Saurabh Jha, MBBS, AKA RogueRad on X/Twitter.

Our understanding of breast tissue density and its relationship to cancer risk has advanced greatly in the last two decades. In this episode of the Imaging Wire Show, we talk to two key opinion leaders on breast density and cancer risk, Stamatia Destounis, MD, of Elizabeth Wende Breast Care in Rochester, New York and Jack Cuzick, PhD, of the Wolfson Institute of Population Health at Queen Mary University of London.

Imaging Wire Sponsors

Bayer  •  Blackford  •  CARPL.ai  •  Change Healthcare  •  Clearpath  •  Enlitic  •  GE HealthCare  •  Intelerad  •  Medality  •  Merge by Merative  •  Nuance  •  PocketHealth  •  Riverain Technologies  •  Siemens Healthineers  •  Subtle Medical  •  Tempus Radiology  •  United Imaging  •  Us2.ai  •  Visage Imaging

Artificial Intelligence

AI Tug of War Continues

The ongoing tug of war over AI’s value to radiology continues. This time the rope has moved in AI’s favor with publication of a new study in JAMA Network Open that shows the potential of a new type of AI language model for creating radiology reports.

• Headlines about AI have ping-ponged in recent weeks, from positive studies like MASAI and PERFORMS to more equivocal trials like a chest X-ray study in Radiology and news from the UK that healthcare authorities may not be ready for chest X-ray AI’s full clinical roll-out. 

In the new paper, Northwestern University researchers tested a chest X-ray AI algorithm they developed with a transformer technique, a type of generative AI language model that can both analyze images and generate radiology text as output. 

• Transformer language models show promise due to their ability to combine both image and non-image data, as researchers showed in a paper last week.

The Northwestern researchers tested their transformer model in 500 chest radiographs of patients evaluated overnight in the emergency department from January 2022 to January 2023. 

Reports generated by AI were then compared to reports from a teleradiologist as well as the final report by an in-house radiologist, which was set as the gold standard. The researchers found that AI-generated reports …

• Had sensitivity a bit lower than teleradiology reports (85% vs. 92%)
• Had specificity a bit higher (99% vs. 97%)
• In some cases improved on the in-house radiology report by detecting subtle abnormalities missed by the radiologist

Generative AI language models like the Northwestern algorithm could perform better than algorithms that rely on a classification approach to predicting the presence of pathology. Such models limit medical diagnoses to yes/no predictions that may omit context that’s relevant to clinical care, the researchers believe. 

In real-world clinical use, the Northwestern team thinks their model could assist emergency physicians in circumstances where in-house radiologists or teleradiologists aren’t immediately available, helping triage emergent cases.

The Takeaway

After the negative headlines of the last few weeks, it’s good to see positive news about AI again. Although the current study is relatively small and much larger trials are needed, the Northwestern research has promising implications for the future of transformer-based AI language models in radiology.

AMRIC’s Multimodality Approach to Medical Imaging

AMRIC is a new premium imaging clinic on Manhattan’s Upper East Side. Learn about their multimodality approach to medical imaging and why they turned to United Imaging for cutting-edge CT, MRI, and molecular imaging technology.

Starting a CT Lung Cancer Screening Program

There’s no doubt CT lung cancer screening reduces lung cancer mortality. In this Imaging Wire Show, we discuss how to set up a lung screening program and the technologies available to help with Chip Gilkeson, MD, of University Hospitals Cleveland Medical Center and Steve Worrell of Riverain Technologies.

The Wire

• Algorithms at the Gate: Radiologist Saurabh Jha, MBBS, has always had a knack for a clever turn of phrase, and he doesn’t disappoint in a new JAMA paper entitled “Algorithms at the Gate” that addresses radiology’s dilemma in adopting AI. Jha views AI through the lens of radiology’s work structure and the incentives radiologists face, pointing out that AI could enable more work to be done by fewer radiologists – a recipe for burnout. To realize AI’s full potential, radiologists must relinquish some of their work to the algorithms.

• ChatGPT Now Analyzes Images: The latest version of the ChatGPT large language model (LLM), ChatGPT-4 Vision (GPT-4V), can now analyze images. Users can even input medical images and get an output that looks like a radiology report – a frightening prospect for radiologists. But a deeper dive reveals faults, as explained in a thread on X/Twitter. The potential of GPT-4V is also explored in a new paper posted to the arXiv server, but at 166 pages you might just want to ask ChatGPT for a summary. 

• ChatGPT Reports for Patients: One potential use case for ChatGPT that isn’t scary – and might actually be realistic – is creating patient-friendly radiology reports. In European Radiology, German researchers asked ChatGPT to create three fictitious radiology reports that could be understood by a child. Fifteen radiologists rated the reports’ quality, finding them to be factually accurate and complete, and not potentially harmful to patients. Some incorrect statements were found, indicating that ChatGPT-generated content still needs some radiologist oversight.

• Qure Partners with USARAD: Qure.ai has cemented a three-year partnership with teleradiology provider USARAD Holdings to deploy its qXR chest X-ray AI software through USARAD’s network. USARAD provides on-call and daytime radiology and teleradiology services through a network of 116 US radiologists; the company sees the Qure software as helping its radiologists make swift decisions and prioritize critical cases. Qure has FDA clearance for 12 AI algorithms so far, the most recent being its qXR-CTR application for measuring cardiothoracic ratio (CTR) from chest X-rays.

• Fine-Tuning CT Lung Cancer Screening: Researchers continue to look for ways to select at-risk individuals for CT lung cancer screening that are more precise than the USPSTF’s criteria. In Cancer, a team from South Dakota proposed the PLCOm2012 model, based on the Prostate, Lung, Colorectal, and Ovarian (PLCO) trial. The model incorporates multiple individualized variables, and in tests with 1.6k people it had higher sensitivity and found more eligible screening candidates than either the 2013 or 2016 versions of USPSTF criteria (91% vs. 58% and 66%). Better selection criteria could improve screening’s cost-benefit calculation. 

• Care Shifting to NPs and PAs: Healthcare is quickly shifting away from doctors, with a Harvard-led study in The BMJ revealing that a quarter of all visits are now delivered by NPs and PAs. After analyzing 276M Medicare visits between 2013 and 2019, researchers found that the share of visits delivered by NPs and PAs climbed from 14% to 25.6%, with wide variations across conditions. Respiratory infections led the way with 41.5% of visits delivered by NPs and PAs, followed by anxiety disorders (36.7%), hypertension (20.4%), and eye conditions (13.2%).

• GE Partners with University Hospitals: GE HealthCare has signed a partnership with University Hospitals of Cleveland, Ohio, to provide equipment and digital solutions. GE will be the sole provider for hundreds of new systems, including digital technology like AI, as well as nuclear medicine, X-ray, vascular and cardiovascular ultrasound, CT, fluoroscopy, surgery, and bone densitometry. The deal is the latest large equipment sale for GE, which recently signed a $30M deal to provide CT scanners to St. Luke’s University Health Network in New Jersey. 

• AI-Driven Guidance Added to POCUS: In other GE HealthCare news, the company is adding Caption Guidance for cardiac imaging to its Venue family of point-of-care ultrasound (POCUS) scanners. Available as an option, Caption Guidance gives clinicians step-by-step instructions for acquiring cardiac images, expanding the number of personnel who can assess patients for signs of heart disease. The Venue scanners are the first to feature the technology, which GE acquired through its purchase of Caption Health earlier this year. 

• Employment Remains Strong: Healthcare employment just saw its best two-month stretch since the start of the pandemic, with Altarum Institute data showing that the sector added 71k jobs in August and 73k in July. Ambulatory care (39,900) and hospitals (14,500) led the pack, bringing healthcare employment to 3.2% above where it was prior to COVID (vs. 2.6% across all industries). Despite the strong job market, year-over-year wage growth clocked in at 3.3% in July, trailing the overall private sector (+4.4%).

• Idonia Installs Cloud Exchange in Florida: Medical data exchange developer Idonia has signed a deal to install its Magic Link software at five facilities operated by Precision Imaging Centers in Florida. Magic Link is a cloud-based platform that enables physicians to share radiology and pathology images and reports as an alternative to sharing records on CDs. 

• Koning Signs Breast CT Partners: Breast CT developer Koning has signed on new partners as it moves to commercialize its Koning Vera Breast CT scanner. Digital radiography developer Swissray signed a deal to sell the scanner in Florida and Arizona, while at-home health testing company Healthyr is partnering with Koning to combine its testing services to detect early breast cancer. Koning is positioning Vera as a solution that can overcome some of the challenges faced by conventional X-ray mammography, such as in dense breast tissue.  

• Cardiac Imaging Shifts: MedAxiom’s 2022 Cardiovascular Provider Compensation and Production Report revealed continued shifts in cardiovascular imaging. CTA is enjoying the greatest growth among advanced imaging modalities since 2018 (from 10 to 16 exams per 1k patients), while cardiac PET and CMRI remained stable (22 to 24 & 4 to 4 per 1k patients). Meanwhile cardiac imaging staples echocardiography and nuclear SPECT are slowly trending in different directions since 2018 (509 to 514 & 117 to 89 per 1k patients).

A Better Way to Deploy AI

CARPL.ai’s end-to-end modular platform for AI deployment is changing how University Hospitals in Cleveland uses AI to enhance patient care. Hear from UH clinicians and administrators as they share their experiences.

Solutions to Transform Medical Image Delivery

Medical providers and health systems are looking to ditch the disc and modernize their patients’ journey. Learn how Clearpath helps them reduce the cost of retrieving and fulfilling patient requests for imaging and medical records.

The Case for Enterprise Cloud

Is your organization ready to move enterprise imaging to the cloud? Check out this Change Healthcare and ACHE webinar detailing cloud-native imaging’s benefits, best practices, and ROI.

The Resource Wire

• Allina Health needed to replace its on-premises PACS to better serve patients. Learn how Allina turned to Amazon Web Services (AWS) and Visage Imaging for a cloud-based implementation of the Visage 7 enterprise imaging platform. 

• What are some of the latest innovations in neuroimaging that can lead to workflow improvements at imaging facilities? Learn more in this presentation from the ASNR webinar series, supported by Subtle Medical. 

• More than 80% of health systems are reporting shortages in their radiology departments. Learn the countermeasures you can take to reduce burnout in this ITN Online article by Intelerad President Morris Panner. 

• What’s the best way to eliminate patient frustration and get them engaged with healthcare again? Find out how technology can revamp the patient engagement experience in this article produced in collaboration with Nuance in Becker’s Hospital Review.

• What are the most common AI applications in neuroradiology? This downloadable e-book from Bayer and Calantic Digital Solutions reviews the most common AI applications for brain imaging and the evidence behind them.  

• Transthyretin (ATTR) amyloidosis is emerging as an under-recognized form of heart failure, especially in the elderly. In this video, Marianna Fontana, MD, PhD, of UCL describes how Us2.ai’s echo AI technology was used to assess ATTR. 

• We talk a lot about radiology practices’ AI adoption, but usually don’t have much evidence to back it up. That changes with this Tempus Radiology report detailing how and why 30 US radiology groups became imaging AI adopters.

• Did you know that 43% of women recommended for diagnostic breast follow-up opt out of additional breast care? Find out how to bring them back with GE HealthCare’s Pristina mammography system with SenoBright HD contrast-enhanced mammography.

• A major US healthcare organization needed a solution to their patchwork system for image sharing with patients. Find out they simplified their system and saved over $100K in this case study from PocketHealth. 

• Join Merge on October 19 for a conversation on how to provide better patient outcomes and user experiences in mammography. Hear from thought leaders on the development and deployment of breast imaging solutions driven by patient needs.

• Data has become an invaluable part of medical imaging, but not all data is created equal, and data standardization has become a real issue. Learn more about tools for standardizing your data in this Imaging Wire Show interview with Beth Santori of Enlitic.

• What are the benefits of AI in radiology? Quite simply, they are game-changing and will drive improved hospital efficiency, according to this article that features Blackford Analysis CEO Ben Panter. 

• Teleradiology can be a force multiplier in radiology. Watch this Medality webinar recording with Daniel Corbett of Radiology Business Solutions to learn the pros, cons, and essential factors for private practice leaders considering teleradiology integration.

• There are three good reasons to add mammography services at your imaging site as providers expect higher demand for breast imaging services. Get the facts about mammography on this resource page from Siemens Healthineers.