“Quick, quantified, precise.”
Dr. Mark Michalski, Director of Center for Clinical Data Science at Massachusetts General and Brigham and Women’s Hospitals, sharing his three-word vision of the future of healthcare.
Attack of the Mid-Levels
We’ve known for a few months that CMS’ 2019 schedule will allow registered radiologist assistants (RRAs) and radiology practitioner assistants (RPAs) to perform outpatient diagnostic imaging procedures under physician supervision, but with that day approaching, this new rule has some of the folks on the Auntminnie.com message boards riled up. These radiologists have seen lower-cost mid-levels (PAs, NPs, CRNAs, etc.) encroach into anesthesia and dermatology and are not excited about that happening in radiology. One warned that when “mid-levels get their foot in the door, they push for greater autonomy and independence and it becomes one big goddamn problem,” calling on fellow radiologists to “prevent the problem before it occurs.” These reactions may not be coming from a place of abundance, but it’s important to know they exist and should be considered if/when providers integrate RRAs and RPAs into their radiologist workflows.
Another Reason to be Paranoid about AI
Swiss researchers created a deep learning neural network algorithm capable of manipulating mammograms by either inserting or removing lesions from the images. The researchers had both good and bad news about their CycleGAN-based algorithm, suggesting that it could help radiologists evaluate scans for cancer but might also allow cyber attackers to cause misdiagnoses by adding or hiding lesions. The researchers trained the CycleGAN algorithm on 680 mammograms from 334 patients (resized to a lower resolution) and then used the algorithm to create altered mammogram images that a three-radiologist panel was unable to distinguish from real images. However, CycleGAN wasn’t able to accurately mimic natural breast patterns when trained with higher-resolution images, suggesting that there’s little threat of algorithm-influenced misdiagnoses today, but it’s something to prepare against in five to ten years.
Should GE Tell You About its AI Platform?
GE Healthcare just published a star-studded interview on LinkedIn Pulse. The interviewer, who happens to be company president and CEO Kieran Murphy, started off by addressing the hyperbole and fatigue swirling around AI, followed by a quick rundown of GE’s AI activities (mainly its new Edison AI platform and new pneumothorax algorithm), before kicking-off a Q&A session with MGH/BWH’s clinical data science guru, Dr. Mark Michalski. Dr. Michalski shared his perspective on AI’s opportunities (care and efficiency), hype (doubt is natural, but AI will deliver), its fit with radiology (AI is good with images; radiology has well-structured data and a solid IT infrastructure), AI ethics (AI needs to be unbiased and democratized), and the future of healthcare (“Quick, quantified, precise.”). With GE’s spinoff looming and the evolution of AI arguably coming faster, you can’t blame GE for working to build its top executive’s brand (here’s another example) and reinforce its position as an AI leader. This can be a hard trick to pull off in an organic and value-added way (especially within a single piece), but in this case GE stuck the landing.
CCTA Before ICA
New research found that coronary CT angiography (CCTA) may be the appropriate first step to determine whether patients with suspected coronary artery disease would benefit from invasive coronary angiography (ICA), potentially reducing the number of patients who undergo the invasive and costly procedure. The study grouped 1,600 patients who were referred for invasive coronary angiography (ICA), with half receiving a CCTA as a first step to determine if ICA was necessary (23% still received ICA, 77% avoided ICA) and the other half immediately receiving an ICA (61.1% of those ended up being unnecessary). After 12 months, 4.6% of patients from both groups had major adverse cardiovascular events, but the CCTA group had a lower rate of coronary revascularization (13% vs. 18%) and saved an average of 57% in initial diagnostic costs ($1,183 vs. $2,755).
The Future of Telemedicine
A new survey (n=283) from Reaction Data suggests that although 67% of physicians and healthcare executives are not aware of CMS’ upcoming telemedicine reimbursement increases, knowledge of these increases is enough to encourage 63% of respondents to speed-up their telemedicine adoption. The survey revealed significant overall growth potential for telemedicine, as 53% of respondents are still not involved with telemedicine at all, with the remaining 23% contracted with a telemedicine provider and 24% using a homegrown solution. Understanding that telemedicine adoption and awareness is almost certainly higher in radiology, it makes sense that these healthcare-wide increases in telemedicine adoption will still have an impact on the uses, perspectives,and expectation of teleradiology.
The Wire
- Chinese medical imaging AI company, VoxelCloud, completed a $50 million Series B financing round that it will use to fund ongoing development and market registration efforts for its portfolio of assisted diagnosis solutions. VoxelCloud is heading into 2019 with serious momentum, already raising $78 million in funding from some big-name investors (Sequoia, Tencent) and attracting a number of clout-carrying western academics to its team.
- A German study (n=172) found that 68Ga-PSMA PET/CT is more effective than PET or CT alone when used for staging before PSA radiotherapy treatment, especially among postoperative patients. The 68Ga-PSMA PET/CT method identified more prostate cancer lesions (171 vs. 156 PET & 85 CT), leading to treatment changes for 62% of patients that would not have happened based on the information in their CT scans.
- Zebra Medical Vision yesterday announced its acquisition of deep learning radiology workflow startup, Radical AI, combining the two Israeli companies with the goal of developing a “premier AI-powered chest X-ray” platform. This appears to be Zebra Medical’s first acquisition.
- A new Kaspersky Lab-funded survey (n=1,758) found that 27% of healthcare IT employees in the US and Canada claim their organization was the target of an attack within the last year, with an amazingly-high 85% of Canadian employees and 78% of US employees previously experiencing ransomware cybersecurity attacks. That’s pretty scary, and so is this quote from a Kaspersky exec: “Healthcare companies have become a major target for cybercriminals due to the successes they’ve had, and repeatedly have, in attacking these businesses.”
- Canadian breast cancer researcher, Martin Yaffe, took a stand against The Canadian Task Force on Preventive Health Care’s (CTF) recently-suggested screening guidelines, claiming that 400 avoidable breast cancer deaths will happen in the country each year if the recommendations are adopted. Yaffe called for earlier screening standards (CTF recommended against screening before 50 years) and a greater focus on informing women with dense breasts so they can seek alternate screening methods such as ultrasound.
- 24x7mag.com published a wide-ranging roundtable discussion on the ultrasound market with leaders from Canon, GE Healthcare, Trisonics, and Philips. The discussion touched on the market’s recent and future evolution (rapid growth, price and application expansion, user/ology expansion, improved image quality, shrinking form factors, automation), top current technological advancements (AI, clinical solutions, resolution, size), and future clinical segment expansion (breast, MSK, liver, acute care, “all”). The article is big, so check it out if you’re interested in going deeper.
- A team of researchers found that radiomic (computer-extracted imaging) features extracted from CT scans can accurately distinguish non-small cell lung cancer adenocarcinomas from granulomas, representing a possible solution for one of thoracic radiologists’ greatest challenges. The retrospective study used radiomic features from non-contrast CT images from 290 patients, organized into a 145-patient machine learning training set (73 adenocarcinomas and 72 granulomas) and a 145-patient CNN and radiologist test set (72 adenocarcinomas and 73 granulomas), with the test set achieving AUCs of 0.75 (intranodular features) and 0.8 (perinodular features) and a pair of trained radiologists achieving AUCs of 0.61 and 0.60.
- The FDA issued a new draft guidance with the goal of clarifying radiation control regulations for manufacturers of diagnostic X-ray systems and their components (focuses on standard compliance, labeling, instructions, and documentation), while revealing plans to harmonize the FDA’s compliance standards with the International Electrotechnical Commission to simplify compliance for global manufacturers. Currently open for comment, the guidance will supersede the FDA’s previous X-ray guidance from 1989.
- Teleradiology marketplace company, DocPanel (300 subspecialty rads, 41 states), signed a pair of deals with NetDirector (HealthData Exchange platform for image and report transfer) and TeraRecon (iNtuition image review and processing system). The announcements highlighted the NetDirector HealthData Exchange’s improvements to sharing speed, access, and costs, while emphasizing TeraRecon iNtuition’s support for processing complex clinical scenarios.
- EhmetDx announced the FDA 510(k) clearance of its 3D CBCT positioning software, used to guide proton beam positioning during treatment. With the software’s FDA clearance, EhmetDx will now focus on commercializing its Mammoknife self-shielded, linac-based breast cancer radiotherapy system.
The Resource Wire
– This is sponsored content.
- Visage Imaging’s Top Five of RSNA2018 blog highlights a major health systems win, some Enterprise Imaging enhancements, an AI entry, a focus on “proving it,” and a shift to FHIRcast. Check it out here.
- The 6th Annual Focused Ultrasound Foundation Symposium included over 250 presentations on the evolution and application of focused ultrasound, including dozens that are now available on the Foundation’s Youtube page.
- POCUS Systems is approved as a Veteran Owned Business with the US Government Office of Veterans Business Development, paving the way for partnerships with the federal healthcare delivery systems.
- Orthopedic Center of Palm Beach County became the latest to install Carestream’s OnSight 3D Extremity System, highlighting the cone beam CT’s image quality and performance as well as its ability to improve patients costs and convenience by performing exams and diagnosis in a single visit.
- This Medmo video details how its healthcare marketplace platform and network of participating radiologists help underinsured patients pay as little as possible for their imaging procedures.
- In this OpenMarkets video, healthcare providers discuss their experience using the OpenMarkets Exchange.
