#370 – The Wire

  • The NTAP Effect: A new report from Vizient revealed that US hospitals significantly increased their use of ischemic stroke triage AI tools after CMS began awarding Medicare New Technology Add-On Payments in late 2020 (NTAPs – up to $1,040 w/ certain scenarios). The number of Vizient member hospitals using LVO AI triage tools doubled from Q4 2020 to Q2 2021 (63 to 130) and LVO AI procedures grew even faster (1,797 to 4,811), with 38% of these AI procedures resulting in mechanical thrombectomy.
  • GE’s Endotracheal Tube AI FDA: GE Healthcare announced the FDA 510(k) clearance of its AI tool used to assess Endotracheal Tube (ETT) placements in intubated patients, which automatically detects ETTs in chest X-rays and provides clinicians with positioning measurements. The ETT tool is one of five algorithms in GE’s Critical Care Suite 2.0, all of which come embedded on its AMX mobile X-ray systems. Given that ETT assessments are common and relatively mundane, this is an AI use case that many radiologists might welcome.
  • Telemedicine Leads to Fewer Treatments & Tests: When patients meet with their primary care providers via phone or video, they receive far fewer prescriptions and diagnostic tests. That’s from a pre-pandemic Kaiser Permanente study (n = 1.1m patients, 2.2 appointments) that found video and telephone visits were far less likely than clinic visits to result in prescriptions (38.6% & 34.7% vs. 51.9% of visits) or lab tests and imaging orders (29.2% & 27.3% vs. 59.3%). Although telemedicine visits were slightly more likely to require office follow-ups within 7 days (25.4% & 26% vs. 24.5%), they did not lead to more ED visits or hospitalizations.
  • Lunit’s CXR Triage FDA: Lunit announced the FDA 510(k) clearance of its INSIGHT CXR Triage tool, marking the South Korean startup’s first FDA approval and a major step in its US commercial rollout. As its name suggests, Lunit INSIGHT CXR Triage triages and prioritizes emergency chest X-rays and notifies physicians of urgent cases. Lunit now plans to accelerate its commercial activity in the US, leveraging its partnerships with Philips and Fujifilm.
  • Radiologist-Supervised Transfer Learning: A new UCSD study showed how AI algorithms can be improved through radiologist-supervised transfer learning. The researchers created 1,466 radiologist-annotated CXRs and used transfer learning to fine-tune an existing pneumonia localization CNN, and then tested it with CXRs from 203 COVID patients. This added transfer learning step improved the CNN’s performance for pneumonia detection (internal data AUCs: 0.756 to 0.841; external data AUCs: 0.864 to 0.876) and severity quantification, suggesting that closed-loop AI systems could be continuously improved by incorporating radiologist interpretations.
  • Medtech OS Vulnerabilities: Cybersecurity firms, Forescout Technologies and Medigate, uncovered 13 vulnerabilities affecting Siemens’ Nucleus Real-time Operating System that hackers could use to cause millions of medical devices to crash or expose data (including imaging devices from multiple brands). There’s no evidence that hackers have exploited these vulnerabilities and Siemens has already released several updates.
  • Mirai’s Broad Validation: A new study in the Journal of Clinical Oncology showed that MIT CSAIL’s Mirai mammography-based AI risk model was able to maintain its accuracy across an incredibly diverse test dataset (seven hospitals, five countries, 129k mammograms, 62k patients, 3,815 diagnosed with w/ cancer within 5yrs). Mirai predicted patient outcomes with 0.75 to 0.84 concordance indices across the seven hospitals, making this the broadest mammography AI validation ever, and suggesting that Mirai could allow broad and equitable improvements in care. This is the latest in a series of milestones that continue to show that Mirai is both accurate and generalizable.
  • MPFS 2022 Breakdown: Healthcare Administrative Partners provided a complete overview of how CMS’ final 2022 Medicare Physician Fee Schedule (MPFS) impacts radiology. Big picture: the conversion factor will fall by 3.71% to $34.89, bringing smaller-than-anticipated cuts to diagnostic radiology (-1%), IR (-8%), nuclear medicine (-1%), and radiation oncology/therapy (-5%). Practices will also see an additional 2% payment reduction when sequestration resumes on January 1, 2022. HAP did a deep dive on all of the 2022 MPFS changes, so check out their report to learn more.
  • Synthetic CT for FUS: A team of South Korean scientists developed an AI algorithm that can generate synthetic CT images from brain MRIs, finding that sCT could be effective for image-guided transcranial focused ultrasound treatments without exposing patients to CT radiation. There’s been quite a bit of synthetic CT research efforts in recent years, but they’ve mainly focused on sCT for radiation therapy planning.
  • Radiologist-Involved Concordance Reviews: A new Beth Israel Deaconess Medical Center study revealed that CT-guided biopsy results are often discordant with diagnostic imaging findings, and highlighted the benefits of involving radiologists in imaging/biopsy concordance reviews. Their analysis of 926 consecutive CT-guided biopsies (453 reviewed w/ radiologists) revealed that 11% of pathology results were discordant with diagnostic imaging (57% proving to be malignant). When reviewed by radiologists, discordant cases were far more likely to be evaluated via repeat biopsies (50% vs. 13% via repeat imaging), leading to faster diagnosis times (median: 41 days w/ rad review vs. 56 without rads).
  • Concussion App: A new NIH study suggests that pupillary light reflex measurements performed with smartphone apps might be able to identify concussed patients, improving diagnosis and treatment (and potentially reducing concussion/TBI imaging). The researchers analyzed data from patients who used the BrightLamp Reflex iPhone app, finding that patients with and without concussions had significantly different pupillary light reflex metrics (e.g. max & min pupil diameter, recovery time). BrightLamp Reflex joins a growing list of non-imaging approaches for concussion/TBI triage, although other options have generally focused on blood testing.

You're signed up!

It's great to have you as a reader. Check your inbox for a welcome email.

-- The Imaging Wire team