AI’s Validation Problem | Quantifying CA AB-72 | Catching Up With ASC

“What if we could predict your future behavior by similarities that your fMRI networks share with those of psychopaths who had been analyzed and whose data now resides in a database?”

Yale radiology and biomedical imaging professor, Evan D. Morris, Ph.D., in an editorial calling for new guidelines protecting data captured in brain scans. This may sound like science fiction, but so did AI a few decades ago.

Imaging Wire Sponsors

  • Carestream – Focused on delivering innovation that is life changing – for patients, customers, employees, communities and other stakeholders
  • Focused Ultrasound Foundation – Accelerating the development and adoption of focused ultrasound
  • Medmo – Helping underinsured Americans save on medical scans by connecting them to imaging providers with unfilled schedule time
  • Nuance – AI and cloud-powered technology solutions to help radiologists stay focused, move quickly, and work smarter
  • Pocus Systems – A new Point of Care Ultrasound startup, combining a team of POCUS veterans with next-generation genuine AI technology to disrupt the industry
  • Qure.ai – Making healthcare more accessible by applying deep learning to radiology imaging

The Imaging Wire

AI’s Validation Problem

Lancet Digital Health published the first systematic review of medical imaging deep learning studies, finding that imaging AI appears to have matched the performance of clinicians, while revealing how much further imaging AI research must progress in order for this performance to be validated.

The study – A review of 82 studies (69 with enough data for contingency tables) found that deep learning models had a pooled sensitivity of 87% (vs. 86.4% for clinicians) and a pooled specificity of 92.5% (vs. 90·5% for clinicians).

Validation – Although this review is a positive sign for deep learning’s potential, it also confirmed persistent methodology challenges, as only 25 of the 82 studies performed out-of-sample external validation and just 14 compared deep learning models and health-care professionals in the same sample.

Advancing DL Research – Because of this, the team called for new reporting standards addressing deep learning’s specific challenges, and suggesting that these standards would allow the AI and radiology communities to have greater confidence in study results. Until then, AI’s potential to achieve the quality of human physicians remains unproven.

Quantifying CA AB-72

California’s 2017 surprise billing law found its way back into the spotlight last week. A new Brookings and USC report detailed how AB-72 influenced California’s out-of-network billings (OONs) and provided new evidence on what could happen if the federal government’s pending surprise billing legislation is signed into law.

California Assembly Bill 72 (AB-72) – The two year-old law limited how much providers could charge patients for out-of-network services, requiring payers to reimburse out-of-network providers at either their local average contract rate or at 125% of Medicare’s fee-for-service rate (whichever is higher). AB-72 can be viewed as a model for the federal bill calling for OON billings to be based on a median benchmark (not arbitration).

The Research – The study found that California’s 2017 surprise billing law drove a 17% decline in the share of specialist services delivered OON at inpatient hospitals and ambulatory surgical centers. Meanwhile emergency medicine services, which weren’t targeted by the California law, did not experience a statistical change in surprise billings.

Specialty Impact – The law drove the greatest shift in neonatal-perinatal OON billings (OON declined by 31%, in-network increased from 88.4% to 92.1%). The study didn’t specify OON billing share declines for radiology, although the specialty saw the smallest increase in in-network billings after the law went into effect (up 2.2 points to 89.3%).

Evidence and Influence – This study joins a growing field of evidence either warning that a California-style median benchmark law would hurt specialists and drive consolidation (supported by specialists) or improve patient access to in-network specialists (supported by insurers). This Brooking/USC study stopped short of calling either side correct (they actually called for more research), but the report could be seen as evidence that the California law was effective at reducing OON billings.

Catching Up With ASC

A new report from Bain & Company forecast significant procedure growth at U.S. ambulatory surgery centers (ASCs) but questioned whether medtech firms can catch up with this shift. Here are the details:

ASC Growth – Bain forecast that procedure volume at ASCs will grow by between 6% and 7% through 2021, up from 4-5% growth over the last three years. This is pretty notable given that U.S. ASCs already perform over half of all outpatient surgeries (vs. 32% in 2005). ASCs’ procedure growth will be led by orthopedic (16%), spine (20%), and cardio (23%) procedures through the mid-2020s, while “other” ASC procedures will grow at a more modest 5% rate.

ASCs’ Strength & Weakness – The key to ASCs’ growth is their ability to charge 35% to 50% less than hospitals, however they also have lower reimbursement rates and are more cost sensitive. ASCs also rely on physicians as their primary medical device decision maker/influencer far more than traditional hospitals (70% vs. 45%) and they value technical support and info on payer reimbursement far more than their hospital counterparts.

Medtech’s ASC Challenge – ASCs are more expensive to sell to (they’re smaller, lower volume, and more geographically dispersed) but also much more cost-sensitive, highlighted by Bain’s finding that 60% of ASCs would switch vendors at a 15% lower cost. Meanwhile many medtech firms have been slow to react to ASCs’ growth due to their historical focus on acute care hospitals.

ASC Solutions – Although its good news for slower-moving medtech firms that hospitals are increasingly acquiring ASCs (25% now have hospitals as shareholders), more ambitious providers are developing lower-cost service models (e.g. remote training), simplifying their products (e.g. reducing inventory requirements), and providing more customized services (e.g. IT service support) to better support the ASC market.

  • New research in the American Journal of Roentgenology found that breast ultrasound was just as effective following DM and DBT. The Brown-led study reviewed 3,183 breast US screenings (45.1% after DM, 52.4% after DBT), finding that breast ultrasound’s additional cancer detection rate was 5 / 1,434 (3.5 / 1k) after DM and 5 / 1,668 (3.0 / 1k) after DBT.
  • GE Healthcare expanded its Edison AI platform to China and announced strategic partnerships with five local developers to create applications for the platform (Shukun Technology, Yizhun Medical AI, YITU Technology, 12Sigma Technologies, Biomind). GE will make the Edison platform available to other Chinese developers, allowing them to produce modules using the Edison portal and distribute them globally.
  • A wired.com editorial by Yale professor Evan D. Morris called for new guidelines protecting data captured in brain scans, suggesting that this data can “reveal as much about you as your DNA” including your diseases (e.g. Parkinson’s or depression), your past (e.g. drug use or trauma), and potentially future (e.g. Alzheimer’s and treatment response). Although the ethical and legal implications of brain imaging are rarely discussed, the editorial suggests that future brain imaging applications may make brain imaging policies as important as DNA-related policies are today.
  • Nearly a year after debuting in Asia, Samsung NeuroLogica announced the FDA clearance of the company’s new Hera I10 ultrasound system, featuring a unique integrated ultrasound examination chair. The announcement clearly emphasized the Hera I10’s ergonomic advantages, including how the motorized chair helps physicians place patients in the optimal position and help patients get on/off the exam table, how its floating console improves sonographer comfort and reduces strain, and how its cable support arm minimizes transducer weight. The new system also has the scanning technology that the Hera platform is known for (Crystal Architecture, reduced shadow tech, blood flow visualization).
  • Brainlab unveiled its new Loop-X mobile intraoperative imaging robot, positioning the medPhoton-developed robot at the center of its spinal Digital Surgery portfolio. Brainlab highlighted the Loop-X’s ability to automate imaging workflow steps and robotically move with the procedure and on command, as well as its ability to image larger and smaller structures.
  • A University of Pennsylvania-led research team unveiled a 2mm nanophotonic microwave imager chip that is far smaller and much more effective than current systems and could allow development of hand-held microwave imagers capable of high-resolution imaging. The new near-field imager uses optical technology (vs. electronic) to process the microwave signal, allowing the production of a chip-based imager similar to smartphones’ optical camera chips.
  • Loyal Source acquired medical imaging services staffing firm, SonoTemps, which specializes in staffing sonographers, mammographers, vascular technologists, and MRI/CT/XR techs. The merger of the two Florida-based medical staffing firms expands Loyal Source’s imaging capabilities, in addition to staffing for travel nurses, physicians, advanced practice, and allied and clinical specialty roles across the U.S.
  • Ryerson, Caltech, and Sunnybrook researchers found that ultrasound can spot whether certain genes are switched on in animals, suggesting that US could eventually be used in new gene regulation applications like measuring tumor growth or nerve cell function. Although ultrasound can’t image at the cellular level, the researchers found that by applying aquatic bacteria that produces microscopic air bubbles to reflect sound waves, it could make cells’ genetic expression visible with ultrasound.

The Resource Wire

  • This Healthcare IT News article details how the University of Rochester Medical Center’s Backstop program leverages Nuance’s mPower Clinical Analytics and PowerScribe Follow-up Manager solutions to track and complete more follow-up imaging recommendations.
  • This Carestream case study compares images of foot trauma captured using the OnSight 3D Extremity System to images captured on 2D X-rays.
  • By partnering with Medmo, imaging centers can keep their schedules full and their equipment busy. Here’s where to learn more and get started.

You might also like

You might also like..

Select All

You're signed up!

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

-- The Imaging Wire team

You're all set!