AI Effectiveness | Nanox’s Next Step | Validation in the Wild

“People that focus on the technology aspect of AI will get tripped up. The questions that people need to ask are: What problem are they solving? What workflow are they optimizing? What condition are they trying to create a positive outcome for?”

MEDNAX Radiology and vRad CIO, Imad Nijim, on the types of questions MEDNAX and Qure.ai focused on as they set out to validate Qure.ai’s qER algorithm.


Imaging Wire Sponsors

  • Focused Ultrasound Foundation – Accelerating the development and adoption of focused ultrasound.
  • GE Healthcare – Providing point of care ultrasound systems, from pocket-sized to portable consoles, designed to support your clinical needs and grow along with your practice.
  • Healthcare Administrative Partners – Empowering radiology groups through expert revenue cycle management, clinical analytics, practice support, and specialized coding.
  • Nuance – AI and cloud-powered technology solutions to help radiologists stay focused, move quickly, and work smarter.
  • Qure.ai – Making healthcare more accessible by applying deep learning to radiology imaging .
  • Riverain Technologies – Offering artificial intelligence tools dedicated to the early, efficient detection of lung disease.

The Imaging Wire

AI Effectiveness

A Lunit-funded study in South Korea added new evidence that AI can improve early-stage breast cancer detection, especially for women with dense breasts, while improving radiologist performance.

The Study – The researchers developed and validated an algorithm using 170,230 mammography exams from five institutions in South Korea, the USA, and the UK (36,468 malignant, 59,544 benign, 74,218 normal). Then in a multi-center, observer-blinded study they had 14 radiologists assess 320 mammograms from two institutions (160 malignant, 64 benign, 96 normal) with and without AI support.

The Results – The algorithm achieved an 0.940 AUC reading the 320 test images, significantly higher than the radiologists’ performance without AI assistance (0.810), while increasing the radiologists’ performance to 0.881 when they were aided by AI. The study found that the algorithm outperformed the radiologists in detecting cancers by mass (53 vs. 46 out of 59) and distortion or asymmetry (18 vs. 10 out of 20), as well as detecting T1 cancers (73 vs. 59 out of 80) and node-negative cancers (104 vs. 88 out of 119).

Implications – We get it that most are tired of the AI beats radiologists storyline (and this study tried hard to emphasize the “AI helps rads” angle), but this study did a lot of things right and appears to add even more evidence that AI really can help.



Nanox’s Next Step

Nanox took a big step towards achieving its lofty market expansion goals, announcing a strategic collaboration with teleradiology company USARAD that it says will lead to the deployment of 3,000 Nanox Systems across the U.S.

The Partnership – Nanox and USARAD will work together to market the Nanox.ARC digital X-ray system and Nanox.CLOUD software to U.S. imaging centers and other potential customers (urgent care, physician practices, etc.) under Nanox’s pay-per-scan business model. Nanox will rely on USARAD’s market knowledge/relationships to guide its big U.S. rollout, while utilizing USARAD’s panel of 300 radiologists to interpret the images once scans start rolling through the Nanox.CLOUD system.

Upsides – All of this would still require regulatory approval and market adoption (plus, it’s not clear what Nanox’s 3k system claim is based on), but there are clear upsides if it works out as planned. Nanox gets a U.S. partner with decent clout (500 medical imaging facility clients) and the panel of radiologists needed to deliver the imaging workflow services that are supposed to come with its systems. USARAD would get a flow of images into its teleradiology network and likely a cut from each Nanox system it helps deploy.



Imaging Wire Q&A: Qure.ai and MEDNAX Validate AI in the Wild

As we enter the “prove it and improve it” phase of the imaging AI adoption curve, Qure.ai and MEDNAX recently completed an algorithm validation partnership that helped Qure.ai do exactly that: validate its qER algorithm at high volumes and across a wide range of variables to help prove and improve its performance.

In the latest Imaging Wire Q&A, we sat down with the teams at Qure.ai, MEDNAX Radiology, vRad discuss their efforts to validate Qure.ai’s qER solution “in the wild.” Here are some of the big takeaways:

Origins – MEDNAX and Qure.ai have been working together for years, leveraging MEDNAX’s massive and diverse dataset to help make sure its algorithms would generalize well, ramping up their latest and largest validation initiative in 2019.

Validation in the Wild – MEDNAX Radiology Solutions applied Qure.ai’s AI models and plugged them into MEDNAX’s own inference engine, sending a flow of 300,000 image studies through qER in nearly the same way active algorithms would be used. MEDNAX captured the results along with the results from its own radiologists to track performance, finding that Qure.ai’s model held a consistent sensitivity and specificity throughout the process.

Takeaways – Although some might say qER’s 0.96 AUC is the main takeaway from this validation, the long-term benefits came from giving Qure.ai a new way to understand how qER would work with this volume and variety of studies and fine-tune the product’s clinical and operational performance.



Share The Imaging Wire

Thank you all for supporting The Imaging Wire and Visage Imaging‘s referral program benefiting the New South Wales Rural Fire Service and WIRES Wildlife.

Please take a moment to consider your colleagues who might find this newsletter valuable and let them know about it. Once again, we’re donating $2 to these Australian Bushfires charities for each new subscriber that you refer.

If you’re a subscriber, please consider using your personal referral link to introduce your industry friends to The Imaging Wire (You can find your personal link in your latest newsletter).

If you haven’t received a referral from one of your colleagues yet, just use this link to sign up.


The Wire

  • The FDA authorized Caption Health’s Caption Guidance software, making it the first AI-guided cardiac ultrasound software available in the U.S. The new solution is used with compatible ultrasound systems (currently Teratech US systems), helping less-trained users capture accurate two-dimensional transthoracic echocardiography (2D-TTE) images.
  • An AJR study detailed how radiology report templates can improve report comprehension / confidence when read by referring physicians and patients. A quality improvement initiative at Brigham and Women’s Hospital designed to increase the use of the term “normal” and other preferred descriptions and decrease vague terms (e.g. “within normal limits”) was relatively successful. The share of reports from attending radiologists using the term “normal” increased from 49% to 59%, although residents did not show the same improvements.
  • Change Healthcare filed to separate from McKesson in a proposed share swap that could be worth around $2.8 billion. Although the separation comes just three years after the companies merged, it shouldn’t come as a surprise given that McKesson revealed plans to separate the companies at the time of Change’s mid-2019 IPO.
  • KHN detailed how the “moral injury” from healthcare inequities and bureaucracies (e.g. unnecessary treatments, high patient costs, preauthorizations, prioritizing money over care) is actually having a greater toll on clinicians’ emotional wellbeing than the factors that have traditionally been associated with burnout (e.g. long hours, inefficient EHRs, etc.).
  • A report from IMV Medical Information Division detailed on AuntMinnie.com found that U.S. CT procedure volume reached a record 91.4 million scans in 2019 (+3% from 2018), driving demand for CT workflow efficiency improvements. The last two years of CT procedure growth actually reversed a downward trend since the previous peak in 2011 (85.3m procedures). Although respondents expect continued CT procedure growth in 2020, they do not believe it will drive a significant increase in demand for new CT scanners in the short term (2/3 said their current CTs are sufficient).
  • Infinx Healthcare launched its AR Optimization Solution (AROS), which uses machine learning to help radiology groups and imaging centers analyze and act upon information in their payer accounts receivables systems. Citing a pilot program that helped reduce write-offs by 60%, Infinix aims to improve the AR recovery process by automatically examining outstanding AR, then using machine learning to predict collectability and make recommendations on how to prioritize charge recovery.
  • USC researchers announced the development of a new high-speed and high-accuracy imaging technique, called SEER (spectrally encoded enhanced representations), that may be able to be applied to everything from detecting lung cancer, spotting counterfeit money, or screening for food safety. SEER reportedly works up to 67 times faster and at 2.7 times greater definition than current techniques (specifically vs. fluorescence hyperspectral imaging) by using mathematical computations to parse data faster and process cellular colors in more detail.
  • A UCSF team developed a deep learning model that was able to grade major hip osteoarthritis features on hip joint X-rays with relatively promising accuracy. The retrospective study analyzed 4,368 participants’ hip joint X-rays captured between 2004 and 2006 (15,364 images) for baseline measurements and compared them to 48-month follow ups, assessing femoral osteophytes (86.7% accuracy w/ internal set, 82.7% accuracy w/ external set), acetabular osteophytes (69.9%, 65.4%), joint-space narrowing (81.7%, 80.8%), subchondral sclerosis (95.8%, 88.5%), and subchondral cysts (97.6%, 91.3%).

The Resource Wire

  • This Healthcare Administrative Partners blog post details how independent radiology practices can build upon their relationships with hospitals, referring physicians, and even neighboring radiology practices to remain strong and independent in the face of ongoing consolidation.
  • 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. Learn more.
  • This GE Healthcare white paper details how its suite of point of care ultrasound AI tools simplify complex patient assessments, enable faster clinical decisions, and calculate precise results.
  • Nuance is preparing to debut new AI-powered systems that automate routine reporting and image analysis, aid in diagnosis, and help uncover incidental findings. Here are the details.

3 Responses to “AI Effectiveness | Nanox’s Next Step | Validation in the Wild”

Leave a Reply

Basic HTML is allowed. Your email address will not be published.

Subscribe to this comment feed via RSS

Get every issue of The Imaging Wire

Delivered right to your inbox. We will never sell or share your information.