Autonomous AI Milestone

Just as the debate over whether AI might replace radiologists is starting to fade away, Oxipit’s ChestLink solution became the first regulatory-approved imaging AI product intended to perform diagnoses without involving radiologists (*please see editor’s note below regarding Behold.ai). That’s a big and potentially controversial milestone in the evolution of imaging AI and it’s worth a deeper look.

About ChestLink – ChestLink autonomously identifies CXRs without abnormalities and produces final reports for each of these “normal” exams, automating 15% to 40% of reporting workflows.

Automation Evidence – Oxipit has already piloted ChestLink in supervised settings for over a year, processing over 500k real-world CXRs with 99% sensitivity and no clinically relevant errors.

The Rollout – With its CE Class IIb Mark finalized, Oxipit is now planning to roll out ChestLink across Europe and begin “fully autonomous” operation by early 2023. Oxipit specifically mentioned primary care settings (many normal CXRs) and large-scale screening projects (high volumes, many normal scans) in its announcement, but ChestLink doesn’t appear limited to those use cases.

ChestLink’s ability to address radiologist shortages and reduce labor costs seem like strong and unique advantages. However, radiology’s first regulatory approved autonomous AI solution might face even stronger challenges:

  • ChestLink’s CE Mark doesn’t account for country-specific regulations around autonomous diagnostic reporting (e.g. the UK requires “appropriate reporting” with ionizing radiation-based exams)
  • Radiologist societies historically push back against anything that might undermine radiologists’ clinical roles, earning potential, and future career stability
  • Health systems’ evidence requirements for any autonomous AI tools would likely be extremely high, and they might expect similarly high economic ROI in order to justify the associated diagnostic or reputational risks
  • Even the comments in Oxipit’s LinkedIn announcement had a much more skeptical tone than we typically see with regulatory approval announcements

The Takeaway

Autonomous AI products like ChestLink could address some of radiology’s greatest problems (radiologist overwork, staffing shortages, volume growth, low access in developing countries) and their economic value proposition is far stronger than most other diagnostic AI products.

However, autonomous AI solutions could also face more obstacles than any other imaging AI products we’ve seen so far, suggesting that it would take a combination of excellent clinical performance and major changes in healthcare policies/philosophies in order for autonomous AI to reach mainstream adoption.

*Editor’s Note – April 21, 2022: Behold.ai insists that it is the first imaging AI company to receive regulatory approval for autonomous AI. Its product is used with radiologist involvement and local UK guidelines require that radiologists read exams that use ionizing radiation. All above analysis regarding the possibilities and challenges of autonomous AI applies to any autonomous AI vendor in the current AI environment, including both Oxipit and Behold.ai.

Incidental Evolution

Last week brought a wave of studies that either highlighted how findings in common imaging exams could add value in completely different clinical areas, or showed how incidentals could find a home in established clinical workflows. That might not be welcomed news among the many radiologists who view incidentals as a clinical slippery slope, but it’s another sign that the incidental evolution is gaining momentum.

Left Atrial Dementia Marker – A new JAMA study showed that echocardiographic left atrial function measurements can be used to identify individuals with higher dementia risks, in addition to supporting cardiovascular diagnosis. Analysis of 4,096 participants’ echo exams and 6-year outcomes (75yr avg. age; 531 developed dementia) revealed that lower left atrial function (e.g. reservoir strain, conduit strain, contractile strain, active emptying fraction, emptying fraction) has a statistically significant association with developing dementia (1.43 to 1.98 hazard ratios).

BACs and CVD – A Kaiser Permanente study added more evidence supporting breast arterial calcifications’ value as a cardiovascular disease risk factor. The researchers analyzed 5,059 women’s digital mammography exams (26.5% w/ BACs), finding that women with BACs had a 51% higher risk of developing atherosclerotic CVD and a 23% higher risk of developing any type of CVD over 6.5-years. This is far from the first study to tie BACs to CVD risk, but it came with a high level of credibility (large/observational study, published on Circulation) and generated quite a bit of media attention.

Auto CAC Pathway – A Journal of Digital Imaging study highlighted how coronary artery calcium scores (CAC scores) could be integrated into standard cardiovascular disease (CVD) risk systems, potentially streamlining CAC AI adoption. The researchers used an FDA-cleared AI model (believed to be from Nanox AI) to screen 14,135 patients’ existing CTs (470 who experienced CVD within 5yrs) and then combined their CAC scores with the ACC/AHA’s PCE risk system. The AI-augmented PCE predictions outperformed standard PCE predictions (sensitivity: 57% vs. 53%; specificity: 70% vs. 67%), without requiring additional scans or diagnostic workflows.

Northwestern Follows-Up – A new NEJM study highlighted the impressive results of Northwestern Medicine’s lung nodule follow-up system, which uses NLP to identify suspicious nodules and then initiates a follow-up workflow (prompts physicians, notifies patients, tracks follow-ups). Over 13 months, the system screened over 570k imaging studies, flagging 29k exams for follow-up (77.1% sensitivity, 99.5% specificity, 90.3% PPV), and tracked over 2,400 follow-ups to completion.

The Takeaway
Last week’s batch of studies serve as yet another reminder that common imaging exams could serve broader clinical roles the future, either by creating new risk-based incidental pathways (LA function for dementia; BAC for CVD), catching more undetected incidentals (AI CAC scoring), or by formalizing how incidentals are brought into clinical pathways (e.g. adding CAC to PCEs; leveraging NLP for follow-ups).

Complementary PE AI

A new European Radiology study out of France highlighted how Aidoc’s pulmonary embolism AI solution can serve as a valuable emergency radiology safety net, catching PE cases that otherwise might have been missed and increasing radiologists’ confidence. 

Even if that’s technically what PE AI products are supposed to do, studies using commercially available products and focusing on how AI complements radiologists (vs. comparing AI and rad accuracy) are still rare and worth a closer look.

The Diagnostic Study – A team from French telerad provider, IMADIS, analyzed AI and radiologist CTPA interpretations from patients with suspected PE (n = 1,202 patients), finding that:

  • Aidoc PE achieved higher sensitivity (0.926 vs. 0.9 AUCs) and negative predictive value (0.986 vs. 0.981 AUCs)
  • Radiologists achieved higher specificity (0.991 vs. 0.958 AUCs), positive predictive value (0.95 vs. 0.804 AUCs), and accuracy (0.977 vs. 0.953 AUCs)
  • The AI tool flagged 219 suspicious PEs, with 176 true positives, including 19 cases that were missed by radiologists
  • The radiologists detected 180 suspicious PEs, with 171 true positives, including 14 cases that were missed by AI
  • Aidoc PE would have helped IMADIS catch 285 misdiagnosed PE cases in 2020 based on the above AI-only PE detection ratio (19 per 1,202 patients)  

The Radiologist Survey – Nine months after IMADIS implemented Aidoc PE, a survey of its radiologists (n = 79) and a comparison versus its pre-implementation PE CTPAs revealed that:

  • 72% of radiologists believed Aidoc PE improved their diagnostic confidence and comfort 
  • 52% of radiologists the said the AI solution didn’t impact their interpretation times
  • 14% indicated that Aidoc PE reduced interpretation times
  • 34% of radiologists believed the AI tool added time to their workflow
  • The solution actually increased interpretation times by an average of 7.2% (+1:03 minutes) 

The Takeaway

Now that we’re getting better at not obsessing over AI replacing humans, this is a solid example of how AI can complement radiologists by helping them catch more PE cases and make more confident diagnoses. Some radiologists might be concerned with false positives and added interpretation times, but the authors noted that AI’s PE detection advantages (and the risks of missed PEs) outweigh these potential tradeoffs.

The Case for Operational AI

A trio of radiologists from Mount Sinai and East River Medical Imaging starred in a recent Aunt Minnie webinar, discussing their paths towards operational AI adoption, and sharing some very relevant takeaways for radiology groups and AI vendors.

The Cast – The Subtle Medical-sponsored webinar featured Mount Sinai’s Amish H. Doshi, MD and Idoia Corcuera-Solano, MD (neuro and MSK subspecialists) and East River Medical Imaging’s Timothy Deyer, MD (CMIO and MSK IR), all of whom were involved in evaluating and adopting Subtle Medical’s SubtleMR deep learning reconstruction solution.

Make it Easy – When discussing their AI evaluation criteria, the panelists placed a major emphasis on ease-of-evaluation and implementation, with one noting that “before even having a conversation” he’d have to be certain these early processes won’t be costly or cumbersome (clear process, no new hardware, minimal IT work, no up-front purchases, etc.). 

Why Operational AI – Much of the discussion focused on why the panelists support operational AI, noting that scan-shortening DLIR solutions like SubtleMR:

  • Allow more revenue-generating scans per day
  • Alleviate technologist burnout and staffing challenges
  • Improve the patient experience (especially pediatric)
  • Eliminate re-scans by reducing movement artifacts that occur in long exams
  • Don’t require changes to radiologist workflows
  • Maintain diagnostic image quality
  • Receive less pushback from admins and physicians than diagnostic AI

Evaluating SubtleMR for MSK – Mount Sinai’s MSK SubtleMR evaluation process included comparing standard of care and SubtleMR-enhanced abbreviated MRI exams from 50-consecutive knee MR patients. They found that SubtleMR cut scan times by 50% (13:27 to 6:45), while achieving comparable image quality, artifacts, and diagnostic performance.

Evaluating SubtleMR for Neuro – Mount Sinai’s neuro evaluation process involved comparing SubtleMR and conventional MRI with 10-15 patients for each potential MR sequence. They then reviewed the scans with key stakeholders, worked with the Subtle Medical team to make requested imaging adjustments, and implemented the solution.

SubtleMR Results – SubtleMR’s list of benefits (scan speed, patient experience, patient throughput, revenue) earned it approval from all key stakeholders. Although one panelist noted that some of their radiologists critiqued the enhanced images, the radiologist pushback wasn’t nearly as strong as what they’ve seen in response to diagnostic AI products.

The Takeaway

We cover plenty of editorials about what it takes to drive AI adoption, but feedback from real world AI adopters is still rare, making this webinar particularly useful for AI vendors and adopters. The webinar also states a solid case for SubtleMR and other deep learning reconstruction solutions, even for groups who might not be ready to adopt the kind of “AI” that we usually focus on.

HIMSS 2022 Reflections

Two years after HIMSS became COVID’s first trade show casualty, healthcare’s leading IT conference returned to Orlando with a very post-COVID vibe and a surge in imaging activity. 

Hope you had a blast if you made it to HIMSS, and here’s some highlights in case you didn’t:

The HIMSS Crowd – Unlike the Delta-impacted HIMSS 2021 conference, this year’s event boasted a full exhibitor list and reportedly solid health IT leadership attendance. However, exhibitor staff often appeared to outnumber potential customers on the show floor, prompting conversations about whether HIMSS is evolving into a B2B event and causing some vendors to question where imaging sits on IT executives’ list of priorities. 

The Mixed Cloud – PACS and enterprise imaging vendors continued to ramp up their cloud capabilities and cloud leadership messaging, with nearly everyone agreeing that the future will bring far more cloud adoption. It was also clear that many radiology practices and hospital systems (and even some PACS vendors) are still taking it slow on their path towards the cloud. 

AI in the Aisles – Only a handful of imaging AI companies had booths this year, but it wasn’t hard to find folks from AI startups walking the show floor or in meeting rooms. That’s actually consistent with previous HIMSS conferences, and it makes a lot of sense given AI startups’ limited budgets and the low count of radiology leaders at the show.

AI in the Enterprise – Although we didn’t hear much about all those PACS-based AI platforms / marketplaces that were announced several years ago, AI was positioned at the center of quite a few PACS vendors’ future diagnostic workflow strategies. These strategies still largely focused on integrating third-party AI tools, but several major enterprise imaging players (e.g. Canon, Fujifilm, Siemens) also forecasted a greater future role for their own homegrown AI tools.

The Productivity Press – With imaging growing in volume / complexity at a much faster rate than imaging teams’ own headcounts / capabilities, just about every product message focused on improving productivity and efficiency. HIMSS 2022 saw imaging vendors address this in a wide variety of ways, including remote modality operation, ultrasound AI automation, automated scanner setup, and hanging protocol standardizing (to name a few).

Expanding Ologies – HIMSS also revealed more multi-ology progress as enterprise imaging players better connected their solutions, added new ology-expanding partnerships, and integrated their acquired companies. That said, it seems like the majority of “enterprise” imaging engagements are still limited to radiology, or at least starting there.

Looking Beyond Imaging – A walk around the show floor suggests that healthcare tech is evolving at a much faster pace outside of imaging, with major adoption and technology advances in telemedicine, patient monitoring, at-home and hybrid care, and patient engagement. Although most of these solutions have little to do with radiology right now, these efforts could change how and where many patients get their care, which would have an impact on nearly all specialties. By the way, we have an excellent newsletter about this space for those looking to keep up with these trends. 

The Takeaway

After one year of digital conferences and another year of minimally-attended hybrid events, the bar has been set pretty low for 2022 trade shows. That said, HIMSS had everything that you would expect from a successful post-COVID trade show (plenty of vendors, exciting tech, strong attendance, good vibes), which is a good sign for future events as long as the pandemic cooperates.

Although HIMSS 2022 didn’t necessarily reveal any major focus changes for imaging IT, it did showcase some solid progress advancing the major imaging trends that we’ve seen over the last few years (cloud, AI, productivity, enterprise-expansion), and we’re excited to see what else this year has in store.

NeuroLogica’s Photon Counting CT

Samsung’s NeuroLogica subsidiary announced the FDA 510(k) clearance of its photon counting-based OmniTom Elite PCD, significantly expanding the mobile head/neck CT system’s diagnostic potential and adding to photon counting technology’s recent momentum.

About the OEPCD – The OmniTom Elite with PCD is now available as an optional upgrade (including field upgrades), swapping the standard OmniTom Elite’s energy integrating detector (EID) for a single-source cadmium telluride-based photon counting detector. Beyond the OmniTom Elite with PCD’s imaging advantages (2x higher spatial resolution, spectral CT images at multiple energy levels), all other key features are shared between the two configurations (16 row, 40cm bore, 30cm FoV).

The Photon Counting Race – NeuroLogica’s photon counting CT launch comes about six months after Siemens Healthineers’ NAEOTOM Alpha became the first FDA-cleared PCCT. The OmniTom Elite with PCD’s launch also comes amid major R&D and M&A efforts from essentially all major OEMs, as they compete for photon counting CT leadership. 

The Photon Counting Advantage – Those efforts seem warranted, as PCCTs produce far higher quality images and provide far more imaging data, while potentially allowing lower radiation exposure and contrast dosage. For the OmniTom Elite’s head and neck applications, that could mean improved visualization and segmentation of bones, blood clots, plaques, hemorrhages, and intracranial tumors.

NeuroLogica’s Next Steps – Even if photon counting’s advantages are widely agreed upon, its potential clinical applications are still being explored. Because of that, NeuroLogica’s announcement emphasized ongoing research efforts to evaluate the OmniTom Elite with PCD’s performance with certain patients (e.g. post-trauma and post-surgical patients) and its plans to develop the mobile PCCT’s “full potential.”

The Takeaway

The OmniTom Elite PCD’s head/neck imaging design (vs. whole body) and use of a single-source detector (vs. dual) make it quite different from the other PCCTs being developed, but it’s launch is still a notable milestone for photon counting CT technology. It’s also a testament to Samsung/NeuroLogica’s R&D efforts, coming 4.5 years after showing the detector at RSNA 2017, and reaching the market before most of the biggest CT players released their own PCCTs.

Radiology’s Nonphysician Expansion

A new JACR study detailed nonphysician practitioners’ (NPPs) expansion across US radiology practices, mirroring a trend already seen in other parts of healthcare and raising questions about how much further radiology NPPs might expand.

The Study – The study reviewed 2017-2019 Medicare data for nurse practitioners and physician assistants (together “NPPs”) employed by US radiology practices, finding that:

  • Radiology practices employing NPPs increased by 10.5% (228 to 252 practices), while the number of overall radiology practices declined by 36.5% (2,643 to 1,679)
  • As a result, the share of radiology practices with NPPs on staff nearly doubled (8.6% to 15% of US practices)
  • NPP-employing practices expanded their NPP workforce at a much faster rate (+17.5%, 588 to 691) than they added radiologists (+10.4%, 6,596 to 7,282)
  • The growth of urban practices employing NPPs (10% to 17% share) significantly outpaced rural practices (5% to 7% share), despite a greater need for radiology coverage in rural areas
  • Radiology practices were also more likely to employ NPPs if they were larger, staffed more interventional radiologists, or had a high number of early-career radiologists

The study was limited to radiology-only practices, which employ two-thirds of U.S. radiologists, but excludes many academic, hospital-employed, and multi-specialty groups. That said, it’s possible that radiology NPP growth would be even greater if these groups were included.

The Takeaway

Although 85% practices didn’t employ NPPs and radiologists still outnumbered NPPs by a 32:1 ratio (as of 2019 anyway), this study reveals a clear trend towards more practices employing NPPs and rising overall radiology NPP headcounts. That’s probably not surprising given the historical growth of NPPs within other specialties, and radiology’s continued shift towards national and PE-owned practices, but it’s still interesting to see how it’s taking place. 

It’s also interesting that this study wasn’t met with the level of radiologist uproar that we saw the last few times radiology NPPs made it into the industry news cycle. Even though NPPs’ expansion across radiology practices doesn’t mean that they will start encroaching into radiologists’ clinical territory (as some rads fear), it does suggest that we’ll see a lot more blended rad/NPP workforces going forward.

A CT-First Approach to CAD

A major new study from the DISCHARGE Trial Group showed that coronary CT is as effective as invasive coronary angiography (ICA) for the management of patients with obstructive coronary artery disease (CAD), potentially challenging current guidelines. 

Background – Invasive coronary angiography (ICA) is the reference standard for diagnosing and managing CAD and it’s performed over 3.5 million times each year in the European Union alone (many more millions globally). However, over 60% of these exams prove negative and theoretically could have been diagnosed via non-invasive CT exams.

The Study – The randomized, multi-center trial (26 sites, 16 EU countries) used CT or ICA as the initial diagnostic and treatment guidance exam for 3,523 patients with stable chest pain and intermediate probability of obstructive CAD (1,808 patients w/ CT). By the end of the study’s 3.5-year follow-up period, patients in the CT group had: 

  • A lower rate of major adverse cardiovascular events (2.1% vs. 3% w/ ICA)
  • A far lower major procedure-related complication rate (0.5% vs. 1.9% w/ ICA)
  • A slightly higher rate of reported angina (8.8% vs. 7.5% w/ ICA)

The Takeaway

These results suggest that following a CT-first strategy for evaluating patients with a medium risk of CAD produces similar longer-term outcomes as the current ICA-first strategy (maybe even better outcomes), while significantly reducing major complications and unnecessary cath lab procedures.

That’s pretty compelling and could actually influence procedural changes, given the size / credibility of the DISCHARGE Trial Group and the fact that CT was already proposed in the Chest Pain Guidelines as a gatekeeper for invasive coronary angiography.

A Case for Multiparametric Ultrasound

A new Lancet study out of the UK provided the strongest evidence yet that multiparametric ultrasound might deserve a core role in prostate cancer screening, either as a complement or alternative to multiparametric MRI. That could be a big deal given mpMRI’s cost, time, and accessibility challenges, and makes this study worth a deeper look.

The Study – The researchers performed mpUS and mpMRI exams on 306 patients with signs of prostate cancer (either elevated PSAs or abnormal rectal exams), and then conducted targeted biopsies on the 257 patients who had positive imaging findings. 

The biopsy results revealed cancer in 133 patients, including 83 clinically significant cancers, while showing how mpUS might contribute to prostate cancer diagnosis:

  • mpUS was positive in 272 patients (89%)
  • mpMRI was positive in 238 patients (78%)
  • mpUS identified 66 clinically significant cases (79%)
  • mPMRI identified 77 clinically significant cases (93%)
  • mpMRI and mpUS combined to detect all 83 clinically significant cancers
  • mpUS exclusively detected 6 clinically significant cancers 
  • mpMRI exclusively detected 17 clinically significant cancers

In other words, mpUS was only slightly less accurate than mpMRI for clinically significant cancer detection (-4.3%), but led to far more biopsies (+11.1%), while the combined modalities notably improved clinically significant cancer detection (+7.2%). 

The Takeaway

mpMRI’s role in prostate cancer screening is still secure, but this study shows that mpUS could improve cancer detection if the modalities are used together. Perhaps more importantly, it suggests that mpUS could be a valid prostate cancer detection option for the half of the world that doesn’t have access to advanced imaging or for the many patients who can’t/won’t undergo MRI (orthopedic implants, claustrophobia etc.).

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

You're signed up!

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

-- The Imaging Wire team