Everyone Gets an Ultrasound | AI’s Inconvenient Truth | Outsourced AI

“She was not trained as a doctor, but she was helping to teach an artificial intelligence system that could eventually do the work of a doctor.”

A New York Times article on one of the tens of thousands of people in developing nations making a living from labeling medical images to train AI.

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

Everyone Gets an Ultrasound

UC Irvine Medical School just gave all 104 of its incoming students Butterfly iQ ultrasound systems at their White Coat Ceremony, positioning the handheld devices as a way to make ultrasound imaging more accessible to its students. Here’s why this is newsworthy:

  • The Symbolism – Given some folks’ vision that handheld ultrasounds will eventually be in the pocket of every physician, this was widely seen as a symbolic event.
  • The Realism – UCI’s tech savvy reputation and history of giving gadgets during White Coat Ceremony (they gave iPads in 2010) suggests it may be awhile before other med schools follow suit.
  • The Marketing – Few healthcare brands get as much social media buzz as Butterfly Network and they did it again with this move. The UCI ceremony was quickly followed by a surge of online praise, with numerous people calling the one-to-one MD-to-US ratio “the future of medicine” and just as many folks saying they “wanted one.” Not sure if Butterfly gave UCI a discount, but if they did it’s probably worth it.

AI’s Inconvenient Truth

A Nature op-ed from a Harvard and MIT team shared “the ‘inconvenient truth’ about AI in healthcare,” suggesting that the AI tools detailed in research articles are not usable in the clinic. Here’s why:

  • No Incentives – They argued that healthcare operates off of a fragmented system (due to political and economic factors, medical practice norms, and commercial interests) that determines how care is delivered, and “simply adding AI . . . will not create sustainable change.”
  • No Infrastructure – The team believes most healthcare organizations don’t have the infrastructure to collect the necessary data to train algorithms so that they match the local population nor do they have the ability to test algorithms for bias.
  • No EHR Data – Although the emergence of cloud storage and rapid EMR adoption is reason for optimism, the paper suggests that the powerful EHR players benefit too much from the “status quo” to make the necessary data infrastructure changes.
  • Here’s the Yes – In order to realize healthcare AI’s potential, the team believes we’ll need public discourse and policy intervention to address “who owns health data, who is responsible for it, and who can use it.” Data will also have to be protected, technologically and legally. “Without this however, opportunities for AI in healthcare will remain just that—opportunities.”
  • Two Routes – If these issues are addressed, then the necessary data infrastructure to support “tomorrow’s AI enabled workflows” could be built either through evolution (creating data infrastructure from existing work) or revolution (a gov’t mandate that all providers store their clinical data in commercially available clouds).

Outsourced AI

A recent NYT profile detailed the booming healthcare AI labeling industry in many of the world’s developing regions, where thousands of workers are labeling medical data to train AI systems every day.

Described in the article as a “ticket to the middle class,” the workers perform a wide range of labeling tasks, ranging from listening to cough recordings to identify “bad coughs” to circling intestinal polyps in colonoscopy videos. This industry also includes plenty of non-medical labeling gigs, such as training social media AI to catch obscene content or training autonomous cars to spot stop signs.

The article didn’t make any specific mention of “our” kind of medical imaging, which makes sense given the education needed to label a CT, but also suggests that this AI labeling boom might not alleviate “our” labeled image shortage.

The Wire

  • Abbott’s in-development i-STAT Alinity blood test could diagnose mild traumatic brain injuries in 15 minutes, even diagnosing patients with normal CT scans. A study using the i-STAT Alinity evaluated 450 ED patients with suspected TBI but negative CT scans, using the blood test to confirm that brain-specific glial fibrillary acidic protein (GFAP) could be a TBI biomarker. Among the 90 patients with the highest levels of GFAP detected, 64% were later confirmed to have a TBI when they received an MRI scan.
  • A new article published in Radiology reviewed the risks associated with 7T MRI (e.g. heating of metallic implants, vertigo) and shared some strategies to reduce these risks (e.g. increase time in 7T field, use earplugs or headphones). The team’s strongest suggestion was for further 7T guidelines and testing, which is evident given that only about 300 implants and RF coils have been tested with 7T MRI versus over 6,000 metallic devices tested with 1.5T and 3T MRIs.
  • Butterfly Network launched Butterfly Education, an educational video library with short instructional videos across a range of categories and applications. Butterfly Education is available to everyone with a Butterfly iQ subscription, helping to add value to the system’s cloud service and building Butterfly’s credentials in the education-centric POCUS segment.
  • NTT DATA Services will use Google Cloud to deliver cloud, analytics, and AI healthcare solutions across its large healthcare IT customer base. NTT DATA was light on specifics, revealing only that it would share more later this year, but NTT does have a solid footprint in medical imaging, and made several moves deeper into imaging AI within the last few years (here are two from 2018 and 2017).
  • Duke biomedical engineers developed a new higher-resolution OCT technology, called optical coherence refraction tomography (OCRT), that overcomes OCT’s lateral resolution challenges (historically worse than depth resolution) and should help improve OCT’s imaging of objects with irregular shapes. OCRT was created by combining OCT images from multiple angles to extend the depth resolution to the lateral dimension, then using machine learning to address distortion created by OCT light refraction.
  • DiA Imaging Analysis announced a partnership with Edan Instruments to integrate DiA’s AI-based cardiac solution, LVivo Toolbox into the Chinese medical equipment manufacturer’s ultrasound devices. LVivo Toolbox is gaining momentum, as this partnership comes a few weeks after a similar ultrasound alliance with Terason and previous PACS and ultrasound partnerships with Konica Minolta, Esaote, and GE.
  • A JMIRS paper from a Sunnybrook-led team suggests that the combination of imaging AI and pathology AI, computer vision, and radiomics is poised to improve breast cancer treatment decision-making. The paper suggests that AI’s role in breast oncology will grow as its ability to leverage “multi-omics” data expands, creating new understanding of cancer and a tumor’s biology that will help improve treatment.
  • NVIDIA and NIH researchers developed an AI tool using NVIDIA’s Clara Train SDK that helps detect prostate cancer using multiparametric MRI (mpMRI) in less time. The team trained the tool on 465 mpMRI scans from diverse sources (multiple centers, brands, protocols) and evaluated it on 98 patients, achieving a DICE score of 0.922 (versus a panel of radiologists’ 0.919 DICE score.

The Resource Wire

– This is sponsored content.

  • Nuance’s latest blog highlights how ImageBiopsy Lab’s KOALA (Knee Osteoarthritis Labeling Assistant) algorithm helps physicians spot signs of knee osteoarthritis.
  • This Carestream case study compares images of foot trauma captured using the OnSight 3D Extremity System to images captured on 2D X-rays.
  • Yale University research reveals that the average patient drives past SIX lower-cost providers on the way to an imaging procedure, due in large part to patients’ and physicians’ limited cost consciousness. Medmo helps address this issue by letting patients enter what they can afford for their scan, then booking them at a nearby imaging center willing to accept that price.

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!