“Garbage in, Cabbage out.”
Australian radiologist, Luke Oakden-Rayner, in his review of Stanford’s CheXNet AI paper, suggesting that although the AI system’s training labels were flawed, it was still able to overcome labeling noise and achieve decent results. Much like cabbage, the results were “edible, but not 5-star dining.”
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.
- 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
AI May Replace Radiologist …Typos
“Mistakes make it into many radiology reports. Deep learning can help fix that.” This bad news/good news combo is from a Columbia University team that trained a neural sequence-to-sequence model (seq2seq) to identify and correct errors that often make their way onto radiology reports (a previous Mayo study found errors in 9.7% of all reports). The team first used its seq2seq model on head CT and chest X-ray reports that intentionally contained errors (incorrect word insertions, deletions, and substitutions), and performed quite well, detecting 90.3% and 88.2% of all errors with 97.7% and 98.8% specificity. Next, the team tested the model on actual reports (with unintentional errors), correctly identifying a solid 98.6% of error-free sentences (789 of 800) while only correctly identifying 38.6% of all error-included sentences (157 or 400). The team admitted that more advancements are needed before a solution like this is ready, although they were firm in their recommendation that a similar autocorrect feature should eventually become a part of radiology report creation.
Hologic and Fujifilm Out of Court, On the Market
Hologic and Fujifilm are reportedly settling their respective ITC and federal court cases, allowing both companies to continue to sell their mammography systems in the US. Although not yet publicly announced, the deal would bring an end to over 18 months of litigation, which started with Hologic’s DBT patent infringement claim against Fujifilm in mid-2017 that was matched with a federal lawsuit from Fujifilm accusing Hologic of patent infringement and illegal monopolistic activities in spring 2018. Things really started to heat up in mid-2018, when the ITC ruled in favor of Hologic and recommended that Fujifilm be barred from importing and selling its Aspire Cristalle DBT system stateside. And that may have been enough to lead to this settlement. We’ll learn more about the settlement in the near future, but at least for now it appears that Hologic and Fujifilm are out of court and both of their DBT lines are staying on on the market.
A team of Northwestern University researchers developed a new imaging technology capable of capturing 3D images in “the smallest capillaries,” potentially improving clinicians’ ability to measure oxygen delivery and allowing for earlier detection of diseases associated with low/no capillary blood flow (e.g. CV disease and cancer). As its lengthy name suggests, Northwestern’s new Spectral Contrast Optical Coherence Tomography Angiography (SC-OCTA) system combines spectroscopy with conventional optical coherence tomography (OCT), allowing imaging of both still and moving tissues (unlike ultrasound) without imaging agents/dyes or exposing patients to radiation (unlike most modalities). The drawback is, SC-OCTA is unable to image deeper than 1mm, although the researchers are working on attaching SC-OCTA to an endoscopic probe in order to image organs.
Healthcare Growth Partners’ January 2019 healthcare private equity survey (n= 105 PE funds) found that most investors remain bullish about health IT, with 35% reporting more deals in health IT than in other sectors (vs. 42% “about the same” and 22% “less active”), and only 5% revealing that their health IT investments are performing worse than expected. However, a notable 28% of all investors believe that health IT is in a bubble (up from 25% in 2016). Given the money flowing into healthcare IT companies, this 28% minority may have a valid point, and the 64% of all investors who saw health IT companies as overvalued in a late-2018 KPMG / Leavitt Partners survey would probably agree with them.
Siemens’ MAGNETOM Lumina 3T FDA Cleared
Several months after debuting at RSNA, Siemens Healthineers announced the FDA clearance of its new MAGNETOM Lumina 3T 70-cm MRI, bringing BioMatrix technology (patient personalization / positioning tech) downstream to its “broad market” MRIs for the first time. The new system also features Siemens’ Turbo Suite (up to 50% faster exams), Innovision (in-bore infotainment option), and Tim 4G and Dot (automates reproducible scan procedures) technologies, plus an optional dockable table. The MAGNETOM Lumina is a net new addition to Siemens’ 70cm wide bore lineup, positioned between the MAGNETOM Skyra and MAGNETOM Vida.
- Canon Medical Systems announced its new 16-slice Aquilion Start CT system in Japan and at Arab Health 2019 this week, bringing what appears to be net new addition to the bottom of it’s Aquilion CT lineup. Targeted at smaller and regional healthcare facilities, the new CT combines technologies found across the Aquilion platform (780mm gantry, low dose imaging thru image reconstruing, navigation mode operation, metal artifact reduction, patient motion correction) with a compact design (9.8m min footprint), and likely lower pricing.
- “Radiology should be the medical specialty most primed to incorporate artificial intelligence (AI) into our workflow; however, it seems that many of us are not even sure exactly what that means. This disconnection can lead us to have a negative view of the future of technology-enhanced radiology, which can then discourage medical students from entering the field.” That’s University of Wisconsin neuroradiologist, Allison Grayev, MD, in a recent ACR editorial suggesting that radiology’s AI laggards and naysayers may scare future talent away from the profession.
- In related news, the Royal Australian and New Zealand College of Radiologists (RANZCR) created a working group that will evaluate the potential impact of AI on radiology, AI safety and ethics, accreditations and regulatory implications, and training needs for radiologists who adopt AI. The working group includes radiologists, data scientists, computer scientists, and other AI professionals, who will work in collaboration with Standards Australia (Australia’s non-profit standards development agency).
- Kromek signed a $58 million, seven-year contract to provide an unnamed medical imaging OEM with cadmium zinc telluride (CZT) detectors, representing a significant win given that Kromek signed $80 million in contracts over the last three years combined. Considering that GE and Spectrum Dynamics are the only players with CZT SPECT systems, the identity of Kromek’s “unnamed” OEM may be a pretty straightforward guess, unless another player is about to enter the CZT-based segment.
- A study from University of Vermont researchers (n=86,349 DBT and 97,378 FFDM screening exams) found that DBT has lower recall rates than full-field digital mammography (FFDM) alone (7.9% vs. 10.9%), but achieves “very similar” biopsy rates and “effectively equivalent” cancer detection rates. Noting that other studies have found DBT to achieve higher cancer detection rates (CDRs) than FFDM, the researchers suggested that their study’s “effectively equivalent” CDRs may be due to their sample’s relatively high FFDM CDR compared to other studies (UVM 5.6 cancers per 1k exams vs. other studies <5 cancers per 1k exams).
- Konica Minolta Healthcare released a software upgrade for its SONIMAGE HS1 ultrasound system, bringing a range of clinical workflow improvements. The SONIMAGE HS1’s software additions include its UltraAdjust one-touch image optimization feature (automatically changes imaging parameters by adjusting depth), new AI-based voice recognition/command technology (allows hands-free operation, improve ergonomics), and a panoramic view option (stitches together a series of images for a broader field of view).
- A team of healthcare heavyweights are teaming up to launch a new blockchain healthcare network intended to improve transparency and interoperability. The IBM-led team includes PNC Bank and insurers Aetna, Anthem, and Health Care Service Corporation, who will collaborate on “promoting efficient claims and payment processing, to enable secure and frictionless healthcare information exchanges, and to maintain current and accurate provider directories.”
- Researchers identified an MRI-based prostate cancer grading system that helps clinicians understand the risk of extraprostatic extension (EPE – the extension of a tumor beyond the prostate glands), potentially allowing for earlier detection and improved treatment. The four-grade system includes: 0) no evidence of EPE 1) curvilinear contact length of 1.5 cm OR capsular bulge and irregularity (24% of EPE patients), 2) both of the above features (38% of EPE patients), and 3) MRI-verified EPE or tumor expansion past the prostate (66% of EPE patients). When this EPE MRI grading system was combined with clinical features, EPE prediction outperformed MRI-alone.
- Perspectum Diagnostics announced the FDA 510(k) clearance of its MRCP+ biliary visualization software, which uses quantitative MRI and AI algorithms for MRCP images to allow “improved visualization of intra-hepatic ducts and measure the widths of bile ducts, biliary tree volume and gallbladder volume.” The new solution is intended to help diagnose and monitor PSC (Primary Sclerosing Cholangitis), which has historically been difficult due to PSC’s lack of effective biomarkers, and is expected to support the development of new PSC treatments and early management of the disease. MRCP+ already achieved CE Marking in Europe.
- A team of Virginia Tech researchers trained a machine learning algorithm to diagnose mental illnesses by combining patients’ brain fMRI scans with clinical data (behavioral data, speech data, psychological assessments, and soon saliva and blood samples). The scientists are initially trying to understand how the brains of both healthy patients and patients with mental illness react to situations, and eventually evaluate the most effective mental illness treatments.
The Resource Wire
–This is sponsored content.
- A new Gallup poll revealed that 13.7% of American adults are uninsured, representing the highest rate since 2013 when it was 18% (before most of ACA went into effect). These are the exact patients who can be helped by the Medmo platform, which connects high-deductible or uninsured patients with radiology centers, ensuring the best value for patients and a profitable revenue stream for imaging centers.
- In this article, a UC San Francisco team details how it is using high-intensity focused ultrasound (HIFU) technology in a new thermal ablation technique able to completely, noninvasively destroy tumors within the body.
- Carestream will showcase the latest version of its Vue Clinical Collaboration Platform at HIMSS 2019, a platform that manages millions of images and is in use at a growing number of large healthcare organizations.
- 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 system