Bernstein et al, in a new European Radiology study on incorrect AI.

Imaging Wire Sponsors

Annalise.ai  •  Arterys  •  Bayer  •  Blackford  •  CARPL.ai  •  Change Healthcare  •  Enlitic  •  GE HealthCare  •  Hyperfine  •  Intelerad  •  Medality  •  Merge by Merative  •  Nuance  •  Riverain Technologies  •  Siemens Healthineers  •  Subtle Medical  •  United Imaging  •  Us2.ai  •  Visage Imaging

Artificial Intelligence

When AI Goes Wrong

What impact do incorrect AI results have on radiologist performance? That question was the focus of a new study in European Radiology in which radiologists who received incorrect AI results were more likely to make wrong decisions on patient follow-up – even though they would have been correct without AI’s help.

The accuracy of AI has become a major concern as deep learning models like ChatGPT become more powerful and come closer to routine use. There’s even a term – the “hallucination effect” – for when AI models veer off script to produce text that sounds plausible but in fact is incorrect.

While AI hallucinations may not be an issue in healthcare – yet – there is still concern about the impact that AI algorithms are having on clinicians, both in terms of diagnostic performance and workflow. 

To see what happens when AI goes wrong, researchers from Brown University sent 90 chest radiographs with “sham” AI results to six radiologists, with 50% of the studies positive for lung cancer. They employed different strategies for AI use, ranging from keeping the AI recommendations in the patient’s record to deleting them after the interpretation was made. Findings included:

• When AI falsely called a true-pathology case “normal,” radiologists’ false-negative rates rose compared to when they didn’t use AI (20.7-33.0% depending on AI use strategy vs. 2.7%)
• AI calling a negative case “abnormal” boosted radiologists’ false-positive rates compared to without AI (80.5-86.0% vs. 51.4%)
• Not surprisingly, when AI calls were correct, radiologists were more accurate with AI than without, with increases in both true-positive rates (94.7-97.8% vs. 88.3%) and true-negative rates (89.7-90.7% vs. 77.3%)

Fortunately, the researchers offered suggestions on how to mitigate the impact of incorrect AI. Radiologists had fewer false negatives when AI provided a box around the region of suspicion, a phenomenon the researchers said could be related to AI helping radiologists focus. 

Also, radiologists’ false positives were higher when AI results were retained in the patient record versus when they were deleted. Researchers said this was evidence that radiologists were less likely to disagree with AI if there was a record of the disagreement occurring. 

The Takeaway 
As AI becomes more widespread clinically, studies like this will become increasingly important in shaping how the technology is used in the real world, and add to previous research on AI’s impact. Awareness that AI is imperfect – and strategies that take that awareness into account – will become key to any AI implementation.

Imaging AI’s Next Wave

We may be entering a third wave of imaging AI’s rapid evolution, one that brings a shift from narrow point solutions to comprehensive multi-finding AI systems. Hear this discussion with Annalise.ai Chief Medical Officer Rick Abramson, MD, exploring how this transition could take place, how radiologist and VC perspectives on AI are changing, and how AI might continue to evolve in the future.

Use of AI in Clinical Domains

If you’re at the AI Med conference in San Diego this week, be sure to attend a case study on the use of AI in clinical domains on June 6, with a stellar speaker lineup that includes Pamela Habib, MD, head of commercial development at the Digital Solutions Business, Americas at Bayer.

The Wire

• CMS to Cover AD Drugs: CMS confirmed that it will provide Medicare and Medicaid reimbursement for a new class of Alzheimer’s disease drugs if they have FDA approval and patients participate in a registry. The move improves the commercial prospects of drugs like Esai’s Leqembi that target amyloid buildup in the brain. Growing use of amyloid-lowering drugs could be a boon for neuroimaging, as Leqembi’s approval requires patients to undergo a PET scan for diagnosis and brain MRIs for treatment monitoring.

• Mixed Results with CTPA Dashboard: In JACR, researchers from New York saw mixed results with a clinical dashboard designed to reduce overuse of CT pulmonary angiography exams for pulmonary embolism that don’t comply with clinical guidelines. In 201,912 patients with suspected PE presenting to emergency rooms from 2017-2019, the dashboard improved guideline concordance (77.5% vs. 66.9%), but order rates per 1,000 patients also increased for CTPA exams (18.4 vs. 17.1) and D-dimer tests (37.3 vs. 30.6). 

• United Imaging Installs CT: United Imaging Healthcare has made the first US installation of a new 4-cm version of its uCT ATLAS CT scanner, at Russellville Hospital in Alabama. United first launched uCT ATLAS in a 16-cm configuration, and at RSNA 2022 debuted a 4-cm, 160-slice version that creates a new entry point for the technology. All models of uCT ATLAS include 0.25-second gantry rotation and an integrated uAI Vision 3D camera. 

• Diabetes Takes Toll on Brain Age: People with type 1 diabetes had brains that appeared 6 years older on 3-tesla MRI scans than normal controls, says a new study in JAMA Network Open. Researchers scanned 416 people with T1D from 2018 to 2019 and used machine learning to compare their brain structure to 99 normal controls. While T1D subjects had older-appearing brains (β=6.16 vs. β=1.04), they did not have signs of brain atrophy in areas typically affected by Alzheimer’s disease. 

• FDA Clears Ezra Fast MRI: Direct-to-consumer MRI cancer screening company Ezra said the FDA has cleared its Ezra Flash AI software for brain imaging, a move that enables the company to roll out a 30-minute full-body MRI scan. Ezra Flash enhances MR image quality to support faster image acquisition; as a result Ezra cut its price for a full-body MRI by 30%, to $1,350. Ezra previously received clearances for prostate image analysis and reporting. 

• No BPE Link to Breast Cancer: A new study in Clinical Radiology is raising questions about whether background parenchymal enhancement (BPE) on breast MRI scans is an accurate predictor of breast cancer. Researchers analyzed 327 breast MRI scans acquired from 2007-2016 of women at high risk of breast cancer; they found no correlation between BPE and breast cancer. The study contradicts previous research suggesting BPE could be a biomarker for breast cancer.  

• Philips Adds Xenon MRI: Philip has signed a deal with Polarean to give its 3-tesla MR 7700 scanner the ability to perform hyperpolarized xenon imaging for lung ventilation studies. Polarean’s Xenoview technology – which was cleared by the FDA in December 2022 – will enable Philips MRI customers to perform scans in which patients inhale hyperpolarized xenon-129 gas that’s distributed throughout the lungs and can be visualized on MRI. Philips is highlighting the relationship at this week’s ISMRM 2023 meeting in Toronto. 

• Qure.ai Partners with InHealth: AI developer Qure.ai is partnering with InHealth in the UK on the deployment of Qure.ai’s qXR software for classifying chest X-ray images into normal and abnormal exams. InHealth will deploy the software across its entire teleradiology service arm, identifying cases that are clear of clinically relevant findings to help radiologists read them more accurately and quickly. The deal illustrates AI’s growing role in helping radiologists handle large volumes of medical images.  

• New Model for CT Lung Screening: A new model for CT lung cancer screening is being implemented in Oklahoma, where Oatmeal Health has partnered with Stigler Health & Wellness Center, a large rural community health center. Oatmeal identifies eligible patients and facilitates their referral to screening sites, then leverages a virtual nodule clinic model and supplemental AI image evaluation to ensure proper care and follow-up. Such partnerships could boost low rates for CT lung screening, currently mired in the single-digit range.

• CARPL.ai Deploys AZmed in Ohio: CARPL.ai has deployed software from AZmed at University Hospitals in Cleveland. The hospital used the CARPL.ai platform to implement AZmed’s Rayvolve tool for fracture detection, giving UH radiologists seamless access to the application through existing clinical workflow. The deployment builds on a relationship between CARLP.ai and UH that was announced earlier this year. 

• Volpara Joins CancerX: Breast density software developer Volpara Health Technologies has joined CancerX, the Biden Administration’s public-private partnership to spur cancer research innovation. Volpara is one of 91 companies, universities, and healthcare organizations named as founding members, and will help guide CancerX’s research portfolio and demonstration projects and participate in collaboration opportunities. The administration launched CancerX in February as part of its Cancer Moonshot project to speed cancer research and cut cancer death rates by 50% over 25 years.

The Forefront of Neuroscience Research

The University of Iowa MR Research Facility (MRRF) is at the forefront of neuroscience research using ultra-high-field MRI. Find out how MRRF opened up new research capabilities by upgrading to GE HealthCare’s Signa 7.0T platform.

Crouse Hospital Untethers Cardiology with CVIS

Crouse Hospital is a nationally recognized cardiac care center in the Syracuse, NY, area, but the hospital’s cardiovascular service relied on separate data islands. That is, until Crouse Hospital adopted the HealthView CVIS from Intelerad’s Lumedx business.

RadNet’s AI-Powered MRI Efficiency

Efficiency and quality are the name of the game at RadNet, and that’s exactly what the imaging center giant achieved when it adopted Subtle Medical’s SubtleMR solution, optimizing its already-accelerated MRI protocols by 33-45% while maintaining consistent diagnostic image quality.

The Resource Wire

• How can your imaging facility create a foundation for MRI safety? Learn from MRI safety expert Tobias Gilk on the best practices for creating a safe environment in this podcast with Medality. 

• Imaging AI is evolving fast, but radiology leaders’ expectations for their AI technologies might be evolving even faster. In this Imaging Wire Show with Dr. Charlene Liew of SingHealth and Dr. Nina Kottler of Radiology Partners, we explore radiology leaders’ current and future expectations for AI, and the central role platforms play in their AI roadmaps.

• See how cloud-native imaging avoids traditional software’s resource utilization constraints and eliminates unexpected disruptions in this Change Healthcare animation.

• Imaging AI deployments face a long list of challenges that often emerge before any value is delivered. This Enlitic post details the top 10 AI deployment challenges organizations must understand in order to make sure their own deployments are successful.

• 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.

• MRI machines require 7,000 tons of helium every year, with per-machine helium costs of up to $39k annually – but it doesn’t have to be that way. See how the low-field Hyperfine Swoop avoids helium and infrastructure costs, while bringing MR neuroimaging into completely new clinical settings.

• This Blackford Analysis video details how imaging AI can improve radiology efficiency and patient care, and discusses the key role that AI assessments and curation plays in achieving these improvements.

• If you’re in the business of using or providing AI, there’s a good chance you spend a lot of time managing AI evaluations. But are your evaluations as efficient or effective as they could be? Check out this Imaging Wire Show with Riverain Technologies CEO, Steve Worrell, detailing the best practices for mitigating AI adoption risks, today and into the future.

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

• Join the conversation in this June 7 webinar and hear from PACS administrator Griff R. Van Dusen of Memorial Health System how Nuance PowerScribe One’s next-generation reporting experience helps streamline workflow and improve report quality so radiologists can get more done in less time. 

• The One Viewer philosophy from Visage Imaging 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.

• New technology from Us2ai called Us2.connect allows you to add AI automation to any echo device. Any echo machine can now have 100% automated reporting with disease detection and editable measurements – all generated in realtime as you scan.

• The number of patients eligible for low-dose CT lung cancer screening has expanded, and so has the need to reach at-risk patients closer to where they live. That’s why Siemens Healthineers’ Mobile Lung Screening Solution combines the quality, ease of use, and flexibility needed to create a program that meets the real-life needs of your community.

• See how United Imaging’s uCT ATLAS combines advancements in image quality, patient comfort, and operator efficiency to bring you one step closer to your masterpiece.