Big AI Study, Debated Results | More Coronavirus Imaging

“. . . studies like this strongly suggest that radiologists will be masters of their domain for quite some time.”

University of Iowa radiology professor, Claudia Mello-Thoms, MS, PhD, sharing one of her takeaways from a recent study on how AI can help improve radiologist accuracy.


Imaging Wire Sponsors

  • Focused Ultrasound Foundation – Accelerating the development and adoption of focused ultrasound.
  • GE Healthcare – Providing point of care ultrasound systems, from pocket-sized to portable consoles, designed to support your clinical needs and grow along with your practice.
  • Healthcare Administrative Partners – Empowering radiology groups through expert revenue cycle management, clinical analytics, practice support, and specialized coding.
  • Nuance – AI and cloud-powered technology solutions to help radiologists stay focused, move quickly, and work smarter.
  • Qure.ai – Making healthcare more accessible by applying deep learning to radiology imaging .
  • Riverain Technologies – Offering artificial intelligence tools dedicated to the early, efficient detection of lung disease.

The Imaging Wire



Big AI Study, Debated Results

A massive study published in JAMA found that mammography screening algorithms can’t beat radiologists head-to-head, but AI can improve accuracy when an ensemble of algorithms is combined with radiologist assessments. Whether those improvements are all that impressive is up for debate.

  • The Study – The study analyzed over 310k screening mammograms using crowdsourced algorithms developed by 126 teams (144,231 scans / 85,580 women from the U.S.; 166,578 scans / 68,008 women from Sweden). These numbers make this one of the largest AI studies performed so far.
  • Results – The study’s top-performing algorithms achieved AUCs of 0.858 (U.S. dataset) and 0.903 (Swedish dataset) and specificity of 66.2% (U.S. dataset) and 81.2% (Swedish dataset), falling well short of the radiologists’ specificity of 90.5% (U.S. dataset) and 98.5% (Swedish dataset). However, the combination of an 8-algorithm ensemble with U.S. radiologist assessments resulted in solid improvements to AUC (0.942) and specificity (92%).
  • Takeaway – The authors touted these results as evidence of the “potential of using machine learning methods for enhancing mammography screening interpretation.” However, a related JAMA response editorial countered that the study mainly “showed that AI is not there yet,” pointing out a number of shortcomings (e.g. not better than double reading, couldn’t show locations, wouldn’t continue to learn) and assuring any AI-concerned rads that “studies like this strongly suggest that radiologists will be masters of their domain for quite some time.”



More Coronavirus Imaging

The spread of coronavirus is quickly becoming one of the biggest health and business issues in years, and although its impact extends far beyond radiology, the disease remained a major imaging topic again this week. Here’s the latest news on that front:

  • X-Ray Insufficient – It’s pretty clear by now that chest CT is the go-to option for coronavirus imaging, but a new study out of South Korea revealed that chest X-ray isn’t nearly as good of an alternative as some might have believed. The study reviewed CT and CXR scans from nine patients with COVID-19, finding that only three patients had suspicious chest X-rays (no signs of COVID-19), while the CT scans showed signs of COVID-19 in all nine patients.
  • COVID-19 ExplosionIn a recent interview, Yale radiology professor Howard P. Forman MD forecasted an “explosion in” identified COVID-19 across the U.S., arguing that coronavirus testing resources and information has been “woefully inadequate.” Dr. Forman suggested that this surge in patients will stretch healthcare resources, including radiologists who may be burdened by COVID-19 more than other illnesses due to the role of chest CTs in diagnosis.
  • ECR Postponed – Almost a week before ECR 2020, The European Society of Radiology postponed the 23k-person event until July due to coronavirus concerns. Although not shocking given the surge in convention cancellations and postponements, it’s still hard to believe that ECR was cancelled. It also brings more attention to HIMSS, which is set to start next Monday with a no-handshakes policy, and hasn’t been cancelled yet (although the number of exhibitor no-shows keeps growing).



The Wire

  • Another Nanox Win: Nanox kept its momentum going, announcing a $174M agreement with The Gateway Group to deploy 1,000 Nanox.ARC X-ray units across Australia, New Zealand, and Norway over the next three years. This comes less than a month after Nanox announced a 3,000-unit agreement with USARAD, and these two agreements (in addition to its big-name backers: Foxconn, Fujifilm, SK Telecom), suggest that Nanox could have a much greater market impact than many assumed when it came on the scene last year.
  • Value Despite Volume: A recent paper in Current Problems in Diagnostic Radiology detailed steps radiology practices can take to make sure they are providing value-based care as imaging volumes continue to rise. Here they are: 1. Reduce unnecessary imaging; 2. Reduce scan times by using abbreviated MRI and other methods; 3. Provide 24/7 radiologist coverage; 4. Effectively communicate results using available technology and staffing reading room assistants; 5. Work to avoid radiologist burnout.
  • US Funding Growth: Signify Research detailed the flow of capital towards diagnostic ultrasound startups, which peaked in 2018 ($390m, 26 deals) due to massive rounds from Butterfly Network and Aisono ($250m and $73m). Although funding activity fell in 2019 ($136m, 9 deals), diagnostic ultrasound startup funding has been on the rise over the last decade (2010-2014 annual avg: $64m; 2015-2019: $174m). Signify’s report goes a lot deeper than this if you’re interested in learning more.
  • ClearRead Detection: A team of Swiss researchers found that vessel suppression (VS) significantly improves nodule detection, interreader agreement, and reading time with oncologic chest CT scans. The researchers created vessel-suppressed reconstructions of 100 patients’ contrast-enhanced chest CTs using Riverain Technologies’ ClearRead CT software. The two sets of images were read by two groups of three radiologists, finding that VS-CTs had far greater nodule detection rates (+21%), higher interreader-agreement rates (k = 0.431 “moderate” vs. 0.209 “fair”), and shorter average read times (154 ± 134 vs. 194 ± 126).
  • TeraRecon’s AI Patents: TeraRecon was awarded a pair of imaging AI patents protecting technologies used by the company’s EnvoyAI and the Northstar AI Results Explorer platforms. The new U.S. patents also protect TeraRecon’s AI interoperability features for tracking/storing AI results and physician preferences.
  • A Call for More mpMRI: The nonprofit group Prostate Cancer UK called for more mpMRI-capable scanners and increased prostate imaging staff across Scotland, revealing that just 11 out of the country’s 14 NHS health boards are currently meeting PROMIS trial standard. The last three years already brought an “impressive” mpMRI rollout across Scotland, and earlier this year, NHS Scotland made mpMRI a standard procedure prior to prostate cancer biopsies.
  • Philips & Paracelsus: Philips announced a strategic partnership with Paracelsus Clinics that will modernize the Swiss healthcare provider’s medical imaging devices and workflow over the next eight years. The partnership will upgrade Paracelsus’ CT, MRI, angiography, X-ray, and ultrasound systems, while also bringing in Philips’ Informatics AI Research Suite (IntelliSpace Discovery).
  • Curium Buys Zevacor: Curium announced its acquisition of Zevacor Molecular, expanding the large nuclear medicine company deeper into the cardiac PET segment. Zevacor is a leading producer of Strontium-82, which is used in the production of Rubidium-82 generators for cardiac PET imaging.
  • Salaries Up: AuntMinnie.com’s SalaryScan found that U.S. radiologists earned a $408k average base salary in 2019 (vs. $394k in 2018), with modality subspecialties ranging from ~$405k averages for ultrasound and MRI to $442k for mammography. Radiologic technologists also hit a new peak in 2019, earning a $74k average base salary, with subspecialties ranging from $71k for mammography to $86k for nuclear medicine/PET.
  • Consultation Works: A study out of New York’s Montefiore Medical Center found that implementing preliminary consultations before nuclear medicine therapy improved patients’ understanding and helped them plan for their treatment. A survey of 38 patients before and after their consultations found that these talks improved: patient familiarity with the term “nuclear medicine doctor” (44.7% vs. 86.8% answered “somewhat or extremely”), understanding of their treatment plan (37.1% vs. 5.4% answered “no or vague understanding”), and emotional state (68.4% vs. 91.9% answered “generally or perfectly calm”).
  • NorthStar’s New Agent: NorthStar Medical Technologies signed a global licensing agreement with Capella Imaging, giving NorthStar exclusive rights to Capella’s FibroScint (Tc-99m F4A) fibrin-targeted cardiac SPECT imaging agent. All though still at an investigational stage, FibroScint could eventually support a range of cardiac diagnoses (e.g. thrombus in heart failure patients with assisted devices, deep vein thrombosis, pulmonary embolism, and acute coronary syndrome).
  • Parkinson’s Tracer: German researchers reached a milestone towards developing a new PET tracer that could improve image-based differential diagnostics of Parkinson’s disease. The team of contest participants invented a stable PET-detectable molecule that can dock onto nerve cell receptors. The team beat out 21 other contestants and now plans to bring their potential radiotracer into clinical studies.

The Resource Wire

– This is sponsored content

  • Visit Healthcare Administrative Partners at the leading conference for radiology business professionals – RBMA PaRADigm 2020 in Hollywood Beach, Fl from April 5-8. They’ll be at booth #406 to answer your toughest revenue cycle questions.
  • ClearRead CT from Riverain Technologies is the first FDA-cleared system for the automatic detection of all lung nodule types, allowing radiologists to reduce search and reporting time and improve nodule detection rates.

Leave a Reply

Basic HTML is allowed. Your email address will not be published.

Subscribe to this comment feed via RSS

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

Join thousands of imaging professionals.