Author: Jake Fishman

Bad AI Goes Viral

Posted on by Jake Fishman

A recent mammography AI study review quickly evolved from a “study” to a “story” after a single tweet from Eric Topol (to his 521k followers), calling mammography AI’s accuracy “very disappointing” and prompting a new flow of online conversations about how far imaging AI is from achieving its promise. However, the bigger “story” here might actually be how much AI research needs to evolve.

The Study Review: A team of UK-based researchers reviewed 12 digital mammography screening AI studies (n = 131,822 women). The studies analyzed DM screening AI’s performance when used as a standalone system (5 studies), as a reader aid (3 studies), or for triage (4 studies).

The AI Assessment: The biggest public takeaway was that 34 of the 36 AI systems (94%) evaluated in three of the studies were less accurate than a single radiologist, and all were less accurate than the consensus of two or more radiologists. They also found that AI modestly improved radiologist accuracy when used as a reader aid and eliminated around half of negative screenings when used for triage (but also missed some cancers).

The AI Research Assessment: Each of the reviewed studies were “of poor methodological quality,” all were retrospective, and most studies had high risks of bias and high applicability concerns. Unsurprisingly, these methodology-focused assessments didn’t get much public attention.

The Two Takeaways: The authors correctly concluded that these 12 poor-quality studies found DM screening AI to be inaccurate, and called for better quality research so we can properly judge DM screening AI’s actual accuracy and most effective use cases (and then improve it). However, the takeaway for many folks was that mammography screening AI is worse than radiologists and shouldn’t replace them, which might be true, but isn’t very scientifically helpful.

Hospitals’ Outside Rebound

Posted on by Jake Fishman

A new McKinsey survey found that healthcare system leaders expect patient volumes to surpass 2019’s levels by next year, while revealing an interesting shift in how/where many of these patients will be getting their care.

The Rebound – The leaders from 100 large private US hospitals reported that their ED and inpatient/outpatient volumes returned to 2019 levels in July 2021 (so, somewhat pre-Delta) and forecast that 2022 volumes will be 5%-8% above 2019.

The Outpatient Shift – Outpatient procedures are expected to drive much of this future patient growth (+8% in 2022, +9% in 2023), with the biggest outpatient increases in orthopedics, psychiatry, and cardiology.

The Virtual Shift – Although it would take another lockdown to return to 2020’s virtual care numbers, the leaders expect virtual visits to represent around 15% of their outpatient volume in 2022/2023, 300% higher than in 2019.

It’s Not Just McKinsey – You probably don’t need a prestigious consulting firm to tell you that more procedures are moving to outpatient settings and more patient visits are being held virtually. The outpatient shift has been going on for some time, and the recent evolution of telehealth tech and home care delivery has brought some major home care commitments from the biggest systems in the country. We even launched an excellent new newsletter to help providers keep up with healthcare’s virtual shift.

The Radiology Impact – Technically the McKinsey forecast didn’t mention imaging once, but patients’ continued shift to beyond hospital walls will definitely have an imaging impact, including more virtual radiologist consultations, more outpatient image-guided procedures, and more at-home and near-home imaging. It could also mean less in-hospital imaging.

Unsupervised COVID AI

Posted on by Jake Fishman

MGH’s new pix2surv AI system can accurately predict COVID outcomes from chest CTs, and it uses an unsupervised design that appears to solve some major COVID AI training and performance challenges.

Background – COVID AI hasn’t exactly earned the best reputation (short history + high annotation labor > leading to bad data > creating generalization issues), limiting most real world COVID analysis to logistic regression.

Designing pix2surv – pix2surv’s weakly unsupervised design and use of a generative adversarial network avoids these COVID AI pitfalls. It was directly trained with CTs from MGH’s COVID workflow (no labeling, no supervised training) and accurately estimates patient outcomes directly from their chest CTs.

pix2surv Performance – pix2surv accurately predicted the time of each patient’s ICU admission or death and applied the same analysis to stratify patients into high and low-risk groups. More notably, it “significantly outperformed” current laboratory tests and image-based methods with both predictions.

Applications – The MGH researchers believe pix2surv can be expanded to other COVID use cases (e.g. predicting Long COVID), as well as “other diseases” that are commonly diagnosed in medical images and might be hindered by annotation labor.

The Takeaway – pix2surv will require a lot more testing, and its chance of maintaining this type of performance across other sites and diseases might be a longshot (at least right away). However, pix2surv’s streamlined training and initial results are notable, and it would be very significant if a network like this was able to bring pattern-based unsupervised AI into clinical use.

Veye Validation

Posted on by Jake Fishman

A team of Dutch radiologists analyzed Aidence’s Veye Chest lung nodule detection tool, finding that it works “very well,” while outlining some areas for improvement.

The Study – After using Veye Chest for 1.5 years, the researchers analyzed 145 chest CTs with the AI tool and compared its performance against three radiologists’ consensus reads, finding that:

  • Veye Chest detected 130 nodules (80 true positive, 11 false negative, 39 false positives)
  • That’s 88% sensitivity, a 1.04 mean FP per-scan rate, and 95% negative predictive value
  • The radiologists and Veye Chest had different size measurements for 23 nodules
  • Veye Chest tended to overestimate nodule size (bigger than rads w/ 19 of the 23)
  • Veye Chest and the rads’ nodule composition measurements had a 95% agreement rate

The Verdict – The researchers found that Veye Chest “performs very well” and matched Aidence’s specifications. They also noted that the tool is “not good enough to replace the radiologist” and its nodule size overestimations could lead to unnecessary follow-up exams.

The Takeaway – This is a pretty positive study, considering how poorly many recent commercial AI studies have gone and understanding that no AI vendor would dare propose that their AI tools “replace the radiologist.” Plus, it provides the feedback that Aidence and other AI developers need to keep getting better. Given the lack of AI clinical evidence, let’s hope we see a lot more studies like this.

Imaging Wire Q&A: Regulating Service with Bayer & MITA

Posted on by Jake Fishman

The “Right-to-Repair” movement’s migration from iPhones to medical devices has brought service to the forefront of the medical imaging industry, and it could have major ramifications on who can service medical imaging devices and how they’re regulated.

In this Imaging Wire Q&A we sat down with Bayer Radiology’s Dennis Durmis and MITA’s Peter Weems to discuss the evolving medical device service issue and how ongoing regulation might impact patients, clinicians, and OEMs.


Tell me about yourselves and how you became involved with the Right-to-Repair movement?

Peter: I’m Peter Weems, Senior Director of Policy and Strategy with the Medical Imaging and Technology Alliance, also known as MITA. We’re the primary trade association and standards development organization representing manufacturers of medical imaging devices.

The service issue is one of our top priorities given its very serious implications for patient safety. We’ve been working on this for several years, after our member companies started flagging instances where their devices had been improperly serviced by an unregulated third-party, creating concerns for patient safety and device performance.

They were reporting situations where third-party servicers bypassed X-ray system radiation controls and instances where devices had been repaired with twist ties or duct tape or non-validated parts that you could find at the hardware store.

Clearly, this raised all sorts of concerns about patient safety and device performance, motivating us to engage the FDA and Congress to make sure all servicers were held to the same quality, safety, and regulatory requirements.

Dennis: I’m the head of Bayer Radiology in the Americas and former chair of MITA, so I’m part of the group that Peter’s been talking to. I’ve been involved with the service discussion for quite some time as well.

Peter makes all the valid arguments. These Right-to-Repair bills continue to grow broader, bringing in medical devices to some state bills. That of course creates concerns over whether these medical products are safe and efficacious, and the risks to patients or even to medical practitioners.



What made you realize how serious this issue is?

Peter: A number of our member companies have flagged instances where imaging devices were improperly serviced by an unregulated third-party servicer, creating enormous risks to patients and healthcare providers in terms of safety and the quality of the devices’ performance.

We’ve accumulated a number of these case studies over the years to the point that we eventually felt that the FDA needed to take action to ensure that everybody is held to consistent quality, safety, and regulatory requirements no matter who is servicing the device.

Dennis: I’ve viewed this as a serious issue for quite a while. At one point in time, I was the head of Bayer’s multi-vendor service business, so I’m actually very familiar with what third-party servicing is. I’d like to think that I have a good idea of the right way to manage a third-party service business.

Since we operated as part of a medical device OEM, we had very stringent quality standards. We had engineers who reverse engineered products. We drew up our own schematics. We understood how the products worked. Basically, we did everything but register a 510(k) with the FDA.

As a result of seeing that and being involved in the third-party space, it was easy to see where some of the good providers were and where some of the really bad providers were. Some servicers were just going through the motions, including not having a quality system, not having training requirements, and not guaranteeing that a product would work in the manner it was expected to work.

This goes back a number of years for me, almost ten years, and it’s continued to raise its head. This previously involved discussions with the FDA about defining repair versus remanufacturing. The Right-to-Repair language has been a more recent issue in the medical device area, and it’s in direct opposition to what we believe in: safety and efficacy.



What are the current medical device service rules?

Peter: Right now, the FDA regulates servicing of medical devices only when it’s performed by a medical device manufacturer. Manufacturers have to register with the FDA, report to the FDA whenever they encounter a death, serious injury, major malfunction, and they’re required to have a quality management system in place that dictates certain things like training and documentation requirements, verification and validation of replacement parts, etcetera.

If you’re not a manufacturer, you’re not FDA regulated. So essentially there is no oversight over the third-party servicing industry. These third-party servicing businesses are not required to register with the FDA, they don’t have to file MDRs, and they don’t have to have any quality or safety controls in place.

Our priority is to make sure that these businesses are held accountable and are registered with the FDA and follow certain basic quality, safety, and regulatory requirements.



What are the patient safety implications of these service rules?

Peter: Fundamentally, this is about patient safety from our perspective, full stop.

Given the critical role they play in our healthcare system, we need all medical devices to perform safely and effectively for their intended use in every case.

That’s why manufacturer service organizations are FDA regulated and that’s why the FDA should regulate third-party servicing businesses in this space. If one of these devices fails to perform, the risks are real. We’re talking about anything from electrical shock to overexposure to ionizing radiation to infection to air embolisms.

Never mind that as our devices become more integrated with software, the cybersecurity risks associated with improper servicing grow ever greater.

For MITA member companies specifically, we’re talking in large part about devices used for screening or diagnosis. If an imaging device fails to perform, diagnosis could be missed or patient care could be delayed. That has a real material tangible impact on the livelihood of patients.

So clearly patient safety is motivating our engagement on this issue.

Dennis: Peter said it well. A patient should never have to worry about whether the equipment that’s being used on them, either for treatment or therapy, has been properly serviced and maintained. That shouldn’t be a concern of any patient or any health professional. Unfortunately, that’s not necessarily the case today and there are multiple examples of where that isn’t happening.

I don’t think that MITA’s recommendation to the FDA is too much. It’s simply to ask the third-party service companies to register with the FDA and maintain a quality system. That’s just good business practice. That’s nothing more than doing what’s supposed to be done regardless of whether you’re in a regulated space or not, to be honest with you.

So those shouldn’t be hurdles or significant cost drivers for any company. It’s simply just good business practices and that’s really what we believe in.



What will it take for Congress to introduce more stringent medical device standards?

Peter: We’ve been working with congress for a number of years on this issue. As part of the previous user fee reauthorization cycle, Congress required the FDA to publish a report on the servicing issue. That report came out in May of 2018.

Legislation was also introduced in that congress, H.R. 2118 by Costello and Peters, which would have required servicers to register with the FDA, report MDRs, and have a complaint management system.

I fully expect that we’ll continue to engage Congress on this issue. I believe that educating Congress about the serious risks being presented to patients and to healthcare providers by unregulated third-party servicing will motivate Congress to act.

Dennis: Like Peter said, it’s all about educating Congress about the lack of third-party standards and the safety risks that creates.

There are a lot of third-party service providers out there that perform work on behalf of OEMs, so they’re considered authorized service providers. When that happens, they fall under the quality system of the OEM that they’re performing the work for.

Typically, within our quality system, we would look at supplier quality. We would perform audits on them. We would make sure that their training is to the same standards that we would expect as an OEM. They would have the latest revision levels of software schematics, antivirus updates, and the like.

So, the extension of the FDA umbrella falls through the OEM to the third-party service providers when they’re performing authorized work. When they’re not providing authorized work, as Peter said, really there are no guidelines, requirements, or standards.

Then you get into the questions about what parts are being put on? What’s the training and background of the person performing the work? What documentation are they providing, if any? All those things come to play and are really some of the big red flags that we see on our end.



Why is this topic particularly important right now?

Dennis: With regards to Right-to-Repair, there’s been increased efforts to expand service rights to third parties. These efforts mainly came from the consumer goods side, but third-party service providers who participate in the medical device space obviously jumped on that bandwagon.

This was heightened during COVID. There was some publicity suggesting that OEMs didn’t have enough bandwidth to service the ventilators required to support patients with COVID-19. We’ve unequivocally demonstrated that that is not the case and not the truth, but that triggered a lot of the recent discussion.

Peter: From our perspective, this is an important issue because we’re talking about complex devices involved in diagnosis or treatment of serious medical conditions. Making sure that these devices always perform safely and effectively for their intended use is our top priority. Action is long overdue.

Now, the other side of this issue has been very aggressive lately in pursuit of their business interests. They had their own legislation introduced in the previous Congress that would have compelled manufacturers to turn over intellectual property to their unregulated competitors. And now there are a number of Right-to-Repair bills at the state level. Many of these bills are broadly applicable to all sorts of consumer goods, but there’s also a number of these bills that are medical device-specific.

If these bills were to pass, it would be disastrous for patient safety and device performance. We’re talking about having to turn over company confidential information–intellectual property– to unregulated competitors who are not required to have quality or safety controls in place or even make themselves known to the FDA.

Uncontrolled use of proprietary, highly technical service materials by unregulated businesses could lead to improper servicing of a medical device, dramatically increasing risks to patient safety, device performance, and cybersecurity. Safe and effective medical device servicing is not merely acquisition of certain documentation or materials—it is the implementation of and adherence to a set of policies, practices, and procedures which consistently return the device to a state of safe and effective operation.

Open distribution of schematics, wiring diagrams, software, what have you, would be very damaging to the safety and performance of these devices and would have serious implications for ongoing competition and innovation in this space.

Dennis: Even if the schematics and testing equipment was provided, if the third-party service company isn’t trained on how to use the product, how to troubleshoot a product, how to ensure that the product is up in a working condition, the product might not be repaired in a way that’s expected. That’s one of the big concerns.



What advice would you give healthcare providers who are considering working with an unauthorized service company?

Dennis: Healthcare providers should ensure that any service provider has a quality system in place and they should ask a lot of questions to confirm that. Providers should ask:

  • How do they ensure that the repaired product meets performance requirements?
  • Do they have an engineering staff or team that looks at reverse engineering and creating service procedures in a controlled manner?
  • What are their training requirements and how do they stay updated in a very dynamic technological market?
  • How are they keeping products updated to recent software revisions or antivirus software requirements?
  • How do they ensure parts equivalency when they purchase or replace a part?
  • How quickly do they respond to emergency calls?
  • And will they assume liability for the products’ performance after they perform the repairs and the work is completed?

If I was evaluating a third-party service provider to work on my hospital equipment, those are some questions I would want answered before I engage with them.


Imaging Wire Q&A: Looking Forward to SIIM21

Posted on by Jake Fishman

With Christopher Roth, MD, MMCI, CIIP
Duke University, Vice Chair of Radiology for Clinical Informatics and IT
Duke Health, Director of Imaging Informatics Strategy

Like many in healthcare, informatics professionals just made their way through an unprecedented year that included plenty of imaging problems to solve and lessons to learn, but very few opportunities to meet with their peers and improve together. That’s about to change at the upcoming SIIM21 annual meeting, and there’s going to be a lot to talk about.

In this Imaging Wire Q&A, we sat down with Duke’s Christopher Roth, MD, to discuss what he’s expecting at SIIM21, what makes this year’s meeting different, and how informatics professionals can make the most of it.


Tell us about your work and what you do at SIIM.

At Duke University, I’m the Vice Chair of IT and Informatics in the Department of Radiology and the Director of Imaging Informatics Strategy for the health system.

At SIIM, I chair the Annual Meeting Program Committee and I’m currently a chair of the HIMSS-SIIM Enterprise Imaging Community.

I’m also on the RSNA Informatics Committee and the ACR Informatics Commission.


What can we expect from the SIIM21 Annual Meeting?

SIIM21 will be a virtual meeting held from May 24th to the 27th.

The first day will be kicked off by Dr. Kimberly Dyan Manning, Associate Vice Chair of Diversity, Equity & Inclusion at Emory University, followed by strong didactic lectures, interactive roundtables, research and applied informatics sessions, and some Global Track sessions to close out the day.

Day two will follow a new format. In the past, it’s been a continuation of those kinds of educational sessions, but the second day of this year’s meeting will focus on the interface between industry and the people on the clinical frontlines. It’s going to focus very hard on getting people together to fuse ideas and learn from each other .

The second day will kick off with a keynote address by Dr. Jonathan Rothberg, a serial entrepreneur, and co-founder of companies such as 4catalyzer, Butterfly Network, and Hyperfine.

In addition, we’re going to have roundtables, but they won’t just be clinical leaders describing problems and imaging vendors at the table participating, it’s going to be a joint conversation. For example, an imaging vendor would partner with clinical thought leaders to approach a difficult problem and then discuss it with the community.

We’ll also have #AskIndustry Sessions where a thought leader will actually present a difficult problem to three or four industry vendors to get their perspectives on it. The point of this is to really attack the middle ground, the interstitial space between industry and users who could be physicians, scientists, analysts, or the technological staff that we all employ to take care of patients.

Day three will start with a keynote address by Harold F. Wolf III, President & CEO of HIMSS, and will largely focus on the breadth of imaging informatics specialties around the health system. So, as I said, that would include pathology, dermatology, obstetrics, cardiology, ophthalmology, and different specialties who create multimedia. Those groups are going to share their successes and their failures.

The highlight and the grand finale of day four will be the Samuel L. Dwyer Lecture, named after one of the real pioneers in imaging informatics, and presented by Dr. Adam Flanders, Professor of Radiology & Rehabilitation Medicine and Enterprise Vice-Chair for Imaging Informatics at Thomas Jefferson University.


Going back to the second day of SIIM21, what are the types of roles that would represent the vendor and frontline sides of these discussions?

The breadth of people and roles at SIIM are what makes it special. We have everyone from the frontline entry-level analysts at our hospitals up through chief technology officers. We have very early-stage members in training who are still residents all the way up to CMIOs.

The second day isn’t going to be so role-focused as it will be expertise-focused. You will hear from some really terrific speakers with engaging personalities and subject matter expertise in the areas that SIIM is interested in sharing. That could even be someone who lives on the procurement side, let’s say, who can speak to a successful RFP as opposed to an unsuccessful one.


Does SIIM21’s virtual format create any other opportunities to do things differently?

Yes. This meeting will be held under the theme of ‘Connecting Without Boundaries,’ supporting one of SIIM’s strategic goals to broaden its international reach, and the virtual format will definitely support this goal. Many of the meetings that we tend to go to have a distinctly North American flavor, and that’s really not the way we should be working.

We developed a global imaging track where thought leaders from around the world will bring experts from their own neck of the woods to share their approach to problems. We’ll have thought leaders from six geographies – South America, Europe, Africa, Middle East, Asia, and Australia – who will have their own opportunities to lecture and meet before, during, and after SIIM21.

The intent of those meetings is to facilitate collaborations between the regional experts and educate SIIM members who just don’t get the opportunity to learn how other parts of the world approach difficult problems.


Are there certain things that you’ve learned at previous meetings that have been particularly helpful in your day-to-day job?

This happens at every single meeting.

When I go to these meetings, it’s almost a joke with my team at Duke that I will come back from SIIM with a wish list of four or five new things. It’s almost an every-year thing that I’ll see something that I get psyched about and I’ll bring it back. We’ll start talking about these ideas and employ several of them over the course of the year.

In addition to that, I go to see the industry partners because I want to know what they’ve been doing over the past year. They’re busy getting new customers, they’re busy doing new development, they’re making new inter-corporate partnerships. I want to know who they’re working with, what they’re doing, and if there’s anything that is beneficial to me.

I love to be able to say to them, “I think this feature is really, really cool. If you did this it’d be even cooler. Can you do that, or can we do that with you?” Really advocate for the things that we need to have in the imaging space to be successful.

In addition to that, SIIM is where I do my shopping. If I know I’m in the market for something, SIIM is probably the first place I’m going to go look for it, because it’s a very practical meeting. There are other industry meetings that are humongous and difficult to navigate. SIIM is a more intimate, practical place to have some of those difficult conversations.


What should people planning to go to SIIM do to make the most out of this year’s event?

To have success in any in-person meeting you have to plan ahead. You need to know when the sessions are going to be and make sure you block off the time to be there.

It’s the same thing with a virtual meeting. You have to look ahead and be sure you’re not going to be doing a biopsy at the time that you’re supposed to be attending a session.

I would also think about your infrastructure, the tools that you have, the applications you use and if there’s any problems that you may be able to improve or solve at SIIM21. Be prepared to have challenging discussions with vendors.

Additionally, I would take a look at your imaging strategic plan and see what gaps there have been over the past year and a half as COVID has consumed so much of our day-to-day activities. So, there’s probably a lot of opportunity in your imaging strategic plan that you can catch up on and the industry partners that you would be engaging have been developing as well. You may have a lot more to learn than in typical years.

I would make time to visit the vendor booths and have individual meetings or group meetings with thought leadership from other hospitals who are all facing the same thing. I know that there are a few of us who all share the same viewer, the same vendor-neutral archive, the same image sharing platform. There’s nothing against getting a group of you together to go talk to that group of corporate entities to provide a little more oomph to the things that you’re asking.

I would also recommend talking to your clinical colleagues if you’re in the hospital. Find out what they like, what they don’t like, and take those things to those corporate partners to see if anything can be done about them.


What’s the type of person that should be going to SIIM21 but doesn’t realize that it applies to them?

I’ll steal an anecdote from a dermatologist, Veronica Rotemberg, from Memorial Sloan Kettering that I heard recently. She is essentially doing dermatology informatics, but she had no idea that she was doing it.

The people that should be coming to SIIM are the people that work in imaging, IT and clinical quality improvement and process improvement, people who are trying to handle the integrations of their images with the electronic health records. Those are informaticists and they oftentimes don’t even know that they are.

It could be derm, it could be path, it could be cardiology, it could be a number of different specialties. People that care about clinical care, care about integrating the images that they’re capturing and longitudinally following the problems that their patients bring to them. These are people that are doing imaging informatics but maybe haven’t been a part of SIIM before. And really, those are the kinds of people that we need to be engaging.

Similarly, there are industry and researchers and clinical people, like myself, who are doing artificial intelligence, who are creating algorithms and trying to deploy packages and get them into the point of care. Maybe they’re in pathology. Maybe they’re in dermatology. Maybe they’re looking at wounds on a smartphone using AI or they’re looking at a mole and evaluating whether or not its cancer.

Those are people that really have a home at SIIM but they just may not have gotten exposed to it yet. That’s really where the enterprise imaging community comes in.

That’s where a lot of SIIM’s educational outreach and liaison committee work comes in. We are trying to reach these people and help them realize that SIIM is the right place to come when you are looking to interact with and learn from people who face the same day to day challenges.


What are you most looking forward to at SIIM21?

What I look forward to most at every single SIIM meeting is seeing my friends. It’s quirky. It’s kind of cliché and folksy, but SIIM really has created a community of people who all see eye to eye. We’re all seeing the same problems. We’re all actually very similar type personalities. And it’s a special team to feel a part of, because it’s people and conversations that resonate with me.

The longer I’m at SIIM, the more I learn from them. I learn a lot from the other docs like me around the country and around the world who do some of this. And the deeper I go, the more I realize there’s a wealth of knowledge on the industry side that I’m still learning.

Imaging Wire Q&A: Nanox.AI Thinks Big

Posted on by Jake Fishman

With Zohar Elhanani
Nanox.AI, General Manager

The role of imaging AI continues to grow, as radiology workflows increasingly utilize these tools to prioritize patients and support diagnoses. This already represents a big change for healthcare, but it could be just the beginning of imaging AI’s far greater public health evolution that extends well beyond the radiology department and could change how and when many diseases are diagnosed.

In this Imaging Wire Q&A we sat down with Nanox.AI General Manager, Zohar Elhanani, to discuss Nanox.AI’s view of how imaging AI is helping healthcare today and how AI’s role in public health could be much bigger than many of us imagine.


You had a front row seat during two key periods in the medical imaging industry’s evolution. What are the major themes that connect those periods and how are they shaping imaging’s future?

I started my career in medical imaging right when we were shifting from analog to digital. My company’s products moved images between healthcare facilities, radiologists, teleradiologists, and referring physicians. That was step one of the digital evolution.

Fast forward 20 years, we’re now seeing a digital image volume evolution, as medical images are being produced, analyzed, and stored at a massive scale. Volumes have grown so much it’s been hard for radiologists to keep up.

This digital image volume growth also made imaging AI possible, which is becoming a larger part of the radiology workflow, and helping radiologists interpret images as efficiently and accurately as possible.

So for me, it’s gone full circle, from the start of the digital imaging evolution and into the imaging AI evolution.


How do you view the next phase of the AI evolution?

AI is already becoming a driving force in medical imaging diagnostics. It’s becoming commonly used across healthcare facilities and providers, and not only in radiology. This is really a tectonic shift in healthcare.

The COVID pandemic and the focus on clinical and revenue cycle efficiency has made AI much more than just a buzzword. AI is actually becoming more focused on validated use cases and generating real tangible ROI.

For Nanox.AI, as an medical imaging AI pioneer, this has been a journey. We initially targeted detection of low prevalence findings, triaging acute conditions, and improving turnaround times for radiologists. That was a very good entry point. It was a valuable way to substantiate how AI can detect abnormalities and prioritize reads.

During our AI journey, we also realized that although these are valuable use cases, they don’t necessarily always present a clear ROI. As part of our evolution, we’re now looking to expand and we’ve already introduced products targeting larger populations at scale, focusing on high prevalence, chronic conditions that have not been detected.

We feel that promoting preventative care for treatable illnesses will expand AI to broader populations and more use cases, while supporting the shift from fee-for-service models to value-based care.

We’re committed to population health AI. We’re building out our population health product offering and roadmap and we’ll introduce more solutions over time, in addition to our coronary calcium scoring and vertebral compression fracture solutions. We think that’s a path for the future and an area that AI can play a bigger role.


We don’t hear AI companies talk about population health very often. Can you tell me more about how AI supports population health?

The pathway to value-based care involves making healthcare systems more efficient and offering patients preventative care, rather than waiting for undetected diseases to get worse.

Our population health solutions focus on catching diseases that have the highest rates of morbidity and mortality. Coronary heart disease and osteoporosis are silent killers, and they get worse over time.

Radiologists don’t always note or look for these findings. Generally, someone walks in for a specific condition, like a broken rib, and incidental findings are not necessarily caught or communicated.

Our solutions yield more information from existing CT scans and EMR data. By applying these algorithms, we can spot undetected diseases and alert physicians to initiate a pathway to care that improves patient health and reduces costs for healthcare systems.

This is where the whole shift to value-based care is heading and we think that’s an area where AI and Nanox.AI could play a bigger role.


How does AI economics work for population health programs?

So obviously there are two sides.

First, there’s a revenue cycle side that involves the actual income from providing medical care. And obviously, in value based care systems, these are capitated programs.

Second, there’s the cost reduction side, achieved through early intervention and avoiding expensive care for under-treated and non-treated conditions.

So the idea is to create enough incentive for both payer and provider to look at AI as a way to reduce cost but also manage patient risk.

The radiologists need to be motivated and incented to identify and confirm these findings. So that’s one area that needs to be looked at. The medical imaging AI industry has been struggling to find the right way to make radiologists more motivated to look into findings that are different from the purpose of the original study. Nanox.AI is always at the forefront of finding solutions and we aim to do that here too.


Who would be involved in evaluating and implementing AI-based population health initiatives?

In our population health projects, we generally work with chief revenue officers and chief population health officers, who look at the breadth of cost and quality of care across their population. The two have to go hand-in-hand. What is the cost and what is the quality?

There also needs to be buy-in at the point of care by the radiologist. That’s where the finding is detected. But in terms of the program as a whole, it’s orchestrated by the chief revenue officer, chief population health officer, and the chief medical officer. They prescribe the pathway to care and define what needs to trigger that pathway based on AI-detected incidental findings.


What’s the best way for these population health executives to involve the radiology department?

There needs to be some kind of economic benefit for the radiologist to take action on these findings. One incentive is obviously just quality of care and the breadth of the report itself, but a financial incentive is also required. That’s part of the equation and that’s something that needs to be sorted at the IDN level between the payer and the provider as part of a value based care paradigm.


When population health programs use imaging AI to identify incidentals at scale, follow-up management becomes really important. What’s the best way to do follow-up management in a program like this?

The emphasis here has to be on establishing pathways to care from the point that the AI and the radiologists confirm a finding. And I think that’s again part of the shift to a value-based care paradigm where these findings make their way to actual treatment, which reduces costs and improves patient care.

That’s exactly where we’re focusing our efforts in order to make sure that a finding doesn’t just stay there in the report itself. It actually triggers a call for action to take the finding to the right stakeholder at the provider level or beyond.

That’s a critical part of it. What is the pathway to care and what are the incentives around that pathway under a value-based care program or plan?


Would population health programs achieve any benefits from AI that they weren’t expecting?

Definitely. We’ve run our own tests on data to compare what’s written in the EMR and patient records, and we found many new findings that did not exist. And that’s simply by running algorithms retrospectively on existing data and substantiating the value of AI.

So definitely, the response has been very, very favorable to the fact that things go missed and are under-reported and there’s value there.

Now, the question is how to deploy that at scale and how to create the actions and the pathway to care from these detections?


Do you have any advice for healthcare systems considering using AI to support their own population health efforts?

One of our larger customers recently shared with us that his three priorities for AI-enabled population health are improving patient care, reducing liability risk, and adding financial value.

I completely agree. Combining the improvement of patient care, the financial value for the system, and reduction of liability risk is critical.

I think that’s something the industry as a whole is still looking for. How do you substantiate the value of AI in terms of the financial benefit? How does it really improve patient care as a whole? And specifically, if we look at triage solutions, how do they really impact low prevalence acute findings versus what we see in population health with high prevalence chronic illnesses?

That’s the goal for this whole pathway that we’re discussing. AI for population health isn’t here to replace referring physicians or regular checkups. It’s here to serve as kind of an early warning signal for chronic disease. That’s really the idea, serving as a safety net for any finding that exists and is not detected or is under reported. It’s another layer that would augment whatever is done by the primary care physicians or any ongoing radiologist interpretations.

Long term, obviously it provides better cost structure for the entire system and offers comprehensive preventative care for the patient. And as I said earlier, we won’t be simply looking at a handful of conditions. It will involve a longer pathway to covering many incidentals and making sure that they’re all accounted for in terms of at least knowing that they’re there and considering potential care pathways to ensure that nothing is ignored or under-treated.

As a whole, it’s another layer of detection that it doesn’t currently exist. That’s how we see AI playing a big role in the population health domain.

Imaging Wire Q&A: Arterys’ AI Journey

Posted on by Jake Fishman

With John Axerio-Cilies, PhD
Arterys CEO & Co-Founder

It’s been quite a decade for AI and cloud technology in the healthcare space, with some major milestones and learning moments along the way.

Arterys had a front seat view for many of these milestones, as the industry’s first cloud-native imaging AI developer and one of the only companies that serves as both an AI developer and a multi-vendor AI marketplace platform provider.

In this Imaging Wire Q&A we sat down with Arterys CEO and co-founder, John Axerio-Cilies, PhD, to discuss medical imaging’s AI and cloud evolution and how Arterys works with its Center of Excellence partners to make AI real.


Arterys’ 10-year anniversary makes you imaging AI veterans. What are some of the key milestones you’ve witnessed during this journey?

When we started back in 2011, imaging AI as we know it wasn’t really a thing.

At the time you could say launching Arterys was a leap of faith, based on our vision for patient-driven insights, data-driven medicine, and a commitment to the cloud.

AI and machine learning have been around for decades, and some imaging vendors began exploring machine learning-like approaches in the early 2000s. However, back in 2011, the forward-looking part of the healthcare industry was mainly focused on precision health and big data. Those were the key buzzwords and cloud wasn’t even part of the equation.

It was still extremely early for cloud, especially in healthcare. Early on, we even had an IT leader at a major academic medical center tell us that they would “never do cloud.” It took 10 years, but now everyone who’s educated on the subject recognizes cloud’s benefits, even the IT team from that same academic medical center. At that same center if it’s not cloud enabled, they will not consider it.

Imaging AI as we know it primarily got its start because of deep learning, beginning with a key paper that came out in 2012, and leading up to the current surge in industry interest that started in 2017. We’re now seeing the AI hype curve slow down into more of a reality. There haven’t been any monumental events that immediately changed the way people in healthcare think about AI. Instead, imaging AI is slowly going through the expected adoption cycles, making its way from early adopters and towards the laggards.


How often do you come across radiologists who are concerned that AI will endanger their job?

I rarely hear concerns that AI could eliminate radiologists’ jobs among the physicians who I work with, but these folks are already interested in AI. Still, there is certainly a pool of radiologists who are concerned that AI might replace them.

However, this happens in every industry and it’s still very early in AI’s evolution. I think it’s going to be decades before AI would realistically challenge radiologist’s current jobs, plus radiologists’ jobs are going to keep evolving along with AI.

It’s been far more common to see radiologists realize that AI can accelerate their workflow and help them day-to-day, and we expect that trend to continue.


How has Arterys’ own AI platform evolved?

In the last few years we’ve opened up our platform to support more and more use cases, including more modalities and more top service lines. The most notable expansions have been in cardiovascular imaging, neuroimaging, women’s health, and within acute and X-ray service lines.

We had to add more functionality to our underlying platform in order to support this portfolio expansion, opening up our platform to the point that it’s almost self-serve. By opening the platform we’ve seen expanded adoption from not only our 40+ AI vendor partners but also healthcare institutions using the Arterys platform to deploy, share, and refine their own AI tools, fully in their control.

That’s exciting because we want to have an entire ecosystem to support early innovators, researchers, and academic medical centers. Even larger IDNs have strategic initiatives around AI and are funding researchers to develop AI models that they want to integrate.

We want to help support that evolution and push these models from research into clinical practice and to ultimately become commercial products. We’re helping manage that too, because we have folks that can help for regulatory support, commercial go-to-market support, and the entire commercialization trajectory.

We also continue to refine these products and their implementations, thanks in part to our Center of Excellence program.


What inspired you to create Arterys’ Center of Excellence program?

The imaging industry lacks clinical evidence, and the data to prove the value proposition of products. Healthcare marketing folks say all these grandiose statements but when you double-click on these statements, there’s often not a lot of data to support them.

This is also where most AI providers and users are lacking, and it’s the reason we created the Center of Excellence program.

Through the Center of Excellence program, we work with our major medical institution customers to go one step deeper and make sure their AI adoption is happening and it’s actually impacting whatever needs to be impacted. These improvement targets usually include patient outcomes and efficiency, so we’re often trying to create an infrastructure that solves both of those problems.

With our Centers of Excellence we translate actual data to show clients how we were able to accomplish their AI goals because we worked with them to change their workflow and helped them guide behavioral changes.

AI success is so much more than a working product. People talk about AUC, sensitivity, and specificity, but that’s less than 5% of the problem. You still need to have the infrastructure and the clinical workflow and the behavioral change to adopt this stuff.


Who would be involved in a Center of Excellence partnership?

Every partnership starts with clinical users, but the things we measure and improve would be very different depending on the specific product, its users, and the organization.

For example, our X-ray product targets ED physicians and to a lesser extent radiologists, giving
them a tool to quickly triage, treat, or discharge patients. We’d work with that partner to confirm that the X-ray solution actually improves outcomes and helps treat patients faster.

It’s very different with our cardiac product, which is used by cardiologists and radiologists, and is absolutely required for diagnosis. With these partners, we’d work with them to help confirm that the cardiac product works as needed.

In any scenario, the clinical users would be a starting point but we’d also work closely with senior leadership like CIOs, CMIOs, and CFOs to make sure institutional goals are being met.

We’re actively looking for more Center of Excellence partners, especially partners in the neuroimaging and in the oncology space.

How are your Center of Excellence partners’ improvements communicated?

We’ve done a good job making this as non-intrusive as possible. Because we’re completely cloud-based we can usually integrate with our partners in a few minutes, and we can also collect more detailed clinical information for partners interested in understanding their patients’ pre- and post-imaging pathways.

We provide Center of Excellence partners with all outputs from each patient session to any of their imaging IT or EMR platforms, allowing them to monitor and analyze their progress.

Can you tell me about your most successful Center of Excellence partners?

The most successful Centers of Excellence really care about making AI real and they are willing to dive in, run assessments, and perform trials to make sure that we’re actually impacting whatever we set out to improve. UMass Memorial Health Care here in the U.S.A. and Centre Hospitalier de Valenciennes in France are a couple great examples of sites who are doing this.

These most successful Centers of Excellence truly had clinical pain points that hurt bad enough for them to make solving them a priority. For example, we’ve had some partners who kept ED patients waiting for X-ray results for hours or had to discharge patients without their results. That’s a massive pain point and it’s enough to make hospitals get serious about finding solutions.

The opposite of that is hospitals who say, “oh, let’s get AI in here” but aren’t sure about what’s their clinically unmet need or if they even have one. The fact that these hospitals don’t know where they can improve suggests that they have a lot of ways to improve, but they have to identify these challenges and commit to addressing them before they are ready to become a Center of Excellence.

What should healthcare institutions ask themselves when considering being a Center of Excellence?

The first thing they should ask themselves is if they are committed to making AI real. I think that’s a really important question. Because if they are not, and they’re not truly invested in actually helping the patients or improving workflow, that’s not an ideal candidate. I don’t care about marking an AI adoption checkbox. What I care about is working with our partners to make their AI adoption impactful.

Potential partners should also understand their goals and confirm that they are ready to work together to achieve those goals, because many improvements come from outside of the software, and continuous improvement is a collaborative process.

About Arterys

Arterys is the market leader and the world’s first internet platform for medical imaging. Its objective is to transform healthcare by transforming radiology. The Arterys platform is 100% web-based, AI-powered, and FDA-cleared, unlocking simple clinical solutions.

Winners Announced for 2020 Imaging Wire Awards

Posted on by Jake Fishman

The Imaging Wire is thrilled to announce the winners of the 2020 Imaging Wire Awards, honoring this year’s most outstanding contributors to radiology.

The following Imaging Wire Award winners were nominated by their peers and selected by a panel of judges for their efforts to evolve radiology and improve the lives of clinicians and patients:


COVID Hero: Byron Christie, MD; Associate Chief Medical Officer of Integrations, Radiology Partners

When the pandemic hit, Dr. Christie and nine radiologists from RP’s SEAL team traveled across the U.S. to provide care in hard hit regions. After recovering from a COVID-19 infection that he contracted while treating patients in Florida, Dr. Christie increased his efforts to fight COVID-19 through his work at RP, continued plasma donations, and by educating medical students.

Diagnostic Humanitarian: Daniel J. Mollura, MD; President and CEO, RAD-AID International

Dr. Mollura is the Founder and CEO of RAD-AID International, a nonprofit organization dedicated to expanding radiology care to underserved and resource-poor communities. Over the last 12 years, Dr. Mollura grew RAD-AID to nearly 14,000 members serving over 80 hospitals in 35 countries. Among many accomplishments this year, RAD-AID’s residency program in Guyana will graduate its first class of radiologists.

AI Activator: Jon T. DeVries, CEO; Qlarity Imaging

Under Jon’s leadership, Qlarity Imaging has made significant progress developing the company’s QuantX software, which integrates images from multiple modalities to assist radiologists in the assessment and characterization of breast abnormalities. DeVries continues to expand QuantX’s capabilities and market reach with an innovative approach to product development and partnerships.

Burnout Fighter: Marla B.K. Sammer, MD; Associate Professor of Pediatric Radiology, Texas Children’s Hospital

Faced with Texas Children’s Hospital’s massive imaging volume growth, Dr. Sammer introduced a new initiative to optimize workflow, balance distribution across teams, and improve radiologists’ workdays. These changes reduced Texas Children’s average turnaround for X-rays by 25% and other modalities by over 27%, while helping its radiologists reliably predict their workday, fostering a sense of fairness and control, and reducing burnout.

Insights to Action: Syed Zaidi, MD, MBA; Associate Chief Medical Officer for Integrations, Radiology Partners

Dr. Zaidi has consistently tackled imaging waste throughout his career, participating in Choosing Wisely and serving as a leader in the ACR’s Imaging 3.0 initiative. Dr. Zaidi also developed a utilization management program at his local hospital to limit unnecessary chest CT scans for pulmonary embolism, while helping to roll out a best practice recommendations program across Radiology Partners.

Continued Care: Jinghong Li, MD, PhD, University of California San Diego

Dr. Jinghong Li is an attending physician and associate professor specialized in pulmonary diseases and critical care at University of California San Diego. While caring for COVID-19 patients in UCSD’s ICU, Dr. Li also worked with engineers and scientists to develop a wearable ultrasonic patch to allow continuous bedside ultrasound monitoring. This patch would alleviate infection control concerns associated with manual bedside imaging, while helping predict respiratory failure due to COVID-19 pneumonia.

Cornerstone: Karen Holzberger, SVP and GM; Healthcare Diagnostics, Nuance Communications

As the leader of Nuance’s healthcare diagnostics team, Karen’s top focus is to drive innovations that advance the practice of radiology. That was on display this year, as Ms. Holzberger led the development of new capabilities that prioritize and add insights to COVID-related exams, delivered on Nuance’s promise to enable “AI at scale” through the Nuance AI Marketplace, and continued to enhance PowerScribe One.

Diversity & Inclusion: Kristina Elizabeth Hawk, MD; President, Matrix East Pod A, Radiology Partners

Dr. Kristina Elizabeth Hawk is a founding member of the RP Belonging Committee, which designs tracks and programs intended to amplify the roles of minority groups in the practice. Dr. Hawk has led outreach to diverse radiology residents and fellows, and serves on the ACR’s Commission for Women and Diversity, Stanford’s Radiology Diversity committee, and Ambra’s #Radxx board.

Congratulations to this year’s Imaging Wire Award winners and nominees, who’s efforts to elevate radiology is truly inspiring. Also, thanks to this year’s amazing judges and everyone who nominated these very deserving imaging professionals!

The 2020 Imaging Wire Award judges include: Bill Algee of Columbus Regional Hospital, Dr. Jared D. Christensen of Duke University Health, Dr. Keith J. Dreyer of Partners Healthcare, Dr. Allan Hoffman of Commonwealth Radiology Associates, Dr. Terence A.S. Matalon of Einstein Healthcare Network, and Dr. Syam Reddy of University of Chicago Ingalls Memorial.



Imaging Wire Q&A: Evolving With Hitachi VidiStar

Posted on by Jake Fishman

With John Stock, MD, FACC
Pediatric Cardiologist
Pediatric Cardiac Care of Arizona

The role of imaging in pediatric cardiology has evolved tremendously in recent years, so in order for these practices to operate successfully, their PACS systems have to evolve at the same pace. That can be easier said than done, but it’s exactly what happened with Pediatric Cardiac Care of Arizona and its VidiStar PACS system.

In this Imaging Wire Q&A, we sat down with Dr. John Stock of Pediatric Cardiac Care of Arizona to discuss the evolving role of imaging in his practice, how Hitachi’s VidiStar PACS has evolved along with it, and what pediatric cardiology practices should look for in their own PACS systems.


Tell us about your practice and how you use imaging.

We perform and interpret approximately 3,000 pediatric studies per year. I interpret all the cardiac ultrasound studies independently after reviewing and confirming measurements.

From there, VidiStar generates a report that is often faxed to the referring physician. The studies are digitally archived on our server and in the cloud, with reports maintained in the electronic medical records. We follow all appropriate use guidelines and quality assurance initiatives, and we are an IAC accredited lab.


How has your practice been impacted by the COVID-19 pandemic?

COVID-19 has definitely affected my practice, but not how some might think. We experienced a 20% to 30% drop in patient volumes during the shutdown’s peak months. There appears to have been a rebound, as children and adolescents returned back to their pediatricians, schools, and sports.

Pediatric patients with congenital heart disease have a higher risk of complications. As a result, we are cautious in our follow up and in some cases evaluating for possible findings related to COVID-19. There is also a subset of the disease called multi-inflammatory syndrome of childhood (MIS-C), which can result in decreased ventricular function and coronary artery dilation. This requires prompt management and follow-up.


You’ve been using VidiStar for quite a while, can you share how you use it?

I use VidiStar on a daily basis for interpreting and completing reports on my pediatric, adult congenital, and fetal cardiology patients. This includes looking at the study as the sonographer performs an evaluation, followed by an independent review with measurements confirmed by the VidiStar reporting package, and then saving the study to our server.


How has VidiStar changed over time?

VidiStar has come a long way since Hitachi acquired the platform two-plus years ago, turning it into a system that is affordable, user friendly, and can support the simplest and most complex pediatric cases.

I’ve benefited most from the improvements to VidiStar’s pediatric reporting package. At first, VidiStar’s pediatric package was very basic and utilized an adult format, requiring me to do a lot of work outside of the platform. Kids are not small adults. They have their own complexities. The reports need to reflect the variation in anatomy that can occur in congenital heart disease.

Hitachi came in, made a commitment to pediatrics, and VidiStar now fits the needs of most pediatric cardiology practices. In just the last two years, the pediatric package improved many measurement parameters and Doppler measurements, which allow me to perform comparisons over time.


What advice can you share for pediatric cardiologists evaluating new PACS systems?

Any independent pediatric cardiology provider considering a new PACS system should evaluate how each system would meet their clinical and workflow needs and whether it fits their budget.

Most important for me clinically, is the ability to track changes over time and knowing that I can be confident when I send out reports to some of the best centers in the country. The reports must also look professional, with appropriate identification of pertinent impressions, as well as documentation of pertinent positive and negative findings.

From a workflow perspective, it is also very important that the PACS system interfaces well with the electronic medical records, and that it’s easy for both the sonographer and the physician to use.

It’s also crucial that the PACS system works consistently. By that, I mean that the system always works and its output is reproducible and consistent over time, which isn’t guaranteed with some platforms.

Pediatric cardiologists should also look for reporting packages that clearly document Z scores and Doppler velocities, which are necessary for appropriate billing. Incorporation of 3-D and strain will also be necessary going forward.

About Pediatric Cardiac Care of Arizona

Based in Tempe, Arizona, PCCA’s mission is to partner with patients, families, and referring physicians in order to provide excellent outpatient cardiac care in an environment of trust, openness, and professionalism.

Dr. John Stock has cared for patients with congenital and acquired heart disease for over 20 years, after receiving his medical degree from Upstate Medical Center, completing his pediatric residency at Phoenix Children’s Hospital, and undergoing fellowship at Oregon Health Sciences University.


About Hitachi VidiStar

The Hitachi VidiStar Platform gives physicians and healthcare providers the ability to read and interpret diagnostic studies over the internet for timely interpretation, improved patient diagnosis, clinical decision support, and advanced patient data analytics and notification.

VidiStar provides highly customizable infrastructure for multi-modality viewing, reporting, and analytics while interfacing with existing IT systems for one seamless solution.

Get every issue of The Imaging Wire, delivered right to your inbox.

You might also like..

Select All

You're signed up!

It's great to have you as a reader. Check your inbox for a welcome email.

-- The Imaging Wire team

You're all set!