“Doctors are human, and humans are biased.”
University of Toronto professor Marzyeh Ghassemi on the human origins of AI algorithms’ inherited bias.
If this feels like a good week to vote and you know someone who made outstanding contributions to radiology this year, nominate them for an Imaging Wire Award.
Imaging Wire Sponsors
- Arterys – Transforming medical imaging by reducing the tedium of interpretation and empowering physicians with our web-based AI platform.
- 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.
- 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.
The Imaging Wire
Explainable Mistakes
Many folks view explainable AI as a crucial next step for AI adoption, but new research out of the University of Toronto suggests that making healthcare AI more ‘explainable’ might actually lead to more medical mistakes. Here’s an explanation:
- The Research – The study by U of T professor Marzyeh Ghassemi found that even though medical experts generally view machine-based recommendations as less reliable than advice from a human, they’re still just as likely to follow AI recommendations.
- Trust & Transparency – Because of this, Ghassemi argues that explainable AI might cause physicians to over-trust and over-rely on their AI tools, leading to more mistakes. She’s not alone, as previous research found that people were less likely to catch errors made by a transparent AI model than a “black-box” model.
- The Transparency We Need – Ghassemi suggests that explainable AI will still be valuable, but for telling physicians when a model could be wrong or warning of potential bias, rather than just explaining how the model arrived at a conclusion.
- The Takeaway – Explainable AI (at least the type many envision) might not be the AI adoption breakthrough many forecast. Even if explainable AI does lead to a new wave of AI adoption, “we need to treat all advice, including AI-based advice, with appropriate suspicion because advice that seems transparent can cause overconfidence.”
Brain MRI Breakthrough
Northwestern University researchers developed a new MRI technique, dubbed T1RESS, that significantly improves the visibility of brain tumors in MRI scans and could lead to major diagnosis and treatment improvements. It could eventually expand to other cancers too.
- About T1RESS – T1RESS uses MRI signals’ radio waves and magnetic fields to manipulate the signals from various brain tissues in a way that doubles the contrast between tumors and normal brain tissue (vs. traditional MRI techniques).
- T1RESS Benefits – This two-fold increase in tumor visibility could allow physicians to spot tumors when they are smaller, still operable / treatable, and “before they have invaded other tissues or metastasized to other regions of the body.” T1RESS’ benefits might also expand beyond diagnosis, potentially advancing radiation therapy (depicting tumor size and shape, catching smaller tumors) and surgery or radiotherapy (improved visibility of tumor margins).
- Next Steps – There’s reason for optimism after this proof-of-concept study, but also more research to do, including confirming these initial findings with a larger multi-site trial. The team also plans to study T1RESS’ ability to detect breast and prostate cancers.
The Wire
- Covera Health’s AI Play: Radiology clinical analytics and certification company, Covera Health, announced the integration of ScreenPoint Medical ’s mammography AI tools into its quality analytics platform. The partnership represents an interesting (and logical) expansion to its practice quality improvement service that will lead to the addition of other third-party applications into the Covera Health ecosystem.
- Developing an Easy Ultrasound: Norwegian research organization SINTEF is developing a portable ultrasound system that any clinician could use, even if they aren’t ultrasound specialists. The still in development INtelligent Handheld Ultrasound Device would use a combination of AI tools, 3D imaging (ultrasound is typically 2D), and an anatomy “navigation system” to support its usability value proposition.
- Pandemic Imaging Disparities: A new JACR study of NY state patients revealed that the COVID-19 pandemic drove significant imaging utilization shifts across a range of socioeconomic groups. The study identified the greatest increases among patients who are older (60-79yrs), male, minorities, earn <$80k annually, and have Medicaid / uninsured status, suggesting that these increases were related to COVID infections. Meanwhile, imaging volumes significantly decreased among opposite groups (<18yrs, female, white, commercial coverage, >$80k income), due in part to delayed screenings and care.
- Nanox & Ambra’s Transfer Partnership: Nanox announced that it will use Ambra Health’s enterprise image exchange solution for medical image storage and transfer, making Ambra a core part of the Nanox.CLOUD platform and Nanox’s “medical screening as a service” (MSaaS) strategy. The partnership helps answer some of Nanox’s critics, who (among other things) have pointed to a mismatch between Nanox’s futuristic strategy and its limited infrastructure. The partnership also lends credibility to Nanox’s supporters who often cite its list of established partners as evidence of the startup’s legitimacy.
- COVID Cough AI: MIT researchers developed an AI model that can reportedly detect asymptomatic COVID-19 infections using cellphone and laptop-recorded coughs with 98.5% accuracy (trained on 200k coughs). This story doesn’t directly involve imaging, but it’s pretty interesting, and it’s the second smartphone+AI triage tool announced in as many weeks (the last one used smartphone videos to detect strokes).
- USPSTF Calls for CTCs at 45: The US Preventive Services Task Force (USPSTF) officially recommended lowering CTC screening’s starting age from 50 to 45 and increasing efforts to screen minority groups, aligning with recommendations from both the ACR and American Cancer Society.
- Microsoft’s Medical Imaging Server: Microsoft’s big Healthcare Cloud launch brought the tech giant deeper into medical imaging with its new Medical Imaging Server for DICOM open source solution. The Medical Imaging Server for DICOM combines with Microsoft’s Azure API for FHIR, allowing healthcare organization to link clinical health data in FHIR with structured metadata from imaging files and perform tasks “that would be difficult and expensive to execute” with on-premises systems (e.g. creating research cohorts, access longitudinal views of a patient during diagnosis, offsite image storage).
- Cardiac MRI GBCAs’ Low Reaction Risk: A new RSNA study found that cardiac MRI gadolinium-based contrast agents have a “very low risk” of adverse events. The researchers analyzed data from 154,779 patients who underwent cardiac MRI (145,855 w/ GBCA), finding only 556 acute adverse reactions (47 severe). That’s low considering that 2.59% of the 8,924 patients who didn’t receive GBCAs also had adverse events.
- DiA’s Cardiac US FDA 510k: DiA Imaging Analysis announced the FDA approval of its LVivo Seamless AI-based cardiac ultrasound application (its 7th FDA 501k), which streamlines echocardiography workflows by automatically selecting the optimal views from each cardiac ultrasound exam.
- Final Transparency: Just in time for the election, CMS announced that its healthcare cost transparency ruling is now “final,” requiring payers to significantly expand how they disclose pricing and cost-sharing information. Payers will have to publish standardized data files for research use by January 2022, launch online comparison tools by January 2023, and disclose cost-sharing information on all the services and goods they cover by the start of 2023. However, some doubt how final this ruling will be, even if HHS secretary, Alex Azar, says it will “be impossible to walk backwards on this.”
- Eon Adds MR & XR for IPN: Eon announced that its Essential Patient Management platform can now identify incidental pulmonary nodules (IPNs) on MRI and X-ray radiology reports (previously only CT), allowing facilities to catch 25% more IPNs. The platform uses computational linguistics to identify incidental pulmonary nodules in MRI and X-ray reports with 97% accuracy (98.95% w/ CT).
- 18F PET/MRI for Prostate Staging: A new study in AJR found that 18F PET/MRI could be effective for the initial staging of high-risk prostate cancer and evaluating response to androgen deprivation therapy (ADT). The researchers performed 18F PET/MRI and MRI pretreatment staging exams on 14 men with newly diagnosed high-risk prostate cancer who previously underwent conventional staging exams with negative or ambiguous results. The researchers identified biopsy-proven lesions in each of the 14 men’s MRI and 18F PET/MRI exams, while 18F PET/MRI detected more suspected nodal metastases (7 vs. 3).
- Rad’s Top-10 Pay: Doximity’s new Physician Compensation Report (survey of 44k physicians in 2019-2020) found that radiologists’ $485k annual average compensation was the 10th highest, well above the average of all physicians ($383k) but well below neurosurgeons ($746k). The survey suggests radiologist comp increased by about 13% since the 2019 survey ($429k), although it’s unclear how many responses were pre-COVID, and unlikely that many rads actually scored a 13% pay bump this year.
The Resource Wire
– This is sponsored content.
- Learn how one critical access hospital in a California ski town used Nuance PowerShare to accelerate care, drive change, and #ditchthedisk in this upcoming webinar.
- Acknowledging the pressures that hospitals and imaging centers are under during the COVID-19 emergency, this Hitachi blog details the equipment financing programs have become most in demand and unveils its new post-COVID programs launched in partnership with Key Equipment Finance.
- 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.
- Check out NVIDIA’s blog post about AI Startups, like Arterys, and their contributions to the fight against COVID. The post includes the epic story of Arterys’ Marketplace launch, and details the various AI models developed to detect COVID in CT images and X-rays.
- They say that in times of crisis, you get to know who your real friends and partners are. This Q&A session details how Healthcare Administrative Partners stepped up to guide their client Triad Radiology Associates through the challenges presented by the COVID-19 pandemic.
- This GE Healthcare article details the company’s long history of image quality advancements and how its new technologies are bringing new image quality and reconstruction breakthroughs.
- Check out how Siemens Healthineers’ Tin Filtration is like selecting better quality sunglasses.
- These two independent Phase III trials provided new evidence supporting the accuracy and effectiveness of using Bayer’s Gadavist contrast agent for cardiac MRI.
