#423 – The Wire

  • MDR Non-Compliance: AI for Radiology reports that only 33 of 198 (17%) CE-marked imaging AI products have been recertified under Europe’s year-old Medical Device Regulation requirements (MDR). The EU gave radiology AI software that was already on the market a three-year grace period to meet the more rigorous MDR requirements, although at the current rate, half of those products will not comply on time.
  • SWE for Pediatric Liver Stiffness: A study out of Cincinnati Children’s Hospital showed that ultrasound shear-wave elastography is comparable to MR elastography for assessing liver shear stiffness in pediatric patients. Using MR elastography as the reference, the authors analyzed images from 44 children and young adults with known or suspected liver disease, finding that US SWE detected abnormal liver stiffness with a 0.95 AUC. 
  • BC Lung Cancer Screening: British Columbia is launching the province’s first lung cancer screening program, revealing plans to screen 10k high-risk patients and catch 150 early-stage cancer cases in its first year. The program will operate through 36 imaging centers, screening participants between 55 and 74 years-old with a smoking history of at least 20 years.
  • Healthcare Worker Breaches: Verizon’s Data Breach Investigations Report revealed that employees are responsible for 39% of healthcare data breaches, more than double the 18% average across all industries. The findings show that most insider breaches are tied to unintentional errors as opposed to malicious misuse, with employees being 2.5 times more likely to make an error leading to a breach (e.g. send an email to the wrong address, lose a document) than purposefully abusing their access privileges.
  • Osteoporosis Radiomics: A team of Korea-based researchers developed a deep-radiomics (DR) model that successfully detects osteoporosis in hip radiographs. When tested against an external dataset of 444 X-rays, their highest performing DR model outshined six radiologists (AUCs: 0.95 vs. 0.77) and significantly improved the radiologists’ performance when they had DR support (AUCs: 0.87 vs. 0.77). 
  • Patients’ AI Perspectives: An Intelerad survey of over 1,000 healthcare consumers suggests that the general public is far more optimistic about diagnostic AI than radiologists. The majority of respondents “either trust or are neutral” about autonomous AI-only diagnoses (64%) or radiologists using AI to support their diagnoses (79%), and they are even more open to non-diagnostic AI tasks like making appointments or prioritizing radiologists’ worklist (88% & 86%). They also have big expectations for AI, as 60% think that AI will perform over half of radiology services in five years, with that number increasing to 75% in the next 20 years. 
  • UPenn’s ARK: UPenn’s ultrasound research lab will utilize PaxeraHealth’s ARK platform, which allows users to develop their own AI algorithms without extensive coding. UPenn plans to use ARK to build its own ultrasound AI models designed to detect liver cancer and evaluate COVID severity, and expects that ARK will reduce the time to bring its AI models to market. 
  • Low on Helium: The dwindling helium supply hit mainstream news last week following the Dollar Tree publicizing concern that the helium deficit will impact its graduation and birthday balloon sales. MRI users and vendors are well aware of the helium shortage, but this may still be a good conversation starter for MRI players with a strong low-helium value proposition. 
  • Novarad Adds CryptoChart Lite: Novarad expanded its imaging sharing portfolio with the launch of its CryptoChart Lite solution. CryptoChart Lite is a no-cost version of Novarad’s established CryptoChart solution, similarly leveraging a printed encrypted QR code or web access code to support image storage and transfer (no CDs, logins, portals, or software installs). Although providers would have to upgrade to the full CryptoChart version to access more advanced features, CryptoChart Lite should serve as a useful starting point for those who are becoming ready to ditch the disk.
  • Move Slow and Test Things: Google AI’s principal research director Greg Corrado recently spoke at STAT’s Health Tech Summit about his plans to address concerns surrounding healthcare AI. Corrado said that Facebook’s infamous “move fast and break things” slogan is the exact opposite of how tech companies should approach health AI problems, and listed several considerations such as performing rigorous testing at each step of algorithm development and ensuring that models are tested with data from populations where they’ll be used.
  • DLR Supports CT-FFR: A new study out of China found that deep learning image reconstruction (DLR) improves coronary CTA image quality, and more importantly, doesn’t negatively impact CT-FFR diagnostic performance. Using QFR and invasive FFR results as a reference, the authors analyzed DLR and four other reconstruction approaches using 182 CTAs from 33 patients, finding no significant difference across their CT-FFR values and diagnostic performance (AUCs: 0.83 vs. 0.81 to 0.86; p>0.05).

You're signed up!

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

-- The Imaging Wire team