A new Lancet paper outlining the path towards improving AI trust/performance, since explainable AI might not be realistic in the short term.

Imaging Wire Sponsors

Arterys  •  Bayer Radiology  •  Blackford Analysis  •  Canon Medical Systems  •  Fujifilm Healthcare  •  GE Healthcare  •  Novarad  •  Nuance  •  Riverain Technologies  •  Siemens Healthineers  •  United Imaging  •  Zebra Medical Vision

Artificial Intelligence

The False Hope of Explainable AI

Many folks view explainability as a crucial next step for AI, but a new Lancet paper from a team of AI heavyweights argues that explainability might do more harm than good in the short-term, and AI stakeholders would be better off increasing their focus on validation.

The Old Theory – For as long as we’ve been covering AI, really smart and well-intentioned people have warned about the “black-box” nature of AI decision making and forecasted that explainable AI will lead to more trust, less bias, and greater adoption.

The New Theory – These black-box concerns and explainable AI forecasts might be logical, but they aren’t currently realistic, especially for patient-level decision support. Here’s why:

• Explainability methods describe how AI systems work, not how decisions are made
• AI explanations can be unreliable and/or superficial
• Most medical AI decisions are too complex to explain in an understandable way
• Humans over-trust computers, so explanations can hurt their ability to catch AI mistakes
• AI explainability methods (e.g heat maps) require human interpretation, risking confirmation bias
• Explainable AI adds more potential error sources (AI tool + AI explanation + human interpretation)
• Although we still can’t fully explain how acetaminophen works, we don’t question whether it works, because we’ve tested it extensively

The Explainability Alternative – Until suitable explainability methods emerge, the authors call for “rigorous internal and external validation of AI models” to make sure AI tools are consistently making the right recommendations. They also advised clinicians to remain cautious when referencing AI explanations and warned that policymakers should resist making explainability a requirement. 

Explability’s Short-Term Role – Explainability definitely still has a role in AI safety, as it’s “incredibly useful” for model troubleshooting and systems audits, which can improve model performance and identify failure modes or biases.

The Takeaway – It appears we might not be close enough to explainable AI to make it a part of short-term AI strategies, policies, or procedures. That might be hard to accept for the many people who view the need for AI explainability as undebatable, and it makes AI validation and testing more important than ever.

Creating Your AI Platform Strategy

Adopting a platform strategy can simplify the deployment and management of imaging applications and AI algorithms, but there’s a lot to consider. In this eBook, Blackford Analysis and its clients detail how AI platforms can benefit clinical and IT teams, and share guidelines to consider when selecting a platform.

Growing with Desert Radiology

Trying to grow your imaging practice? See the strategies Desert Radiology used to manage its growth in this webinar from Aunt Minnie and United Imaging.

The Wire

• Dark-Field X-Ray’s COPD Potential: New research out of Germany shows that dark-field X-ray technology could become an effective way to diagnose COPD “and potentially other lung disorders,” while exposing patients to a fraction of the radiation produced by CT scans. Among 77 patients with COPD, dark-field X-ray outperformed CT for lung diffusion capacity assessments, matched CT for emphysema assessments, and improved emphysema characterization compared to standard clinical tests.

• Aidoc Adds ScreenPoint: Aidoc continued its AI platform expansion, adding ScreenPoint Medical’s Transpara mammography AI tools (2D & 3D). ScreenPoint fills a key gap on the Aidoc platform, joining the vendor’s homegrown triage/detection products (ICH, LVO, PE, IFG, c-spine & rib fractures) and partner products from Riverain (lung nodule detection), Imbio (PE assessment), Icometrix (stoke assessment), and Subtle Medical (image enhancement). The alliance also gives ScreenPoint access to Aidoc’s large client base, while adding to its solid list of partners (Fujifilm, Siemens, Volpara, Sectra, Agfa).

• The Case for LCS Telehealth: A new American College of Surgeons study suggests that virtual lung cancer screening consultations should remain a viable option after the pandemic, and could be valuable for other cancer screenings. Temple University Hospital previously held CT LCS exams and consultations during the same on-site visit, but made the consultations virtual during the pandemic. Analysis of the two periods (n = 637 & 440) revealed that the on-site and virtual visits had similar diagnostic results and follow-up recommendations, and didn’t adversely affect African American patients.

• GE’s ASTRO AI Alliances: GE Healthcare’s ASTRO 2021 announcement revealed a series of new AI vendor partnerships focused on radiation therapy planning and treatment guidance. GE unveiled a new collaboration with Vysioneer (CT/MR brain tumor auto-contouring), an MRI integration with Spectronic Medical (converts MRIs to synthetic CT images for treatment planning), and an AW Workstation / AW Server integration with Mirada Medical (enhances cancer visualization and diagnostics).

• AyrFlo Early Warning: A husband-and-wife duo from the University of Wisconsin developed an ultrasound-based device that can identify when sedated patients are experiencing respiratory distress up to four minutes before conventional oxygen monitors alert clinicians. The AyrFlow device (in FDA regulatory process) uses a modified ultrasound sensor that attaches to patients’ necks and displays their breathing status on an operating room computer screen, although future versions could expand beyond surgery (asthma, sleep apnea, etc.).

• ILD AI Evidence: A new AJR study highlighted how AI could help radiologists overcome traditional challenges diagnosing interstitial lung disease (ILD). The researchers had a commercial deep learning algorithm and six readers detect reticular opacity in chest X-rays (n = 197 patients with ILD, 197 control patients), finding that the AI tool outperformed the readers (with and without AI support) for sensitivity (98% vs. 93.8% vs. 73.3%), specificity (99% vs. 97.3% vs. 92.3%), and accuracy (98.5% vs. 95.6% vs. 84.8%). The readers’ interobserver agreement also significantly improved when they had AI support (κ=0.870 with AI vs. 0.517 without AI).

• Hospital Practice Acquisition: Montana’s Benefis Health System (Great Falls-based, 3,400 employees, 7 locations, 14 counties) acquired Helena Imaging (two imaging centers, two radiologists), allowing Benefis to provide imaging and radiology services in Helena after opening a pair of clinics in the city. We don’t usually see hospitals acquire radiology practices, but this seems like a unique case, rather than the start of a trend.

• Bayer’s 2021 G4As: Bayer’s G4A Digital Health Partnership program added five new “Advance Track” startups, including unique AI-enabled telerad firm Nines and medical image sharing company Zed Technologies (the other startups were non-imaging). Nines and Zed follow over G4A 50 Advance Track alumni that raised a combined $998m, although Blackford Analysis is the only previous imaging company.

• Breast Density Awareness: A new Bayer survey revealed a need to improve breast density education. The survey (n = 500 US women w/ dense breasts, >35yrs old) found that 30% of the women did “not feel informed” about how breast density impacts their cancer risks or screening needs. When they were told about their breast density, 57% did not receive any informational resources, leaving 24% of them “worried” and 26% “unsure of what to do next.”

• Joe Knows About SimonMed: Arizona television news consumer watchdog segment, “Let Joe Know,” detailed allegations of questionable billing practices at SimonMed (60 imaging centers in Arizona). A former employee and several patients alleged that SimonMed would “force” patients to pay more for imaging services than required by their co-pay, and then make them wait months for refunds. A SimonMed response detailed how it ensures up-front accuracy and transparency, and provides refunds when necessary.

• No Image Reconstruction: Hamamatsu Photonics and UC Davis scientists developed the first advanced medical imaging technique that doesn’t require image reconstruction, potentially allowing new scanners that are smaller, more efficient, lower radiation, and equally or more accurate than current CT and PET systems. The new technique was made possible by its use of a pair of detectors (vs. a ring) and its unique approaches to light detection and signal processing.

Salem Regional’s Case for Bayer’s MEDRAD Stellant FLEX

Learn how Salem Regional Medical Center improved its radiology workflows and cut service and syringe expenses after adopting Bayer’s MEDRAD Stellant FLEX system.

The Resource Wire

• See how Arterys’ combination of AI, 4D Flow, and cloud is making Cardiac MR more efficient and effective in this detailed case study.

• Take the AiCE challenge and see why half the radiologists in a recent study “had difficulty differentiating” images from Canon Medical Systems’ Vantage Orian 1.5T MR using its AiCE reconstruction technology compared to standard 3T MRI images.

• This Riverain Technologies case study details how the University of Colorado Hospital enhanced its chest X-ray workflow with ClearRead Bone Suppress.

• See how the Ochsner Lafayette General health system improved its radiology report quality and efficiency when it migrated to PowerScribe One.

• It’s clear that structured reporting is a must for CVIS platforms, but they aren’t all created equal. This Fujifilm Healthcare article reveals what physicians and sonographers view as the “non-negotiable” CVIS structured reporting features.

• See how Novarad’s CryptoChart solution allowed Central Ohio Primary Care (COPC, 70 practices, 400 physicians) to make the transition to digital imaging sharing in this Healthcare IT News case study.

• There might be more “advanced” imaging modalities, but X-ray keeps growing and will be a mainstay modality for the foreseeable future. See the factors driving X-ray’s growth in this GE Healthcare report.

• See why AI-powered cardiovascular screening’s detection and prevention benefits make it the new frontier of population health.

• How can artificial intelligence (AI) help improve clinical and operational performance in MRI? Listen to the latest Siemens Healthineers podcast to learn more.