Uncategorized

Komori Calls Timeout

“It is somewhat paradoxical to use the words ‘small data’ or ‘explainable’ to describe a study that used deep learning . . .”

Harvard graduate student, Hyunkwang Lee, adding some context to his MGH-based team’s recent AI breakthrough, which was indeed able to diagnose and classify brain hemorrhages using a relatively small dataset and then explain the reason behind its decisions.



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.
  • OpenMarkets – A marketplace for healthcare equipment, used by hundreds of hospitals and suppliers to buy and sell imaging equipment in the most efficient way possible.
  • Pocus Systems – A new Point of Care Ultrasound startup, combining a team of POCUS veterans with next-generation technology to disrupt the industry.




The Imaging Wire

MGH’s Small Data and Explainable AI Breakthrough
A Mass General team developed an AI-based system that can quickly diagnose and classify brain hemorrhages using relatively small datasets, while providing explanations for its decisions, potentially overcoming key barriers related to AI’s clinical adoption (delayed diagnosis, large training set requirements, and overcoming AI’s decision evidence “black box”). The system was trained on 904 head CT scans (labeled by hemorrhage subtypes, or no hemorrhage), supporting accuracy by adding steps to mimic the way radiologists analyze images (e.g. adjusting contrast and brightness, scrolling through adjacent CT slices for comparison), later equaling radiologists in detecting and classifying a retrospective test set (100 with, 100 without hemorrhages) and beating non-expert human readers in an additional prospective test set (79 with, 117 without hemorrhages). To overcome AI’s “black box” issue, the system reviewed and saved images from the training set that are representative of the five hemorrhage subtypes in an “atlas of distinguishing features,” later displaying images that are similar to the scan being analyzed as supporting evidence behind its decision. The MGH team is bullish on the new system’s potential to serve as a virtual second opinion, suggesting that it could be “deployed directly onto scanners, alerting the care team to the presence of a hemorrhage and triggering appropriate further testing before the patient is even off the scanner,” revealing plans to next introduce the system into clinical settings and expand from there.


Komori Calls Timeout
“We haven’t given up on acquiring Xerox, but we will not take the initiative at this point. We are not going to try to persuade Xerox.” That’s Fujifilm’s CEO Shigetaka Komori calling a carefully-worded timeout on the company’s efforts to acquire Xerox after months of (very) strong opposition from Xerox’s Icahn-infused leadership. Komori also revealed plans to keep Fuji Xerox and Xerox’s print OEM partnership intact, which should be seen as more good news for those who prefer a stable Fujifilm and want to protect other parts of the company from collateral damage (yes, even its medical division). That last part is the reason we’ve covered this story so much (here’s our Oct, Aug, June coverage). In addition to providing some top-notch corporate drama, Fujifilm’s attempted acquisition of Xerox (risking greater exposure to the troubled print industry) and Xerox’s retaliatory threat to end the OEM partnership (potentially decimating Fuji Xerox, Fujifilm’s largest division) would have posed serious challenges to Fujifilm’s overall health. Even though Komori’s statement suggests Fujifilm technically lost in its bid to acquire Xerox, this one feels like a win for Fujifilm in the long run.


China’s CT Lung Screening Guidelines, Redefined
Chinese researchers published results from a low-dose CT lung cancer screening program (n=14,505) intended to explore new approaches to population selection and define the positive screening result threshold for China’s high-risk cities. The Shanghai-based program screened people who may not be viewed as high-risk in other countries (as young as 35 years old, non-smokers) and did not set a minimum pulmonary nodule size threshold for positive results (any size/density = positive). The screenings revealed lung nodules in 4,336 subjects (29.89%) with lung cancer later confirmed in 178 patients (1.23%, 81% of which proved to be stage I). The Chinese program’s 1.23% cancer detection rate was lower than the well-known NLST and NELSON trials (2.4% and 2.6%), almost surely due to its inclusive population criteria, but the 178 confirmed cancer patients’ 81% rate of stage I cancers was much higher than the NLST and NELSON trials (58% and 64%), revealing early detection benefits. The program also appears to have identified a positive result threshold, as 75% of the 4,336 lung nodules were under 5mm, while 94% of the 178 cancerous nodules were over 5mm, suggesting that 5mm is an effective positive result cutoff in Shanghai and similarly high-risk cities.


GE Healthcare’s Confidential IPO
Sources revealed that General Electric confidentially filed for GE Healthcare’s IPO in mid-December, with a public filing now expected in spring 2019. Although not officially announced, GE’s reported filing represents its biggest step towards creating an independent GE Healthcare since first revealing its spinoff plans last June. The report also gives new insights into timing (previously 12-18 months from June) and suggesting we’ll soon learn more about the division’s spinoff structure (initially targeted public sale of 20% of shares, recently suggested up to 49.9%). Given GE’s need to lighten its debt load and sharpen its focus, it shouldn’t be too surprising that the GE Healthcare IPO is moving along, but this report makes the idea of GE Healthcare going independent in 2019 feel a lot more real.


Merry X-Ray’s Ultrasound Conquest
Merry X-Ray (MXR) ended 2018 with its acquisition of ultrasound parts, probes, and service company, Conquest Imaging, continuing what’s proven to be a busy few years of acquisitions for the San Diego-based company. Perhaps not coincidentally, the acquisition comes just over two months after Conquest Imaging landed in The Imaging Wire for moving its probe repairs in-house, as the cost savings from this in-house shift along with Conquest Imaging’s overall ultrasound service operations will be relied on to allow MXR to offer its clients even lower ultrasound parts and services prices going forward. The acquisition also advances MXR towards its goal of supporting all modalities and its recent push to expand its service, reconditioning, and parts capabilities, which were common themes across the company’s five previous acquisitions since 2017 (here are the previous two in March and April 2018).



The Wire

  • Australian researchers developed a brain biopsy imaging needle, using a miniaturized optical coherence tomography probe to identify blood vessels during biopsies or other neurosurgical needle interventions, allowing surgeons to avoid blood vessels and reduce the risk of causing intracranial hemorrhage. In a study of 11 patients, the researchers were able to detect blood vessels with a sensitivity of 91.2% and a specificity of 97.7% by shining infrared lights onto the brain tissue to identify blood flow.

  • South Korea’s JLK Inspection unveiled its AIHuB medical image diagnosis system, which combines deep learning technologies (neural network solvers and libraries), algorithms, and image processing technology, with what it claims is a “user friendly” interface. The on-site AI system analyzes images from a range of modalities (MRI, CT, X-ray and mammography) and can reportedly detect over 30 medical conditions in 14 regions of the body (e.g. stroke, aneurysms, Alzheimer’s disease, lung/prostate/breast cancer, coronary artery disease, and “other” digital pathology).

  • 3M plans to acquire MModal’s technology business for $1 billion, using MModal’s technology division (~$200m annual revenue, 750 employees) to strengthen 3M’s revenue cycle management and population health platform. Meanwhile, the deal will allow MModal to sharpen its focus on its transcription, scribing and coding services business, which will continue to work with 3M through a strategic partnership.





The Resource Wire

This is sponsored content.

  • By partnering with Medmo, imaging centers can keep their schedules full, their equipment busy, and increase revenue. Here’s where to learn more and get started.

  • This video highlights how healthcare providers can benefit from selling on the OpenMarkets platform, getting multiple offers on their old equipment, instruments, and supplies.







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!