GE’s Embedded AI | Non-Physicians Not Interpreting | Modified Images

“If doctors can’t or won’t fix a problem that is almost universally acknowledged in our profession, should we act outraged or surprised when an outside agency tries to do it for us?”

A NYT Op-Ed from cardiologist, Sandeep Jauhar, criticizing U.S. physicians for complaining about the upcoming appropriate use criteria requirements even though they couldn’t or wouldn’t address the unnecessary imaging issues that led to it.


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



GE’s Embedded AI

GE Healthcare announced the FDA clearance of its Critical Care Suite, an “industry-first” set of device-embedded AI algorithms that flag pneumothorax cases in chest X-rays and alert the clinical team for prioritization.

  • The Solution – Built in collaboration with UCSF using GE Healthcare’s Edison platform, Critical Care Suite comes embedded on the GE Optima XR240amx mobile X-ray system. Critical Care Suite immediately flags critical cases scanned on the device and informs the radiologists (via PACS) and technologists (via on-device notification) of urgent cases. The solution also includes quality control features that detect acquisition errors for technologists to review and correct (protocol, field of view, rotation) before the images go to the radiologists.
  • Big Emphasis – GE has been promoting Critical Care Suite throughout the last year, including a showcase at RSNA 2018 and regular mentions from top executives who used it as an example of GE’s AI leadership. However, launch messaging from GE Healthcare CEO, Kieran Murphy, suggests that Critical Care Suite may be more significant for its role in GE’s AI roadmap, as he revealed that the rest of GE’s offerings will soon add similar capabilities and “Critical Care Suite is just the beginning.”



Non-Physicians Not Interpreting

Despite increased policy-level support for medical image interpretation by non-physicians (not with always radiologist support), new research published in the American Journal of Roentgenology finds that nurse practitioners and physician assistants still rarely interpret diagnostic imaging. When they do, its “overwhelmingly” for radiography and fluoroscopy services.

  • The Study – A review of Medicare data found that although diagnostic imaging services rendered by non-physicians increased by a whopping 14,711% between 1994 and 2015 (from 36 to 5,332 per 100k beneficiaries), non-physicians still only interpreted 1.27% of the total volume. The vast majority of these non-physician-billed services were for radiography and fluoroscopy (0.01% and 2.1% of all XR and fluoro Medicare services during the 1995-2015 period), which increased from 10,899 non-physician services in 1994 to 1,665,929 in 2015. The team also found considerable state-to-state variations (PA: 478 per 100k, SD: 7,987 per 100k), suggesting that these differences are due to unique state-level regulations.
  • The Significance – The authors don’t expect non-physicians to begin interpreting a substantial share of medical images in the near-term, but called for healthcare providers who employ radiology NPs and PAs to better define these roles and initiate training programs that align with regulations in their state.



Modified Images

A team of Swiss researchers found that machine learning networks can learn what makes a mammographic image look suspicious and modify the images (inject or remove the suspicious areas), representing either a way to support teaching and research applications or a tool for image cyberattacks.

  • The Study – The team trained their CycleGAN algorithm on 680 publicly available images (with and without lesions) and tested it against an internal dataset (n = 302 cancers, n = 590 controls). Three radiologists evaluated low and high resolution versions of the modified images (256 × 256 px and 512 × 408 px).
  • The Results – The radiologists’ overall cancer detection rate was better with the low-resolution modified images (AUC 0.70 vs. 0.76, p = 0.67) than the high resolution images (0.80 vs. 0.37, p < 0.001), although the radiologists were more successful spotting modified images at higher resolutions (0.94, p < 0.0001) than lower resolutions (0.55, p = 0.45) due to better visibility of artifacts.

This is the latest in a number of studies on AI’s vulnerability to adversarial attacks and it’s quite likely that the future will bring even more adversarial AI studies and greater awareness of this potential threat.


The Wire

  • Radiologists from the University of Texas Southwestern Medical Center detailed how they successfully standardized and implemented their imaging protocols. The process involved 1) creating protocol definitions by subspecialty radiologist teams; 2) creating a database to link clinical imaging protocols to specific modalities; and 3) providing a protocol library to all users and locations (and keeping it updated). They noted that 1) assigning a project champion; 2) having an effective modality-specific operational committee; 3) appointing a project lead; and 4) electronically publishing the protocol database were keys to their success.
  • Siemens Healthineers and EquipX announced a partnership that will make EquipX’s clinical equipment portfolio management software available to Siemens Healthineers clients. EquipX provides insights into operating costs, utilization, and business profitability to support informed decision-making regarding capital equipment purchases.
  • A new NYT Op-Ed by cardiologist, Sandeep Jauhar, criticized physicians for complaining about Medicare’s upcoming appropriate use criteria requirements, suggesting that doctors have shown little progress in curbing imaging overuse so they “shouldn’t be outraged when outsiders try to do it for them.” Jauhar recapped the various criticisms against the AUC requirements (ineffective, inefficient, based on a lack of medical understanding, etc.), but focused more on how physicians have allowed unnecessary imaging to become a problem and haven’t made enough efforts to create their own regulations.
  • HeartFlow announced the FDA clearance of its HeartFlow Planner tool, which allows interventional cardiologists to model coronary artery disease (CAD) treatment scenarios in real time before entering the cath lab. The pre-procedure planning tool is based on a model of patients’ HeartFlow FFRct Analysis, allowing physicians to identify blockages and explore treatment scenarios to understand the impact of each modeled treatment strategy.
  • The Australian Nuclear Science and Technology Organization (ANSTO) shut down its Lucas Heights reactor in Sydney for the second time in three months, creating “significant impacts” on the country’s molybdenum-99 supply. Some in the country are expecting the shortage to reach crisis levels by this week, forcing clinicians to only scan patients in most serious need of diagnostics until overseas isotope suppliers are able to get MO-99 to roughly one third of normal levels by the end of September.
  • Bayer announced its new MEDRAD Stellant FLEX CT Injection System (150 ML and 200 mL sizes), which launches with new injector technology and purchasing options, as well as usability and automation improvements. Bayer emphasized the new system’s inclusion in its TechCARE non-obsolescence program, which combines Bayer’s service plan with two hardware upgrades over the course of a three-year contract.
  • New research published in the European Journal of Radiology found that although CT-FFR is highly accurate in detecting lesion-specific ischemia, it may only be superior to cCTA for male patients. The study reviewed 351 patients (73.5% male) with 525 vessels (398 men, 127 women), finding that CT-FFR achieved strong sensitivity (78% men, 75% women), specificity (79%, 81%), positive predictive value (75%, 61%), and negative predictive value (82%, 89%). CT-FFR’s AUC (0.83 for men & women) was much better than cCTA alone (0.76 men, 0.74 women), although CT-FFR’s p value was much better with male patients (0.007 for men, 0.12 women) potentially because there were far more male patients in the study or due to differences in women’s vessel diameter.

The Resource Wire

  • How much does an MRI scan cost? According to Medmo, that depends. Scans made with the exact same device on the exact same body part could cost $225 at one facility and $2,500 at another. Medmo also provides some advice to make sure patients don’t pay too much for their scans, including using the Medmo Marketplace where the average MRI costs between $225 and $700.
  • This Carestream case study compares images of foot trauma captured using the OnSight 3D Extremity System to images captured on 2D X-rays.


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.