Ben Glocker, PhD, of Imperial College London, in a new study on foundation models for AI algorithm development.

Imaging Wire Sponsors

Annalise.ai  •  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 Hits Speed Bumps

There’s no question AI is the future of radiology. But AI’s drive to widespread clinical use is going to hit some speed bumps along the way.

This week is a case in point. Two studies were published showing AI’s limitations and underscoring the challenges faced in making AI an everyday clinical reality. 

In the first study, researchers found that radiologists outperformed four commercially available AI algorithms for analyzing chest X-rays (Annalise.ai, Milvue, Oxipit, and Siemens Healthineers) in a study of 2k patients in Radiology.

Researchers from Denmark found the AI tools had moderate to high sensitivity for three detection tasks: 

1. airspace disease (72%-91%)
2. pneumothorax (63%-90%)
3. pleural effusion (62%-95%). 

But the algorithms also had higher false-positive rates and performance dropped in cases with smaller pathology and multiple findings. The findings are disappointing, especially since they got such widespread play in the mainstream media. 

But this week’s second study also brought worrisome news, this time in Radiology: Artificial Intelligence about an AI training method called foundation models that many hope holds the key to better algorithms. 

Foundation models are designed to address the challenge of finding enough high-quality data for AI training. Most algorithms are trained with actual de-identified clinical data that have been labeled and referenced to ground truth; foundation models are AI neural networks pre-trained with broad, unlabeled data and then fine-tuned with smaller volumes of more detailed data to perform specific tasks.

Researchers in the new study found that a chest X-ray algorithm trained on a foundation model with 800k images had lower performance than an algorithm trained with the CheXpert reference model in a group of 42.9k patients. The foundation model’s performance lagged for four possible results – no finding, pleural effusion, cardiomegaly, and pneumothorax – as follows…

• Lower by 6.8-7.7% in females for the “no finding” result
• Down by 10.7-11.6% in Black patients in detecting pleural effusion
• Lower performance across all groups for classifying cardiomegaly

The decline in female and Black patients is particularly concerning given recent studies on bias and lack of generalizability for AI.  

The Takeaway

This week’s studies show that there’s not always going to be a clear road ahead for AI in its drive to routine clinical use. The study on foundation models in particular could have ramifications for AI developers looking for a shortcut to faster algorithm development. They may want to slow their roll. 

Monitoring AI Solutions in Healthcare

Clinical applications for healthcare AI are rapidly expanding, but many barriers are still preventing widespread adoption. This Nuance post explores a critical set of questions: what happens after an AI model goes into production, and how to know if it continues to perform as expected?

A Perfect Balance of Utility and Nirvana

The One Viewer philosophy from Visage offers end users from across the enterprise the chance to get access to Visage 7’s powerful tools based on their clinical need, with the same #1 rated viewer for multiple workflows. Find out today how it works.

The Wire

• Patient-Friendly Radiology Reports: Patients reported high levels of satisfaction with patient-friendly radiology reports that were created with Scanslated’s natural language processing software. In JACR, researchers surveyed 6.2k patients from March 2022 to February 2023 on their opinions of reports generated by the software, which uses patient-centered language to explain technical medical jargon and present it via online portals. Patients agreed or strongly agreed that they felt better prepared (81%), had a more productive conversation (87%), better understood treatment recommendations (77%), and felt reduced anxiety (70%). 

• Visage Lands Baylor Scott Deal: In one of the biggest enterprise imaging deals we’ve heard of, Visage Imaging has landed a contract worth $140M (Australian) to install its Visage 7 cloud-based image management software at Baylor Scott & White Health in Texas. The deal runs for 10 years, and Visage 7 will replace legacy PACS software at the health network. The contract joins other recent victories for Visage, including Gundersen Health System for $20M.

• UK to Study AI for Chest X-Rays: University College London Hospitals (UCLH) is launching LungIMPACT, a clinical trial to evaluate Qure’s qXR AI algorithm for analyzing chest X-Rays. qXR will flag X-ray exams that have abnormalities for urgent radiologist review, with the goal of speeding up processing of these studies. The UK has been making major investments in AI, such as its launch of the NHS AI Diagnostic Fund this summer. 

• Questions about CAC Scoring: A research letter in JAMA Internal Medicine raises the question whether CAC scoring leads to unnecessary interventions like statin prescriptions and stress tests in patients without symptoms or known atherosclerotic cardiovascular disease. Researchers analyzed records of 6.8k patients from the MESA study, finding that coronary calcium buildup is actually common as people age. Therefore, calcium findings on CT scans “should not reflexively result in a specialist referral or a prescription for a statin and/or aspirin.” Instead, clinicians should discuss risk with individuals.

• Nuance Starts Shipping DAX Copilot: Nuance Communications this week announced the general availability of DAX Copilot, AI-based software for automated clinical documentation first unveiled in March 2023 as DAX Express. DAX Copilot uses conversational, ambient, and generative AI to help doctors streamline the creation of medical documentation. DAX Copilot leverages generative AI technology from OpenAI, the generative AI developer in which Nuance parent Microsoft has an ownership position. 

• Rad AI Reveals Omni Reporting: On the same day as the Nuance announcement, Rad AI lifted the covers on its Omni Reporting software, which the company originally teased in late August. The company discussed how it leveraged its work in generative AI to develop the reporting solution, and promised additional details at RSNA 2023. 

• Sirona Telegraphs Platform Launch: Sirona Medical this week gave a sneak peak at its upcoming platform launch on November 13, when the company said it will debut its unified cloud-native radiology IT platform for image viewing, archiving, and reporting. The platform incorporates Sirona’s RadOS Euphrates radiology operating system, which includes features for data structuring, normalization, and AI orchestration. Sirona will also discuss its Unify workflow service, archiving for $0.33 per exam, and Amplify SDK developer ecosystem. Sirona sees AI as helping radiology thrive in the era of value-based care. 

• Exo Launches New POCUS: Exo has launched Exo Iris, a new handheld POCUS scanner that includes built-in AI support. Exo Iris is based on the company’s pMUT technology that combines the image quality of piezoelectric crystals with the affordability of silicon. On the software side, the scanner includes Exo’s SweepAI to automatically acquire images in a single sweep, while Bladder AI helps providers quickly determine bladder volume. More AI tools are pending FDA clearance, including cardiac and lung AI suites. Exo bought AI firm Medo AI last year.

• Medality Partners with Harbor-UCLA: Online education firm Medality has partnered with the radiology residency program at Los Angeles County Harbor-UCLA Medical Center. Medality will provide its simulated learning platform to Harbor-UCLA, which includes case-based learning in a microlearning format, assignments and visibility into resident activity, fully annotated teaching cases, and customized curriculum. 

• MRI Reveals Long COVID’s Toll: MRI is showing the toll that long COVID has on the body. In the C-MORE study in Lancet Respiratory Medicine, UK researchers used the modality to track 259 people with long COVID who were scanned from March 2020 to November 2021; they were more likely to have multi-organ abnormalities than healthy controls (61% vs. 27%, odds ratio=2.9), and also had higher rates of lung, brain, and kidney abnormalities. But the study’s protocol for selecting a control group drew criticism as part of a debate that erupted over research into long COVID.

• GE/Sirius Highlight CEM Breast Biopsy: GE HealthCare and surgical marker developer Sirius Medical Systems are highlighting their work together at this week’s EUSOBI 2023 meeting in Spain. The companies are collaborating in a workshop on breast biopsy guided by contrast-enhanced mammography (CEM) using Sirius’ Pintution device, a permanent magnetic seed that marks tumors to guide surgical removal without the use of guidewires. 

• McGinty Joins 4DMedical Board: Radiologist Dr. Geraldine McGinty has joined 4DMedical’s board of directors. McGinty is a practicing radiologist and professor of clinical radiology and population health sciences at Weill Cornell Medical Center; she was also the first woman elected chair of the ACR’s Board of Chancellors and also served as ACR president. 4DMedical recently landed the University of Miami as the first US site for its SaaS-based XV Technology for assessing lung ventilation, which it debuted at RSNA 2022. 

• HAP Urges Support for Medicare Reform: Revenue cycle management firm Healthcare Administrative Partners (HAP) is urging support for HR 2474, federal legislation that would modify how the Medicare Physician Fee Schedule (MPFS) is calculated. In a blog post, HAP explains how the current method for calculating the MPFS conversion factor restricts physician payments and doesn’t reflect the real cost of providing healthcare. It further notes that when adjusted for inflation the conversion factor has fallen sharply since 2001. 

• NHS Wait Times Skyrocket: Wait times for MRI exams in the NHS England have hit two-and-a-half years and are up nearly 50% since 2019. An analysis in The Guardian found that NHS England’s wait list in July 2023 had grown compared to June 2019 (1.6M vs. 1.1M); in addition to long MRI waits, patients can wait almost two years for an ultrasound scan and a year for a CT scan. The NHS has been turning to technologies like AI in an effort to reduce waiting times.

Overcoming the Radiologist Shortage

How can radiology practices use innovative training and education techniques to grow and overcome the ongoing shortage of radiologists? Find out in this Imaging Wire Show interview with Daniel Arnold and Deanna Heier of Medality.

The Benefits of AI-Automated Echocardiography

Us2.ai just published what might be the most comprehensive paper we’ve seen on AI echo, detailing the benefits of AI-automated echocardiography, the global need for more scalable and flexible CVD assessments, and how its technology is fit for the future.

Easily Quantify Cardiac Volumes

Tempus Radiology’s Cardio AI solution recently added a new Atrial Volumes feature that allows cardiologists and radiologists to easily quantify volumes for both left and right atria in cardiac MRI images.

The Resource Wire

• True time-of-flight (TOF) PET/CT based on lutetium oxyorthosilicate (LSO) detectors offers numerous clinical advantages, such as better definition of small lesions and improved uniformity. Learn the other advantages in this article from Siemens Healthineers. 

• Data standardization has emerged as a key issue for both clinical and research aspects of medical imaging. Learn about its pivotal role in healthcare in this Enlitic webinar on October 25 featuring radiologist Cheryl Petersilge, MD, founder and CEO of Vidagos. 

• The flow of new AI applications makes it hard for radiology groups to determine which tools would help them and how IT teams can handle increased AI adoption. In this Blackford Analysis white paper, radiology and IT leaders from NYU and Canopy Partners share how a platform approach alongside a curated marketplace can help solve these challenges.

• Still sharing patient images with CDs? It may be time to move on. Find out how PocketHealth can help you make the transition to fully digital image exchange in this article.

• Subtle Medical has been named one of the 50 most promising private generative AI companies by CB Insights, one of just four healthcare companies on the list. Learn more about the Subtle solutions that led to this honor. 

• AI in healthcare has grown quickly, as have discussions about AI regulation. Listen to this on-demand webinar from Bayer and Calantic Digital Solutions to hear expert perspectives on the current and future state of AI governance in healthcare and radiology.

• There is growing awareness that greater collaboration among stakeholders to show data can provide clinicians decision support for providing the right treatment to enhance patient care and outcomes. GE HealthCare shares more on how they are looking at cancer care with data-driven diagnostics and treatment.

• There are many things for a radiology practice to consider when it embarks on the journey to AI for decision support. In this article from Annalise.ai, clinical consultant Dr. Simon Edelstein talks about his experience. 

• SOIN Soluciones Integrales of Costa Rica turned to Merge enterprise imaging solutions from Merative when it wanted to modernize the imaging environments of 50 hospitals across the country. Download this PDF white paper to find out how they did it.

• Early detection is key to improving treatment pathways and patient outcomes in lung cancer. Learn more facts about lung cancer screening on this resource page from Riverain Technologies. 

• 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.

• Did you know one-quarter of healthcare organizations have experienced a cyber-attack in the last year? This Change Healthcare animation explains how third party-certified cloud-native enterprise imaging can help secure IT infrastructure that might be exposed with re-platformed imaging systems.

• Driven by a focus on R&D, United Imaging has increased its brand influence and market share worldwide. Get the details in the company’s first annual report since going public. 

• Generative AI has emerged as a powerful resource for a variety of tasks. Intelerad President Morris Panner surveys the medical imaging use cases for generative AI in this article in Healthcare IT Today.

• When SyntheticMR validated its SyMRI MSK solution, they leveraged the CARPL platform to compare conventional knee and spine MRI image quality with SyntheticMR images. Check out their validation process and results here.