#302 – Are We Being Fair to AI?

“It appears that radiologists and AI, for slightly different reasons, are in the same boat.”

Duke University’s Maciej A. Mazurowski, Ph.D., on our challenges with explainable decision making.

Imaging Wire Sponsors

  • Arterys – Reinventing imaging so you can practice better and faster.
  • Bayer Radiology – Providing a portfolio of radiology products, solutions, and services that enable radiologists to get the clear answers they need.
  • GE Healthcare – Enabling clinicians to make faster, more informed decisions through intelligent devices, data analytics, applications and services.
  • Healthcare Administrative Partners – Empowering radiology groups through expert revenue cycle management, clinical analytics, practice support, and specialized coding.
  • Hitachi Healthcare Americas – Delivering best in class medical imaging technologies and value-based reporting.
  • Novarad – Transformational imaging technologies that empower hospitals and clinicians to deliver clinical, operational and fiscal excellence.
  • Nuance – AI and cloud-powered technology solutions to help radiologists stay focused, move quickly, and work smarter.
  • Riverain Technologies – Offering artificial intelligence tools dedicated to the early, efficient detection of lung disease.
  • Siemens Healthineers – Shaping the digital transformation of imaging to improve patient care.
  • United Imaging – Our mission, Equal Healthcare for All, pushes us beyond conventional boundaries to help clinicians expand modern, digital, intelligent care to more people within their communities.
  • Zebra Medical Vision – Transforming patient care with the power of AI.

The Imaging Wire

Uneven Expectations

A new Radiology: Artificial Intelligence editorial questioned why we expect more from imaging AI than human radiologists, suggesting that these uneven expectations stand in the way of achieving the best possible interpretations. Here’s a breakdown of our uneven expectations and some suggestions on how to fix them:

  • Overrating Explainability – One of the biggest AI criticisms is that algorithms can’t explain their results, but the author argues that we also don’t understand the human decision-making process. We just give humans the benefit of the doubt because “we believe that they are like us,” but that might “be a false comfort.”
  • Judgement Calls – AI skeptics are also concerned with AI’s inability to apply “judgement,” “intuition,” and “common sense.” However, these are generally used to support decisions when we don’t have enough data, and insufficient data isn’t good for human or AI decision making.
  • Comparing Performance – We spend a lot of time measuring whether AI can interpret scans better than radiologists. However, we rarely measure whether radiologists can interpret scans better than other radiologists, even though half of all radiologists’ perform below the specialty’s median. So the question is, which radiologist are we comparing these AI tools against?
  • Focus on Optimization – Rather than measuring AI’s performance versus radiologists (of various performance levels), we should be measuring how AI improves all radiologists’ interpretation accuracy. Initially, that would require a “rigorous assessment of both AI and radiologists” to understand what to optimize (and how), rather than just giving radiologists an AI tool and measuring future improvements.

The Wire

  • ABR’s 6-Week Notice: The American Board of Radiology is facing backlash yet again, this time for proposing that residents can only take “up to six weeks” of family medical leave during their entire four-year residencies. Safe to say the folks on Radiology Twitter hated this proposal for a range of reasons (too short for new moms/babies, doesn’t set a minimum leave, undermines programs directors’ authority), calling it “draconian,” “completely inhumane,” and “not family friendly.”
  • Dementia Emergency Imaging On the Rise: Even though patients with dementia are visiting the ED less often, they are receiving much more emergency imaging per visit. That’s from a new Emory study that measured ED visits and imaging volumes from 2006 to 2014 (n = 428k patients w/ Alzheimer’s, 33.7k w/ vascular dementia), revealing a significant drop in Alzheimer disease and vascular dementia patients’ ED visits (-24.7% & -20.3%), and notable increases in head CT (4.4% to 11.1% & 1.5% to 2.9%) and brain MRI (0.04% to 0.5% & 0.0% to 0.1%) utilization rates.
  • Rad Tech Prep: Staten Island’s St. Peter’s Boys High School announced a radiologic technologist training program that will give its students an unusually early start towards a radiography career (20% of RT credits). St. Peter’s rad tech program will include a combination of onsite and online classes through a partnership with medical training school John Patrick University, and even include an on-campus imaging suite donated by large east coast dealer Prestige Medical Imaging.
  • Upskilling RTs: In other technologist training news, the University of Texas Medical Branch (UTMB) just launched a partnership with the College of Health Care Professions (CHCP, a large medical training school) to “upskill” UTMB’s certified medical assistants into limited X-ray technologists. The first-of-its-kind, UTMB-paid program will train the assistants in six months (half the typical training length), while providing UTMB with a pipeline of future limited X-ray technologists to staff its smaller clinics that don’t have enough volume for experienced rad techs.
  • AI’s Missing Piece: A HealthCare IT News editorial suggested that health systems will have to start viewing themselves as “engineering houses” in order for them to get the imaging AI results they want. The author (Penn Medicine’s chief data scientist) suggested that many healthcare systems often end up implementing AI without defining the problems they’re solving or applying the necessary IT rigor, while AI vendors often turn on their solutions and “then walk away.” To fix this, the author called for a more comprehensive approach to designing a complete AI solution (tech, human workflow, measurable value, long-term operational capability).
  • CEM for Augmented Breasts: A new AJR study provided the first evidence of contrast-enhanced mammography’s (CEM’s) effectiveness imaging newly-diagnosed breast cancer among women with augmented breasts, perhaps as an alternative to MRI. The Mayo Clinic researchers reviewed imaging from 17 women with implants and newly-diagnosed breast cancer. Both modalities were in agreement for all women’s index cancer, CEM identified an additional lesion in six of the women, and no lesions were identified with MRI that weren’t also identified with CEM.
  • Sense’s ICH Headset on Trial: Sense Diagnostics is launching a 400-patient clinical trial to evaluate how its non-invasive brain scanner helps physicians monitor intracranial hemorrhage between CT scans. The company’s SENSE headset uses low-power tailored radio frequency (RF) pulse to detect real-time changes that may indicate expanding brain bleed, potentially allowing earlier treatment and less reliance on CT-based surveillance.
  • SR Adds Delaney: Strategic Radiology added Wilmington, North Carolina’s Delaney Radiology (28 radiologists, 2 imaging centers, services 4 hospitals) to its consortium of independent radiology practices, which now includes 30 member practices and over 1,300 radiologists. Delaney is SR’s fifth practice in North Carolina, all of which joined since early 2019.
  • AI IDs Orthopedic Implants: A UK-based team developed a deep learning model able to automatically identify metallic orthopedic implants better (and faster) than orthopedic specialists, suggesting that algorithms like this could support preoperative planning for implant replacements. The team developed seven different CNNs using hip and knee X-rays featuring 12 different implant models and tested them against a 180 X-ray dataset. The best performing model identified implants with 98.9% accuracy (178 of 180), well above the study’s five specialists (76.1% median, 85.6% best).
  • Cost Transparency’s Slow Start: It’s been over two months since the new healthcare cost transparency rules went live, but most U.S. hospitals still aren’t complying. That’s from new research in Health Affairs that found 65 out of the 100 hospitals analyzed were “unambiguously noncompliant,” including 12 that did not post any cost information and 53 that didn’t post payer-specific negotiated rates. Only 22 hospitals were fully compliant.
  • Arrhythmia Ultrasound: Ultrasound is more effective than electrocardiograms (ECGs, the current standard) for identifying cardiac arrhythmia locations. That’s from a New York-based study that performed electromechanical wave imaging (EWI, a type of ultrasound) and ECGs on 55 patients with cardiovascular disease. EWI outperformed ECG for localizing atrial and ventricular arrhythmias (96% vs. 71% location accuracy), and suggesting that EWI could improve cardiac ablation decision-making and treatment planning.
  • NeuroLogica’s New Mobile Stroke Unit: NeuroLogica launched its new SmartMSU mobile stroke unit, adding the company’s latest OmniTom Elite CT system to its specialized stroke ambulance, which previously featured the dated CereTom CT system. The SmartMSU owes much of its improvements to the updated CT (16-slice vs. 8-slice, smaller form factor, larger gantry, better image viewing, new contrast workflow) while also updating its stretcher and perhaps other non-CT features.
  • Investigating Change: The American Hospital Association asked the U.S. DOJ to investigate UnitedHealth Group’s acquisition of Change Healthcare, suggesting that the merger could reduce competition for health care IT services at the expense of hospitals and providers. The AHA suggested that shifting healthcare data from Change (“a neutral third-party”) to UHG’s Optum subsidiary “would impact (and likely distort) decisions about patient care and claims processing and denials.”
  • Peer Pressure Works: A new study out of Duke detailed how introducing a peer-comparison radiology utilization dashboard intended to reduce unwanted clinical variation also helped reduce imaging volumes and costs. The researchers analyzed radiology orders from 159 Duke network primary care providers, revealing that informing PCPs how their radiology utilization stacked up against their peers led to order changes that cut radiology costs by 19.5% in the first year (led by: CT -34.8%, ultrasound -21.5%, MRI -16.6%).

The Resource Wire

– This is sponsored content.

  • With radiation dose management now largely considered best practice, this Bayer white paper details the top five benefits of adopting contrast dose management.
  • In this Hitachi blog, VidiStar users share how they’ve benefitted from the cardiovascular information system’s flexible SaaS-based pricing model and leveraged its productivity advantages to increase reimbursements.
  • If it were just about the machines, things would be a lot simpler. In this video, part two of their RSNA virtual presentation, United Imaging explains how they go beyond the status quo to improve the process of buying and owning medical equipment.
  • It says a lot when a solution works so well for a radiology department that they decide to perform a study to quantify its benefits. In this Imaging Wire Q&A, University Hospital of Zurich’s Thomas Frauenfelder discusses his experience and study on Riverain Technologies ClearRead CT.
  • With a post-COVID surge in orthopedic surgery looming, this Zebra Med blog details how its 3D imaging orthopedic solution can help surgeons improve their preoperative efficiencies.
  • CMS continues to modify the Quality Payment Program (QPP) and MIPS policies due to the COVID-19 public health emergency, but the changes and their timelines can be hard to manage. Check out this Healthcare Administrative Partners post detailing how radiology QPP and MIPS policies have changed and how radiology practices should react.

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!