“No one knows exactly how the development of artificial intelligence will proceed. But it does have the potential to radically change society and how we work.”
A very factual statement from Dutch finance minister, Wopke Hoekstra, right before he upset the radiology industry with an exaggerated take on how AI has impacted the specialty.
Know anyone who made outstanding contributions to radiology practice and outcomes this year? Nominate them for an Imaging Wire Award.
Imaging Wire Sponsors
- 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
Humans Preferred, AI OK
The majority of women want a radiologist involved in interpreting their screening mammograms, but a decent share are OK with AI playing a role, and many aren’t quite sure who they would fault if an AI-based diagnosis went wrong. That’s from a new JACR study out of Holland (n = 922 women, 16 to 75yrs) and here are the details:
- Human Involvement – The majority of the women believed that a radiologist should review all exams, even if the scans are also interpreted by an AI system (77.8% agree).
- AI First Reader – Most respondents were not quite ready for a process that uses AI for initial reads and only relies on radiologists to review flagged studies (31.5% agree, 26.9% neither agree nor disagree, 41.7% disagree).
- AI Second Reader – The women were more open to using AI as a second reader (after an initial radiologist interpretation), but sentiments were still quite mixed (45.9% agree, 37.1% neither agree nor disagree, 17% disagree).
- AI Responsibility – Responses were also mixed about assigning responsibility for AI-involved misdiagnoses, as “neither agree nor disagree” was the most common response (44.6% for developers, 39.2% for radiologists). However, more of the women would hold radiologists responsible for misdiagnoses (38.7% agree) than the AI developers (34.3% agree).
- The Takeaway – It makes sense that the general public isn’t quite sure how to feel about AI’s role in their cancer detection considering that many in AI and radiology are still debating these same topics. What’s more surprising is that over 22% of these women said they would be OK with an AI system independently performing their mammogram interpretations and only 31.5% need radiologists to review flagged scans.
- JH’s Implantable Ultrasound: Johns Hopkins scientists scored a $13.48m DARPA grant to develop an implantable ultrasound device that “could revolutionize care” for patients with spinal cord injuries. The device will use ultrasound technology to image the site of spinal cord injury and to stimulate nearby blood vessels and tissue in order to optimize oxygen/nutrient delivery. This project is more than just academic, as they plan to make their implantable ultrasound clinically available within the next 5 years.
- Lost in the Supermarket: Dutch finance minister, Wopke Hoekstra, came under fire from the country’s radiology societies for suggesting that AI has already made radiologists redundant and for comparing them to supermarket cashiers. To be fair, Hoekstra’s statement has been somewhat misrepresented. During a larger conversation about the country’s economy, he warned that AI might have an adverse long-term effect on Holland’s job market, using cashiers and radiologists as separate examples (not comparing them). Still, he did say that radiologists have become “significantly redundant because a machine that can read pictures better than a person who studied for them for ten years,” which isn’t currently true but certainly got the attention of the Dutch imaging lobbies.
- Radiology’s Heavy Load: Radiology is the most mentally demanding physician specialty and has among the highest Physician Task Loads (PTL = combined mental, physical, and temporal demands, and perception of effort). This is from a new survey of 5,276 physicians that found radiologists have a 272.2 PTL score (out of 400), which is above the survey’s 260.9 mean PTL, but well below emergency physicians’ 295.3 score. The study also suggested that high PTL scores correlate with high burnout rates.
- GE Vivid Ultra Edition: GE Healthcare announced the FDA approval of its Vivid Ultra Edition cardiovascular ultrasound system, which leverages new AI-based features that shorten exams (up to 80% fewer clicks) and improve measurement consistency (up to 99% accuracy). The Vivid Ultra Edition is highlighted by its new ‘AI Auto Measure – Spectrum Recognition’ tool (detects appropriate spectral Doppler image measurements), ‘AI Auto Measure – 2D’ tool (detects parts of images used to measure the left ventricle), and its ‘AI-based View Recognition’ feature (IDs which standard 2D scan plane is acquired and stores it in the image file).
- Zebra’s AI Optimism: An editorial from Zebra Medical Vision CEO, Ohad Arazi, shared some optimism for radiologists who might be concerned about how AI will affect their profession. Here are some of his key points: 1) The number of radiologists continues to grow and regional shortages persist; 2) Radiologists do far more than analyze images for a single condition; 3) Unless a “complete end-to-end system” is developed, AI won’t replace radiologists; 4) AI will help us battle COVID by aiding management and processing backlogs; 5) AI will allow radiologists to perform more complex diagnoses; 6) AI will enable new image-based analytics breakthroughs (e.g. genetic sequencing).
- Hybrid Improvements: A University of Chicago study detailed how the conversion of a hospital fluoroscopy suite to a hybrid CT/angiography system improved utilization and efficiency. Comparing the 24 months before the room’s hybrid conversion to the 12 months after the transition, the hospital’s monthly shared CT and Angio-CT volumes increased by 46.7% and 12%, while the new hybrid room achieved a 287% efficiency improvement compared to the hospital’s other IR suites.
- Fast & Accurate COVID CT AI: A team of Chinese researchers developed and evaluated an AI system that can accurately distinguish COVID-19 from other lung conditions in CT scans. Developed using 4,260 CT images (patients w/ COVID-19, the flu, and healthy lungs), the algorithm detected COVID-19 with a 97.81% AUC in a 3,199-image test set, and reached AUCs of 92.99% and 93.25% against a pair of public datasets. That’s high enough to outperform five radiologists, while completing the reads in half the time.
- Simplified Reports: An Atlanta-based team developed a new system to help patients understand their radiology reports. The Information Reporting and Data Systems (Info-RADS) adds a pair of messages that use “very general terms” to 1) communicate if there are concerning findings and 2) Note if there are actionable but not critical findings that require follow-up. A survey revealed that patients found the two messages to be valuable (87% & 73%), the first message reduced their anxiety (68%), and the second message increased some patients’ sense of urgency (42%).
- Lunit’s CXR Evidence: A Lunit and Mass General Hospital study demonstrated how AI can improve CXR-based lung cancer detection when used as a second reader. The research team used Lunit INSIGHT to analyze 5,485 CXR images from NLST participants, identifying pulmonary nodules with 94% sensitivity and 83% specificity (higher sensitivity than NLST radiologists). The researchers suggested that the algorithm might improve CXR’s role in lung cancer diagnosis.
- Fluoro Glasses: Japanese researchers demonstrated how commercially available smart glasses (in this case, picoLinker glasses) could be used to support minimally invasive spinal surgery by displaying fluoroscopic images on a small screen in front of the surgeon’s glasses. Their study of 20 consecutive lumbar fusion surgeries (10 w/ glasses) confirmed that the smart glasses reduced surgeons’ head turns to view the surgical monitor (0.1 avg. vs. 82.4), while cutting surgery times (100.2 minutes vs. 105.5) and radiation exposure times (38.6 minutes vs. 41.8).
- Google Transparency: At the very start of 2020, Google published an impressive mammography AI study that showed its model was as accurate as radiologists and caught cancers that the rads missed. It took less than a week for the study’s first critical response to surface, and 10 months later criticism is still coming out. The latest rebuttal came from a large international team who called out Google’s lack of transparency (data, code, model design), which made it difficult for researchers to confirm or add to the study, while encouraging researchers and scientific journals to publish studies with greater transparency and reproducibility.
- Eliminating Blooming Artifacts: A team of scientists from Rensselaer, GE Research, Cleerly, and Weill Cornell will use a $3.7m NIH grant to develop an AI-based technique that eliminates blooming artifacts in cardiac CT scans, without requiring redesigned CT hardware.
- Qynapse Acquires True Positive: French neuro imaging AI firm, Qynapse, acquired Canadian brain analytics software developer, True Positive Medical Devices (TPMD), representing a rare imaging AI acquisition (so far, anyway). The acquisition combines TPMD’s technologies, patents, and expertise with Qynapse’s know-how and product line, creating what they suggest is the “the most advanced AI platform for brain diseases.”
- See-Mode’s 510(k): Stroke prediction and prevention AI startup, See-Mode Technologies, announced the FDA 510(k) approval of its Augmented Vascular Analysis (AVA) software. AVA uses deep learning, text recognition, and signal processing to streamline the analysis and reporting of vascular ultrasound scans (<1 minute), reducing the need for manual drawings and cutting overall reporting time. See-Mode’s first FDA approval comes just after a $7m funding round, which it will use to expand to the US.
The Resource Wire
– This is sponsored content.
- These two independent Phase III trials provided new evidence supporting the accuracy and effectiveness of using Bayer’s Gadavist contrast agent for cardiac MRI.
- Learn how Scott County Hospital’s new CT helped them increase their imaging bandwidth in this upcoming Hitachi webinar.
- Radiology will see a significant cut in Medicare reimbursement in 2021 if the MPFS Proposed Rule is applied without a change to the budget neutrality requirement in the law. Get all the details in this Healthcare Administrative Partners blog post.
- Vessel suppression from Riverain Technologies’ ClearRead CT software was found to significantly improve nodule detection, interreader agreement, and reading time with oncologic chest CT scans.
- This GE Healthcare case study details how Shields Health Care Group reduced its MRI scan times by 30% after adopting GE’s AIR Recon DL image reconstruction technology.
- In its latest Q&A, Nuance sat down with Infervision North America Director, Matt Deng, Ph.D., to discuss the company’s journey and the impact of its InferRead Lung CT solution.
- Siemens Healthineers’ c.cam cardiac camera allows you to perform high-quality scans in a small space without costly renovation fees. Discover how quickly you could be up and running with a budget-friendly cardiac camera.