About AI Users | Fixing Medicare | DBT+US

“Those are informaticists and they oftentimes don’t even know that they are.”

Duke’s Christopher Roth, MD on the growing number of informaticists who haven’t figured out they’re informaticists.

Imaging Wire Sponsors

Arterys | Bayer Radiology | Canon Medical Systems | GE Healthcare
Healthcare Administrative Partners | Hitachi Healthcare Americas
Novarad | Nuance | Riverain Technologies | Siemens Healthineers
United Imaging | Zebra Medical Vision

The Imaging Wire

AI Users & Usage

We talk/write a lot about the technical side of imaging AI (e.g. training, testing, funding, approvals) but rarely about how AI is being used or how radiologists feel about it. That makes sense given where we are in AI’s evolution, and it’s also what makes a new survey from the ACR Data Science Institute so valuable:

  • The Survey – The ACR DSI surveyed 1,427 U.S. radiologists in May 2020, finding that 493 (33.5%) were actively using AI, and then performed a supplemental survey with 366 of those AI users.
  • About the Non-Adopters – A notable 80% of the practices that aren’t using AI don’t plan to adopt the technology within the next five years, in part because they “see no benefit” in AI (80%), can’t “justify the expense” (31%), and are concerned that it will hurt their efficiency.
  • Large Practices Lead – AI adoption is far more advanced at larger practices, as groups with more than 100 radiologists had a 39.8% AI adoption rate, while just 18.8% of practices with 1-5 radiologists were using AI.
  • Narrow Use – The AI adopters used an average of just 1.2 AI products each, spanning just 40 of the 80+ FDA-cleared algorithms. Interestingly, the radiologists used 27 internally-developed AI tools, giving homegrown tools a higher overall utilization rate (9.8%) than any single commercial segment (mammography AI was 2nd at 9%).
  • Interpretative AI – The radiologists primarily used AI to support interpretation (52%), followed by worklist management (15%), image enhancement (11.9%), operations (11%), and image measurements (9.9%).
  • Inconsistent AI – The vast majority of radiologists stated that their AI tools had “inconsistent” performance (461, 94.3%), which they largely attributed to the potential impact of biases (patient, scanner, and conspicuity).
  • AI Trust – However, most AI users were satisfied with their overall experience and found that AI provided value to both them and their patients.
  • AI Preferences – The radiologists named lesion detection and anatomic measurements as the applications that would most benefit from AI-based improvements (73% & 71% of respondents), followed by worklist prioritization (62%), image quality (58%), and departmental processes (51%).
  • AI Needs – When asked what they need from the ACR DSI, over 60% of the radiologists asked for tools to evaluate AI performance with representative data sets, 50% asked for improved reimbursements, over 40% wanted ways to measure ongoing AI accuracy, and more than 25% wanted methods to help them create their own AI.
  • AI Opportunity – The paper estimated AI’s current US market penetration at just 2%, suggesting that full market penetration would mean 560,000 algorithms in use across the country.

The Wire

  • Imaging Wire Q&A – Looking Forward to SIIM21: Like many in healthcare, informatics professionals just made their way through an unprecedented year that included plenty of imaging problems to solve and lessons to learn, but very few opportunities to meet with their peers and improve together. That’s about to change at the upcoming SIIM21 annual meeting, and there’s going to be a lot to talk about. Check out the latest Imaging Wire Q&A, where Duke’s Christopher Roth, MD shares what he’s expecting at SIIM21, what makes this year’s meeting different, and how informatics professionals can make the most of it.
  • Breast Ultrasound’s AI Classifier: Research presented at ARRS 2021 detailed how AI can classify breast cancer subtypes using standard ultrasound exams, suggesting that techniques like this could help improve treatment planning and outcomes. The researchers trained two ML models using the radiomic features from B-mode ultrasound scans to detect two aggressive breast cancer subtypes (triple-negative and HER2) with 0.824 and 0.778 respective AUCs.
  • Philips’ CT Tube Case Proceeds: Philips’ trade secret lawsuit against a pair of ex-employees and two associated Chinese companies will proceed after a U.S. federal judge denied the defendants’ motion to dismiss the case. Philips alleges that the former employees left with 800 files containing proprietary information about Philips’ CT tubes and provided them to their new employer, Illinois-based GL Leading Technologies, which later created two companies in China (Kunshan Yiyuan Medical Technology and Kunshan Guoli Electronic Technology) to produce alternative Philips CT tubes.
  • A Transcranial ULM Breakthrough: French researchers performed the first microscopic mapping of the human brain’s vascular network using a new transcranial ultrafast ultrasound localization microscopy technique that involves injecting patients with microbubbles. The new ULM technique could allow scientists to better understand brain haemodynamics and the relationship between neurological diseases/disorders and vascular abnormalities.
  • DBT+US’ Recall Advantage: When women with dense breasts receive breast ultrasound exams in addition to DBT they have far less abnormal findings, suggesting that this approach could lead to lower recall rates. That’s from a study presented at ARRS 2021 that reviewed 12,649 DBT screenings (8,534 w/ DBT+US), finding that DBT+US produced half as many abnormal results than DBT alone (rates: 6.7% vs. 14%) among women with dense breasts, while detecting 2.9% additional suspicious lesions that DBT missed.
  • Inefficient in Alberta: Alberta’s auditor general found that the Canadian province’s faulty CT and MRI ordering processes (inconsistent guidelines, inefficient scheduling, inflexible funding, poor communication, manual processes, etc.) led to “thousands of examples” of misprioritized exams, unnecessary scans, and long backlogs.
  • Fixing Medicare: A new USC-Brookings paper detailed how to fix the “dysfunctional” Medicare advanced imaging payment policies that drove major volume increases and imaging site shifts, with little evidence that these changes “were anticipated or planned.” To fix these issues, the paper recommended that CMS: 1) Create a more systematic method to establish advanced imaging payment levels based on current data; 2) Align payment levels for physician offices and outpatient locations (PFS & OPPS) to avoid “widely diverging fees” that historically drive imaging site shifts; 3) Apply prior authorization requirements for services paid under the PFS and OPPS, especially if Appropriate Use Controls prove to be ineffective.
  • Brain MRI AI Triage: A new study in Radiology: Artificial Intelligence detailed a deep learning approach that’s able to differentiate brain MRI scans as “likely normal” or “likely abnormal,” suggesting that a similar approach could be used for brain MRI triage. The researchers trained and tested three CNNs using T2-weighted FLAIR images (from two continents, with a range of pathologies), with the best performing model achieving an F1-score of 0.72 and an AUC of 0.78.
  • Akumin Acquires: Relatively large U.S. Imaging center company, Akumin, acquired six Florida imaging centers for $39m, which will give it 133 imaging centers in 7 states and expand its presence in Florida to 78 locations. The acquisition also appears to mark Akumin’s return to growth mode, after making one major acquisition in 2019 (+27 centers, $214m) .
  • A Compression Solution: A Pittsburgh-based inventor unveiled the “Bra-Shapes Imaging Device” that “better conforms to a woman’s anatomy to make mammography less painful.” Although details are very limited, the new device apparently supports different breast sizes to reduce discomfort from compression, and is being made available to medical device companies.
  • DECT for BME: Dual-energy CT (DECT) could serve as an MRI alternative for detecting vertebral bone marrow edema (VF BME), particularly in emergency settings. That’s from a recent study review published in the Radiology Journal (n = 17 studies, 742 of 2,468 vertebrae with BME) that found DETC confirmed VF BME with high specificity (96% overall, 98% w/ bone and soft-tissue kernels), although some studies showed that DECT’s specificity was heavily dependent on reader expertise (96% vs. 79%).
  • Radiology Salaries Up: AuntMinnie.com’s SalaryScan revealed that the COVID pandemic didn’t negatively impact U.S radiologist or technologist salaries in 2020, both of which posted overall increases. The survey showed radiologists’ average base salary reached $422k (vs. $408k in 2019), with subspecialty averages ranging from ~$408k for mammography to $439k for IR. Radiologic technologists also hit a new peak in 2020, with average RT salaries on the West Coast surpassing $100k for the first time, while other regional averages ranged from $68.7k (Midwest) to $82.5k (Northeast).
  • The Case for Survivor Screening: Patients with histories of non-skin malignancies have a similar risk of developing lung cancers and extrathoracic malignancies as lung cancer screening program participants who don’t have history of cancer, making it “imperative” that cancer survivors are considered for LDCT screening. That’s from Brigham and Women’s Hospital researchers who reviewed 5,835 LDCT screenings, finding that patients with and without previous malignancies had similar rates of lung cancer (1.4% vs. 1.9%) and non-lung malignancies (1.5% vs. 2%), contradicting previous studies that found cancer survivors had lower future lung cancer risks.
  • SenseTime’s CXR CE: Hong Kong-based AI developer SenseTime’s SenseCare-Chest DR Pro chest X-ray triage AI tool just gained its CE Mark, joining its CT-based SenseCare-Lung Pro solution in the European market. SenseCare-Chest DR Pro is primarily used to classify normal and abnormal CXR images, but it can also detect and locate lesions/abnormalities and produces quantitative analysis and text descriptions for each scan.

The Resource Wire

– This is sponsored content.

  • Learn how Salem Regional Medical Center improved its radiology workflows and cut service and syringe costs after adopting Bayer’s MEDRAD Stellant FLEX system.
  • Check out how Zebra Med’s Chest Solution prioritizes chest X-ray worklists, while providing specific insights into each study.
  • Did you know roughly half of all follow-up recommendations are never completed? Check out this Nuance panel presentation, where providers share how their organizations are addressing failed follow-ups to ensure that patients don’t fall through the cracks.
  • This new ebook looks at how you can empower your CT technologist team and set up your department for success by looking at four key areas.
  • Live Site Planning Sessions are one of United Imaging’s favorite collaboration moments. Their team comes on-site to take as-built measurements of the Medical Imaging suite and then creates a preliminary drawing for live discussion, giving customers and their design team a chance to adjust equipment placement, make changes to the floor plan, and review equipment specifications in a live virtual environment, saving time in the preliminary drawing process.
  • 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.

Get every issue of The Imaging Wire, delivered right to your inbox.

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!