Positive Citations | Butterfly Gets Fit to Print | Don’t Fear the Reader

“Two-thirds of the world’s population gets no imaging at all.”

Butterfly Network founder, Dr. Jonathan Rothberg, on the dire need for imaging access in developing regions.

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.
  • Pocus Systems – A new Point of Care Ultrasound startup, combining a team of POCUS veterans with next-generation technology to disrupt the industry.
  • Qure.ai – Making healthcare more accessible by applying deep learning to radiology imaging.

The Imaging Wire

Positive Citations
Radiology studies with positive titles and/or conclusions are more likely to be cited in literature related to diagnostic accuracy compared to studies with negative titles or conclusions. The researchers warned that this phenomenon may cause some clinicians to overestimate the accuracy of certain diagnostic tests at the expense of patient outcomes.

This is from a team of Canadian and Dutch researchers who looked at 995 diagnostic accuracy-related studies that were published between 2005 and 2016. They defined the positivity level of each study’s title and conclusion (positive, neutral, or negative), and then measured their ongoing citations.

The researchers found that adopting a negative title is a recipe for low citations, as studies with positive and neutral titles had 8 to 11 times more citations per month than studies with negatives titles (mean citations per month 0.66 vs. 0.50 vs. 0.06). Meanwhile, studies with positive conclusions had a solid, but not as unbalanced, citation advantage over studies with neutral and negative findings (0.54 vs. 0.42 vs. 0.34 mean citations per month).

Considering that some may not submit (or accept/publish) studies with negative findings in light of this widely-understood positivity bias, the role of positivity in research may be even worse than this study suggests.

Butterfly Gets Fit to Print
There are certain imaging stories that I know right away my mom will forward to me and this is one of them. The New York Times highlighted how the emergence of low-cost handheld ultrasounds (in this case, Butterfly Network’s Butterfly iQ) is allowing significantly improved care delivery in remote African villages.

The story mainly focused on how portable ultrasounds are being used, but also highlighted Butterfly Network’s greater good philosophy and lauded the Butterfly iQ’s low-costs, ease of use/training, rugged design, and potential to support a much wider group of developing nations.

Butterfly Network hasn’t exactly struggled to garner press over the last year, but this NYT piece may be its greatest public relations achievement so far, exposing the company to a much wider audience (beyond tech, startup, and medicine) and positioning the Butterfly IQ as feel-good device that may “revolutionize front-line global medicine.”

Crowdsourced AI
In an effort to alleviate radiation oncologist shortages in many parts of the world, a team of Boston-based researchers launched a crowdsource campaign to develop an AI algorithm that might reduce the labor associated with lung cancer radiation therapy planning.

The team set up a 3-phase contest on developer crowdsourcing site, Topcoder.com, awarding a share of $50,000 to coders who develop AI techniques that are able to learn to draw tumors on CT scans for radiation therapy planning. Out of 34 participating contestants and 45 algorithms, the research team chose 10 finalist algorithms and then combined the top 5 algorithms into one “ensemble” program that actually rivaled the accuracy of trained oncologists. All this happened over the course of 10 weeks.

In addition to potentially helping the team get closer to their goal of addressing regional radiation oncologist shortages, the study provides a new example of how crowdsourcing might be able to bring new AI solutions to market relatively quickly.

The Case for Late-Pregnancy Fetal Ultrasound
A team of University of Cambridge researchers just published a study that strongly supports the adoption of late-pregnancy fetal ultrasound screening for all mothers, suggesting that this step would “virtually eliminate undiagnosed breech presentation” (baby positioned to deliver buttock or feet first) and reduce related C-sections and fetal mortality rates. That alone seems worth it, but the researchers estimated that this added screening step would be cost effective if performed for less than £19.80 ($25.73) per woman.

The researchers studied records from 3,879 women who underwent fetal ultrasound screening during the 36th week of their first pregnancy, revealing that 179 of the women had breech presentation (4.6%) and a pretty significant 96 (53.6%) of those breech presentations were previously undiagnosed. With this early detection, the women diagnosed with breech presentation were offered an ECV procedure to try to turn their babies, scheduling C-sections for women who declined ECV or in cases where ECV wasn’t effective. Just as notably, none of the remaining 3,700 deliveries resulted in breech presentations that weren’t diagnosed by the fetal ultrasound.

The researchers suggest that adopting universal late-pregnancy ultrasound would identify 14,826 undiagnosed breech presentations in England each year, eliminating 4,196 emergency C-sections, 6,061 breech deliveries, and 7 to 8 neonatal mortalities annually.

Don’t Fear the Reader
In an effort to soothe many radiologists’ concerns that AI may someday put their careers in jeopardy, Stanford’s Alex Bratt, MD, took to the JACR to share why “radiologists have nothing to fear from deep learning.” Bratt acknowledged that most AI watchers see diagnostic radiology as most susceptible to automation, but revealed the following reasons why deep learning doesn’t pose a threat to radiologists.

  • Deep Learning remains limited and inflexible compared to humans who can apply a wide range of inputs into their diagnoses (e.g. clinical notes, lab values, prior imaging) and operate far beyond DL’s largely 2D realm.
  • Deep Learning is “brittle.” In other words, DNN performance declines with small changes such as using images from different institutions, limiting how it can be used in clinical settings.
  • Deep Learning needs lots and lots of data in order to function effectively, while human reasoning allows diagnosis with far fewer inputs, even if they are seeing a rare condition for the first time (like “left upper quadrant appendicitis in a patient with gut malrotation”).

Although these reasons may come off as a criticism of AI, Bratt was sure to emphasize the benefits that will come from AI automation and called for radiologists to “educate ourselves so that we can cut through the hype and harness the very real power of deep learning as it exists today.” Even without pointing out the limitations of AI, many of which should be solved over time, there are plenty of human-based reasons that radiologists will be just fine and those who make the most of their AI tools will do even better.

The Wire

  • VisualDx announced a grant from Baylor College of Medicine’s Translational Research Institute for Space Health (TRISH) to provide clinical decision support for ultrasound imaging during deep space flight, with a specific focus on user-guidance and being able to operate without an internet connection. There’s some background to this, as TRISH and NASA have an ongoing cooperative agreement in place and VisualDx was previously awarded a separate research grant for AI point-of-care diagnostics during space travel last year. Ultrasound clinical decision support tools like this are particularly relevant for space travel, given astronauts’ limited medical backgrounds and the likelihood that telemedicine may not always be available in space.

  • The ACA’s elimination of mammography screening co-payments in 2010 resulted in a “statistically significant but small” improvement in zero-cost coverage for all women (racial, income, etc.), with the most significant improvements among African American women. The UCSD and University of Michigan study was based on data from 1,763,959 commercially insured women (40–74yrs) between 2010 and 2014.

  • The AMGA (American Medical Group Association) was quick to voice its “strong opposition” to the PIMA act, which is intended to limit in-office self-referrals for some complex Medicare-paid services (including imaging), suggesting that it would hinder value-based care and delay patient diagnosis. AMGA argues that restricting multispecialty medical groups’ ability to refer patients within their groups for advanced imaging “would have a devastating impact on some of this country’s leading healthcare organizations.”

  • There are plenty of products and procedures that are intended to reduce radiologist burnout, but a new editorial from Indiana University’s Dr. Richard B Gunderman emphasized that radiologists and healthcare organizations would benefit even more by adopting a new mindset. Gunderman specifically suggested that radiologists should adopt “appreciative inquiry,” a problem-solving method that focuses on the positive in any situation, shifting conversations to “what works well” and finding new opportunities to improve.

  • ViewRay announced a pair of distribution partnerships that will support the expansion of its MRIdian MRI-guided radiation therapy system throughout Canada, Australia, and New Zealand. ViewRay signed deals with Canadian distributor Minogue Medical and major Australia/New Zealand distributor Device Technologies to market, sell, and support the MRIdian system in each region.

  • Imaging Biometrics introduced its IB Server software, giving healthcare providers immediate access to the same quantitative parameter maps as IB’s other solutions (IB Neuro, IB Diffusion, IB DCE, and IB Delta Suite) from a networked PC. IB Server automatically detects relevant studies, then executes the appropriate IB software algorithm, and sends the processed parameter maps and quality assurance information to the resident PACS for viewing.

The Resource Wire

This is sponsored content.

  • The latest Carestream blog shares how radiographers and technologists view technology and provides some steps to help them adopt new tech.

  • This Medmo video details how its healthcare marketplace platform and network of participating radiologists help underinsured patients pay as little as possible for their imaging procedures.

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!