COVID Severity AI | Risk Milestone | ED Disparities

“Something else is going on here that’s beyond the clinical, that’s beyond the diagnoses,”

UPMC Children’s Hospital’s Jennifer Marin MD, after her study revealed major emergency imaging disparities between Black, White, and Hispanic children.

We’re very happy to welcome Novarad as a Premium Imaging Wire sponsor. Novarad’s enterprise imaging, AR, and diagnostic solutions are changing the lives of providers and patients every day, and I can’t wait to share their message with you.

Imaging Wire Sponsors

  • Arterys – Reinventing imaging so you can practice better and faster.
  • Bayer Radiology – Providing a portfolio of radiology products, solutions, and services that enable radiologists to get the clear answers they need.
  • GE Healthcare – Enabling clinicians to make faster, more informed decisions through intelligent devices, data analytics, applications and services.
  • Healthcare Administrative Partners – Empowering radiology groups through expert revenue cycle management, clinical analytics, practice support, and specialized coding.
  • Hitachi Healthcare Americas – Delivering best in class medical imaging technologies and value-based reporting.
  • Novarad – Transformational imaging technologies that empower hospitals and clinicians to deliver clinical, operational and fiscal excellence.
  • Nuance – AI and cloud-powered technology solutions to help radiologists stay focused, move quickly, and work smarter.
  • Riverain Technologies – Offering artificial intelligence tools dedicated to the early, efficient detection of lung disease.
  • Siemens Healthineers – Shaping the digital transformation of imaging to improve patient care.
  • Zebra Medical Vision – Transforming patient care with the power of AI.

The Imaging Wire

AI-Severity Score

AI and federated learning startup, Owkin, developed a multimodal COVID-19 severity prediction system that outperforms current systems by analyzing a combination of clinical and imaging variables. Here’s how they did it:

  • The Study Group – The team collected 58 clinical/biological variables, CT scans, and radiology reports from 1,003 COVID-infected patients admitted at two French hospitals (931 patients w/ both clinical + CT data).
  • Verifying Variables – The researchers first set out to identify the variables associated with COVID severity, which included 12 clinical variables (e.g. age, sex, oxygen saturation, diastolic pressure, respiratory rate, others) and three radiology report variables (more severe = extent of disease, crazy-paving lesions; less severe = peripheral distribution of lesions).
  • Developing a CT DL – They then developed a deep learning model using unannotated CT images from 646 patients, which predicted high-severity patients from each hospital with higher AUCs (0.76 & 0.75) than the radiology reports (0.73 & 0.66). They also found that the CT scans captured the same key clinical variables detailed above (e.g. crazy-paving lesions), reinforcing the value of CT images in multimodal COVID AI severity scoring.
  • Multimodal AI-Severity System – The team finally built their AI-Severity scoring system using five of the clinical/biological variables (age, sex, urea, platelet and oxygen saturation) and the CT DL model, finding that this CT + clinical AI combo was slightly more accurate (+ 0.03 AUC) than just using clinical data. The multimodal system also outperformed 11 current COVID risk systems, achieving 0.05 to 0.28 AUC improvements.
  • The Takeaway – The study’s main takeaway is that adding CT data to clinical/biological data improves COVID AI severity risk predictions, and using CT deep learning analysis can be even more valuable than using radiology reports because of their reproducibility, speed, and accuracy advantages. This study also contributes to the growing research momentum that we’re seeing towards combining imaging and clinical/biological data in AI (including these studies, and the one right below this).

MIT’s Risk Breakthrough

MIT’s CSAIL team reached another mammography AI milestone, developing a risk-assessment algorithm that proved accurate and consistent across diverse populations and soundly outperformed current risk assessment methods. Here are some details:

  • CSAIL History – The MIT CSAIL team’s new Mirai algorithm expands upon their earlier breast cancer risk efforts, including a hybrid deep learning model (DM + risk factors) that beat the established Tyrer-Cuzick assessment method (0.70 vs. 0.62 AUCs) and was as able to predict breast cancer among Black and White women with the same accuracy (both 0.71 AUCs).
  • Mirai Improvements – With an eye on clinical adoption, Mirai comes with algorithmic improvements (210k training set, optional support for more clinical risk factors) and large-scale validation to prove its robustness (validated with datasets from USA, Sweden, and Taiwan). It can also predict a patient’s risk of developing breast cancer across multiple future time points (1, 2, 3, 4, 5-years) to help guide screening plans.
  • Mirai Results – These improvements worked well, as Mirai was significantly more accurate than prior methods at predicting cancer risk in all three global datasets (0.76 – 0.79 AUCs), while identifying two times more future cancer diagnoses than the Tyrer-Cuzick model. The Mirai model also achieved similar accuracy across different patient groups (races, ages, density categories) and different cancer subtypes.
  • Next Steps – The team’s long-term goal is to make this type of breast cancer risk assessment “part of the standard of care,” but their short-term focus is still on validation, including a clinical integration at MGH and more global validation studies. They also plan to develop future models that utilize patients’ entire imaging history and add support for DBT mammography.

The Wire

  • ED Imaging Disparities: Black and Hispanic children are 18% and 13% less likely to receive diagnostic imaging during ED visits compared with White children. That’s from a new JAMA Network Open study (13m pediatric ED visits, 3.69m w/ imaging, 44 children’s hospitals, 2016-2019), finding that imaging was far more likely to happen during White patients’ visits (33.5%) than Black and Hispanic patients (24.1% & 26.1%), suggesting that ED imaging may be both overused on White patients and underused on minority patients.
  • HHS’s AI Strategy: The U.S. HHS just released its official healthcare AI mission and strategy statement, calling this the “first step towards transforming HHS into an AI fueled enterprise.” HHS plans to drive AI advances within the department and across the U.S. healthcare system, serving as an AI regulator, investor, convener, and catalyst. Within those roles, HHS will focus on: 1) Developing an AI-ready workforce/culture; 2) Encouraging AI innovation and development; 3) Democratizing foundational AI tools and resources; 4) Promoting trustworthy health AI use and development.
  • COVID Vaccine Imaging: A new paper in Clinical Imaging details how medical imaging can play a role in the COVID pandemic’s vaccine research and rollout phases including: 1) Using molecular imaging to research immune cell dynamics within the host body; 2) Using serial CT to research vaccine efficiency (particularly detecting GGOs); and 3) Using FDG PET to measure post-vaccination lymph node activation. The paper also emphasized the importance of taking patients’ vaccination histories before imaging exams, since post-vaccination immune responses “might mimic various confusing imaging patterns” that create challenges for reading radiologists.
  • Hologic’s Ultrasound Expansion: Hologic continued its ultrasound portfolio expansion, adding the new “cost-optimized” SuperSonic MACH 20 below its existing SuperSonic MACH 30 and MACH 40 ultrasounds. The SuperSonic MACH 20 supports a range of applications (breast, liver, muscles, and tendons), and offers B-mode imaging as well as the SuperSonic lineup’s trademark ShearWave PLUS elastography for tissue stiffness evaluation.
  • Incidental Follow-up Gap: A new ACR survey of 247 radiologists and 145 emergency physicians revealed that most believe their hospitals have established follow-up guidelines for incidental findings (86% rads), but few of their hospitals actually close the loop on these findings (26% rads) or have follow-up tracking systems in place (23% rads, 36% EPs). The survey also uncovered a significant disconnect regarding who is responsible for incidental follow-ups, with radiologists assigning more responsibility to the ordering EPs and the EPs assigning more responsibility to primary care physicians.
  • icobrain on Nuance AI Marketplace: icometrix’s icobrain software portfolio (MRI & CT-based solutions for MS, dementia, Alzheimer’s disease, stroke, TBI, and epilepsy) is now available on the Nuance AI Marketplace. icometrix’s addition to the Nuance AI Marketplace further expands its platform/partner network, which also includes Blackford, TeraRecon/Envoy AI, Guerbet, and Siemens Healthineers.
  • Vicariously Liable in NJ: A technologist-related lawsuit against Lourdes Medical Center will move forward after a New Jersey appellate judge reversed a previous dismissal, stating that the plaintiff didn’t require an “Affidavit of Merit” to pursue this case (AOM = an expert statement that a lawsuit has “merit”) because RTs do not require AOMs in negligence lawsuits (many other clinical roles do). The plaintiff was injured during an imaging exam when a technologist asked him to ‘hold weights contrary to the [ordering physician’s] instructions,’ leading to the now-reinstated vicarious liability lawsuit against Lourdes Medical Center (and not the RT).
  • COVID’s Elective Impact: A new study in the Annals of Surgery revealed that delayed electives cost U.S. hospitals $22.3b during the March-May 2020 shutdown, estimating that it will take most hospitals 12 to 24 months to return to pre-pandemic elective volumes. The study found elective revenue declines across all hospital types, although already vulnerable hospitals (urban non-teaching, rural) were more impacted by the elective delays.
  • qEASL’s Advantages: There are a number of tumor imaging response criteria used for the assessment of hepatocellular cancer after transarterial chemoembolization therapy, but the qEASL method is the most effective (qEASL = MRI-based whole liver volumetric enhancement quantification). That’s from a new JACR study that found qEASL beat the alternative RECIST and mRECIST assessment methods based on quality-of-life years (1.06 vs. 1.02 vs. 1.05 QALYs) and costs, suggesting that the U.S. health system could eliminate $575M in annual costs if it standardized on qEASL.
  • When to Stop Screening BC Survivors: A new consensus guideline recommends stopping routine mammography screening for >75yr-old breast cancer survivors when their life expectancy reaches <5 years, while advising physicians to “consider” discontinuing mammography when these patients have 5-10 years left. However, the guidelines recommend continued mammography screening for >75yr-old breast cancer survivors with >10yrs left.
  • Siemens’ Big Manchester NHS Deal: Siemens Healthineers landed a massive 15-year, €140M value partnership with Manchester University NHS Foundation Trust (9 hospitals) that will start with the installation and maintenance of 222 devices and will eventually include roughly 360 installations.
  • COVID Follow-up Findings: A new RSNA study (n = 114 patients) revealed that 35% of patients who recover from severe COVID-19 infections show lung fibrotic-like changes in their CT scans 6 months after symptom onset. These patients generally had more severe cases and/or greater risk factors (e.g. age, acute respiratory distress syndrome, longer in-hospital stays, higher chest CT scores).
  • Diagnostic Error Initiative: Patient safety watchdog, The Leapfrog Group, just launched its new “Recognizing Excellence in Diagnosis” initiative. The 2-year, $1.2M project will create and distribute guidelines/resources intended to help prevent diagnostic errors, and recognize high-performing hospitals.

The Resource Wire

– This is sponsored content.

  • Novarad’s COVID-19 AI Diagnostic Assistant analyzes chest CT scans in seconds, helping physicians quickly identify COVID-19 patients and get them into care. The best news – it’s available to clinicians worldwide free of charge.
  • This Riverain Technologies case study details how Einstein Medical Center adopted ClearRead CT enterprise-wide (all 13 CT scanners) and how the solution allowed Einstein radiologists to identify small nodules faster and more reliably.
  • Zebra-Med’s Bone Health Solution helps radiologists automatically identify patients at risk for osteoporosis, allowing HMOs to enroll more patients in bone health programs.
  • Check out how the UMass AI Lab, an Arterys Center of Excellence, is implementing AI and radiomic methods for thoracic imaging.
  • With radiation dose management now largely considered best practice, this Bayer white paper details the top five benefits of adopting contrast dose management.
  • This Hitachi Healthcare blog outlines the criteria providers should consider for their image and reporting platforms, and how the Hitachi VidiStar platform’s features, service, and vendor collaboration meet providers’ needs.
  • Watch this recorded webinar from Healthcare Administrative Partners where they examine how the 2021 Medicare Physician Fee Schedule (MPFS) and Quality Payment Program (QPP) final rules will impact radiologists.

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!