AI Bubble Buzz | Medicare for All Targets Radiologists | Crowdsourced AI

“The bubble will pop – and the market will consolidate.”

Dr. Hugh Harvey as part of a recent Twitter conversation on the imaging AI bubble.

We’re coming into the final week to nominate make your Imaging Wire Awards nominations. If you or a colleague made outstanding contributions to radiology practice and outcomes this year, please take three minutes to nominate them (or you) here.

Imaging Wire Sponsors

  • Carestream – Focused on delivering innovation that is life changing – for patients, customers, employees, communities and other stakeholders
  • Focused Ultrasound Foundation – Accelerating the development and adoption of focused ultrasound
  • Medmo – Helping underinsured Americans save on medical scans by connecting them to imaging providers with unfilled schedule time
  • Nuance – AI and cloud-powered technology solutions to help radiologists stay focused, move quickly, and work smarter
  • Pocus Systems – A new Point of Care Ultrasound startup, combining a team of POCUS veterans with next-generation genuine AI technology to disrupt the industry
  • Qure.ai – Making healthcare more accessible by applying deep learning to radiology imaging

The Imaging Wire

AI Bubble Buzz

Reports of “layoffs at some prominent radiology AI companies” got the radiology AI Twitter community going last week, as the AI “bubble-ists” were quick to embrace this news as supporting evidence for their bubble theories and the AI thought leaders used it as a chance to discuss how much more AI has to evolve. Regardless of your preferred confirmation bias angle, these threads were packed with big names and delivered some big insights.

The Langlotz Four – AI leader Curt Langlotz broke the news and shaped much of the conversation with his four reasons that radiology AI companies haven’t lived up to investor expectations: 1. High quality clinical training data is hard to get; 2. Commercial radiology AI algorithms mostly solve easier problems, making them “nice to have” features; 3. Radiology AI algorithms haven’t generalized as well as expected, so cautious hospitals are making the AI business development process longer and more costly; 4. Few people/companies have experience delivering AI results into radiology workflow and its difficult to do.

Other Culprits – Imaging AI Twitter had plenty of other points to add to Langlotz’ list, including AI’s unestablished payment/reimbursement structure, the lack of clinical evidence that AI is effective, the fact that AI tools aren’t yet integrated into core healthcare processes, the prevalence of narrow AI tools, and poor/absent PACS integration.

The Bubble – News of the layoffs reinvigorated forecasts that the AI bubble would pop and the market would consolidate, which is a safe bet for any fast-emerging industry (all industries consolidate). Meanwhile, the AI optimists used this thread to remind everyone that a deflating bubble doesn’t mean that the premise/promise of radiology AI is wrong, while assuring that the AI players who emerge after a correction would have major opportunities available to them. Not unlike Forrest Gump’s shrimp boat after the hurricane.

It’s still unclear how many AI companies are laying folks off or how large these layoffs are. In fact, they might not be all that different from layoffs taking place in other industries. Whether or not these layoffs represent a change in the market, the problems and solutions detailed in these Twitter threads serve as a valuable guide for any AI firm who wants to emerge as a leader in the coming years.

Warren’s Radiologist Plan

Sen. Elizabeth Warren unveiled her long-awaited “Medicare for All” funding plan, which quickly caught radiologists’ attention due to its intention to slash rates for “overpaid specialties” like radiologists and orthopedic surgeons.

There’s way more to this 28-page plan than simply cutting radiologists’ pay (tax changes, basically eliminating private coverage, cutting payments to hospitals/docs/drug companies) and most coverage has made it seem like Warren singled out radiologists more than she really did. Still, radiologists are certainly included on Warren’s list of “overpaid specialties” and that’s pretty notable for the folks reading this newsletter.

One bit of “good news” for any concerned radiologists is their reimbursements would still be about 10% above current Medicare rates, keeping their pay well above their peers in most other countries. The other bit of good news for them is even if Warren became president, Medicare for All would face Obamacare-level pushback and have to make Obamacare-level compromises in order for the remnants of this plan to become a reality.

Crowdsourced AI

Australian AI company, Presagen, took aim at healthcare AI’s dataset access and bias challenge with the launch of its new AI Open Projects platform. Here are some details:

AI Open Project – Presagen’s AI Open Projects is an online platform that connects multiple clinics, allowing them to crowdsource globally diverse datasets and co-develop AI models “that are robust, scalable and unbiased.”

Decentralized AI – AI Open Projects is based around Presagen’s patent-pending Decentralized AI Training technique, which allows AI training without moving or centralizing the data. Instead, the technique moves the AI to the data for training and only the AI derived from the data is shared and moved between data sources (never the private data itself).

Participation is Key – Presagen is seeking participants for a range of AI Open Projects (radiology, retinal, and fertility), asking them to provide data, expertise, and clinical support, while “Presagen does the rest.” The “rest” in this case is pretty significant, including building the AI, managing regulatory approvals, and commercialization. AI Open Projects was already used to launch the IVF solution, Life Whisperer, and will soon be used to develop a radiology solution for lung cancer detection.

Like most innovative approaches, AI Open Project will face some skepticism (maybe about its business model, maybe participation, maybe validation/testing), but it’s an interesting way to target some of AI’s biggest challenges.

The Wire

  • A Brigham and Women’s Hospital and Harvard Medical team introduced a pilot radiology teaching consultation service (TCS) intended to expand radiologists’ role as imaging consultants for referring physicians. BWH integrated the TCS into its Internal Medicine Morning Report (IMMR) resident educational program, relying on senior radiology residents to create “dynamic” teaching slides for each referred case that is presented during IMMR sessions. Following the seven-month pilot, a survey of six IM residents found that the slides helped with their IMMR preparation and presentations (83%), helped them engage their audience (100%), gave them more confidence teaching radiology (67%), improved their ability to understand radiology reports (67%), and gave them a greater appreciation for radiologists’ work (100%).
  • Clarius Mobile Health launched its second-generation series of wireless ultrasound scanners, expanding the systems’ image processing power (8x more than most handheld systems) and reducing their size (almost half the size of 1st gen) and cost (C3 Multipurpose starts at $4,900). The 2nd generation series includes two multi-purpose scanners and four specialty-targeted scanners (e.g. sports medicine, anesthesia, OBGYN), expanding upon its 1st generation series’ three-model lineup (Convex, Linear, Microconvex).
  • After months of hospital backlash against CMMS’ proposed healthcare cost transparency rule, including over 1,400 comments, the Trump administration decided to delay the rule that would have required hospitals to share their negotiated rates. However, CMMS clarified that the delay is only intended to produce a new rule covering both hospitals and insurers later this year, which could lead to a legal battle over whether the federal government can force hospitals to disclose what they consider trade secrets.
  • A new paper in Clinical Radiology shared a guideline for the development of “responsible AI,” outlining how radiologists can support AI development by leveraging their clinical/technical expertise and ensuring that solutions target unmet clinical needs, “not just what’s interesting or feasible.” The article also clarified that responsible AI should include radiologist-validated data with a strong ethical foundation (privacy protected), validating models based on traditional “diagnostic test accuracy” and using real-world testing.
  • Major Pennsylvania area healthcare provider Geisinger adopted Life Image’s ‘first-of-its-kind’ Mammosphere solution, allowing patients to request, store, and share their breast health records, including prior mammograms. This is the first time a health system has provided open access to prior breast health images and records, and was highlighted as a significant breast health achievement given the role of previous mammogram access in the early detection of breast cancer.
  • A University of Michigan team found that patients are likely to forgo head CTs or other imaging studies with the right combo of clinical evidence and financial motivation, representing a potential way to curb emergency imaging overuse. The team surveyed 913 ED patients, presenting them with hypothetical minor traumatic brain injury emergency scenarios, different levels of potential benefits (detecting brain hemorrhage) and risks (developing cancer) from undergoing a head CT scan, and different incentives to skip testing ($0 or $100). The team found that increasing the benefits of CT from 0.1% to 1% resulted in a significant increase in test acceptance (adjusted OR = 1.6) and increasing the risk of a scan from 0.1% to 1% significantly decreased test acceptance (AOR = 0.70). Meanwhile presenting the patients with a $100 incentive to forego low‐value testing significantly reduced test acceptance (AOR = 0.6).
  • The CMS named the 25 companies chosen to participate in the first stage of its $1.6 million Artificial Intelligence Health Outcomes Challenge (out of 300), surprising some in the industry by not including any imaging AI firms among the final participants. That said, imaging AI wasn’t the best fit for this challenge, given its focus on using AI/ML to “predict unplanned hospital and skilled nursing facility admissions and adverse events.”
  • Research by a Pittsburgh-based team found that contrast-enhanced mammography (CEM) and molecular breast imaging (MBI) achieved similar malignancy detection rates as breast MRI, possibly serving as lower-cost alternatives to contrast-enhanced MRI for staging newly diagnosed breast cancer. The study of 99 women with 110 index malignancies found that MRI, CEM, and MBI revealed a similar number of malignancies (102, 100, 101), while MRI overestimated the size of more of the malignancies than CEM and MBI (24%, 11%, 15%), and MRI revealed more non-index lesions that resulted in additional biopsies with a lower positive predictive value (MRI: 46, 28%; CED: 27, 53%; MBI: 25, 44%).
  • Last week’s news cycle brought a pair of physician shortage articles and an article on AI’s potential to alleviate doctor shortages, which weren’t planned but certainly identified one of the global challenges facing healthcare. Forbes published a relatively standard story on how AI solutions (particularly from Swedish AI firm Peltarion) could help solve the UK’s “cancer doctor shortage,” while separate stories described how India’s rural radiologist and scanner shortage is creating challenges for the country’s pregnant women and how the U.S.’ upcoming physician shortage may have the greatest impact on the country’s rural communities.
  • German AI developer Merantix Healthcare announced the CE clearance of its Vara automated breast cancer screening solution, which is used to enhance radiologist workflow by filtering out healthy mammograms to allow them to focus on “difficult images.” Vara is trained on an over 2 million image dataset and has been deployed in five European countries so far.

The Resource Wire

  • Did you know that imaging patients are most likely to no-show for their procedures on Mondays and Saturdays? By partnering with Medmo, imaging centers can keep their schedules full, despite the inevitable Monday no-shows.
  • Catch Nuance’s Karen Holzberger and Woojin Kim, MD in this radiologytoday.net article, where they share how AI app marketplaces are bridging the gap between AI creators and users.
  • This spotlight details how the Sunnybrook Research Institute pioneered focused ultrasound and became the word leader in FUS clinical trials.
  • Headed to RSNA 2019? Qure.ai will present four abstracts at the conference covering: a new metric to evaluate radiology AI models, chest X-ray TB screening, segmenting and measuring ventricular and cranial vault volumes with AI, and how clinical context improves AI performance for cranial fracture detection.

You might also like

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!