzword like “big data” or “artificial intelligence” can create confusion in the minds of people and not everybody understands the particular role of these in the imaging and radiology world.
In a near future AI (artificial intelligence) will be an important pillar of healthcare IT imaging solutions using different technologies and techniques to get automated clinical results in a faster way, optimizing efficiency and improving worklist orchestration.
Radiology will be dramatically affected by it in the next ten years, so we need to be prepared to use the right terminologies.
Q. Could you give me an example of how artificial intelligence would be applied to a radiologist’s workflow?
Yes, absolutely. Let me give you an easy example. Think about the patient that comes in a hospital with a headache, a terrible headache causing unusual pain. When the patient comes to the emergency room and the doctors don't understand what is the cause in a few minutes, they are going to order a brain CT scan to be sure that there are no issues there. Right now, what is happening is that they do a CT scan and then the radiologist on call is assigned to read and report the case to help clinicians take the right action. This is what is happening in a classic hospital nowadays.
With artificial intelligence techniques available today, so nothing from the future, as soon as the images are received from the archive from the CT scan a software will run and the if the data shows a brain bleed then, automatically the study will be raised to be read as urgent, alerting the doctors of the new top priority. As the study becomes urgent the study will be assigned to the right doctor that has the closest relationship with that patient or to the best subspecialty, neuroradiology in our case. The right specialist then already knows that there is a brain bleed, and can begin determining if the patient can be treated in the emergency room or needs to be moved. As you know in such cases there is a limited time that you can save the life of a person.
So this is an example of how artificial intelligence can be applied to get quicker results, and bring the right people together to improve treatment. The aim is to simplify the way of working for the radiologist, using artificial intelligence to improve efficiency and productivity.
Q. What types of radiological problems are best suited right now to this technology?
First of all, artificial intelligence works very well on CT scans. So the main modality on the AI landscape right now is CT scanner, both for heads where you can detect brain bleeds, and for abdomen where you can detect fatty liver or spine fractures.
AI also can help doctors to analyze the entire CT scan in the case of specific request for a study. Let me give you an example. A radiologist is asked to report a CT scan looking for a pancreatic disease diagnosis. Given the indication from the referring physician, the radiologist will look at the pancreas, following a protocol to be sure there are no issues connected to it.
But if there is available an AI application that is automatically looking for spine micro fractures and bone density it might provide an early indication of osteoporosis. The radiologist can be alerted about an abnormal value and can review the details improving the efficiency of the CT scan itself. With AI it is like having an assistant that is giving the radiologist an alert to look at another area of the scan.
Artificial intelligence is also being strongly applied to help detect tumors, particularly mammography tumors. Needless to say – AI won’t replace radiologists or specifically breast imagers. Rather, it will augment their ability to find the key data they need to care for their patients and present it in a concise, usable format.
These are the two main radiological areas that I see that are very attractive in this moment for AI because both of them can really help to improve the efficiency of the department itself.
Q. So you feel AI will increase in the number of incidental findings in imaging studies?
Yes, I think that often many people think machine learning and deep learning techniques, being used to work on the more complex examinations, but it should be the opposite. AI should be used to identify the easy stuff doing a sort of triage, to help doctors to prioritize the readings. The complex studies will have to be managed by a very expert human working with multidisciplinary experts, maybe using advanced algorithms to faster identify key findings.
Right now artificial intelligence software will help the radiologist to analyze the entire image set, helping to prioritize the urgent cases in an easier way.
Q. On a practical level do small community hospitals need to invest in some technology to allow them to take advantage of AI, or more simply, what kind of technology is needed?
Interesting question and very good one. I think that they have no choice but to invest in AI in the coming years. This is the reality because an investment in AI for a hospital is something that will substantially improve productivity and quality of care. So it is not question of whether you are a small or medium or large hospital or institution, I think that you will need to improve the efficiency of the radiology department because of changes in reimbursement based on value-based care.
Value-based care is changing the landscape for every care provider. You need to have better productivity with more efficiency in the radiology department and artificial intelligence will help both on analyzing imaging and optimizing the operations of your department.
Q. Do PACS administrators have a role to play in this and how is artificial intelligence going to affect PACS systems?
Radiology PACS are moving to enterprise imaging, so PACS administrators’ work is already broadening. They are becoming imaging administrators instead of only radiology PACS administrators.
AI will impact the overall enterprise, not only radiology. The PACS administrator will help to implement the right software for their hospital because they are a bridge between technology and clinical workflow. They understand very well what radiologists need in order to make their work more efficient.
Additionally in the future they (PACS administrators) will be much less focused on pure technology. Enterprise imaging is moving to the cloud, reducing the focus on technology, freeing time to allow them to focus on clinical workflow orchestration to improve efficiency.
Q. The last question is as you pointed out in your article that futurist (Ray) Kurzweil predicts that machines will have the capacity to be smarter than humans by 2029. My question is, are humans going to be ready to deal with that so soon?
You know, I think that we have time to address artificial intelligence in the right way. At some point AI will probably be smarter than humans, this is true, but in healthcare it will always be used on one dedicated line of thinking in order to improve care and speed up lesion detection. Anyway AI is already smarter in chess and computers are already winning games against humans.
The role of the human will be to use artificial intelligence responsibly, deciding when and where to use it. We have been very well trained by science fiction literature in the last 30 years.
Nevertheless, imaging in healthcare will be very affected by AI. Doctors should be ethically trained and they have to drive the adoption focusing on what’s best for the patient.
Q. Is there anything else I've missed or I you would like to emphasize?
The intent in the article was to put a stake in the ground and clarify what the vendors are developing, what does artificial intelligence, machine learning and deep learning mean and what are the applications in diagnostic imaging today, and what will probably happen in the future.
The future of imaging will be in the cloud in a few years, and artificial intelligence will help care providers to improve efficiency and outcome quality.
Before we were using algorithms based on a process or set of rules, now we are developing AI based software using a real set of data. AI will not replace any humans, but it will help them to deliver better results in improved productivity and increase efficiency.
We have to drive that and not be driven.
Their manager rocks! Thanks to their great team, I'm so happy I chose this venue for my event! I had a great time at event venue Atlanta and think the atmosphere is as good as it gets. I will be astonished if I ever find a negative review on these guys.
ReplyDelete