“I hate it.”
How Dr. Luke Oakden-Rayner really feels about last week’s paper criticizing Google’s mammography AI study (even though he respects the authors “a lot”).
Know anyone who made outstanding contributions to radiology practice and outcomes this year? Nominate them for an Imaging Wire Award.
Imaging Wire Sponsors
- Bayer Radiology – Providing a portfolio of radiology products, solutions, and services that enable radiologists to get the clear answers they need.
- GE Healthcare – Enabling clinicians to make faster, more informed decisions through intelligent devices, data analytics, applications and services.
- Healthcare Administrative Partners – Empowering radiology groups through expert revenue cycle management, clinical analytics, practice support, and specialized coding.
- Hitachi Healthcare Americas – Delivering best in class medical imaging technologies and value-based reporting.
- Nuance – AI and cloud-powered technology solutions to help radiologists stay focused, move quickly, and work smarter.
- Riverain Technologies – Offering artificial intelligence tools dedicated to the early, efficient detection of lung disease.
- Siemens Healthineers – Shaping the digital transformation of imaging to improve patient care.
The Imaging Wire
Efficient Extenders
Faced with rising volumes of high-labor and low-reimbursement chest radiographs, UPenn is leveraging radiology extenders (REs) to help address this challenge, and they’ve found that REs can really speed up the CXR reporting process.
- The Study – Two REs and junior residents drafted 49 CXR reports and were evaluated based on how they improved four attending thoracic radiologists’ flow rates (interpretations finalized per hour).
- The Disclaimer – The goal of this study was to show how REs improve radiology productivity, not compare their performance against residents. However, the because the study measured the REs and residents, the “The Results” section below will compare those measurements.
- The Results – The study’s attending radiologists finalized 93.6 RE-drafted cases per hour, which was 52% faster than when the radiologists produced the reports on their own (62.4 cases per hour), and 36% faster than when the rads finalized the resident-drafted reports (72.6 cases per hour). The RE-drafted reports and resident-drafted reports were found to have “not statistically different” accuracy.
- The AI Angle – Some might argue that AI will take over for these REs soon enough, but the authors suggest that REs’ role will similarly evolve, taking on a greater focus on managing AI workflows.
- The Takeaway – Not everyone is comfortable with using RTs or RAs for drafting reports (reduces resident training, undermines radiology), but this UPenn team has been building its RE case for several years (previously for drafting MSK reports). Given that imaging volume growth has outpaced the supply of radiologists and trainees, the authors make it quite clear that they believe REs deserve a place in radiology reporting workflows.
Google Responds
Google quickly responded to the recent paper criticizing its mammography AI study’s lack of transparency (data, code, model design), agreeing that transparency and reproducibility are important, but mainly focusing on justifying its decisions. Here’s the short version of Google’s response, plus some fun follow ups:
- Releasing Its Data – Google noted that much of its data is indeed available through OPTIMAM, but it can’t share the remaining source data because Google doesn’t own it.
- More Methodology Details – Google added an addendum to its original mammography AI study with more training methodology details.
- Open Sourcing the Code – Google’s model is largely based on CNNs that are already open-source and the descriptions included in its study should be enough to allow AI experts to expand on its approach.
- Releasing the Model – Google detailed the regulatory, legal, and responsibility factors keeping it from releasing a “containerized version” of this model, reminding that the model is still at the proof-of-concept phase.
It’s safe to say that some authors of the paper that criticized Google were not satisfied with this response, accusing Google of abusing its access to patient data while using data privacy as an excuse to not release its code or model details (among other criticisms). This led to Dr. Luke Oakden-Rayner stepping up to defend Google and AI science, suggesting that many of the authors’ criticisms are scientifically incorrect and calling their response unprofessional (among other criticisms).
The Wire
- Follow the AI Money: A new Oxford paper found that the majority of comments on the FDA’s AI/ML regulatory framework were submitted by people with financial ties to AI. Out of 125 analyzed comments, 79 (63%) came from parties with financial ties to AI, only 10 (8%) comments came from folks without financial ties, and they couldn’t confirm “the absence of financial ties” for the other 36 (29%) comments. Perhaps not coincidentally, 86% of the comments didn’t cite scientific literature. It’s hard to blame the AI industry for weighing in on this framework, but the authors still suggest that the FDA should mandate conflict of interest disclosures and encourage the use of scientific evidence.
- Tertiary MRI Discrepancies: A review of 357 body MRI secondary interpretations performed at a tertiary care center and the associated original interpretations from outside hospitals found discrepancies in 68.9% of cases. The AJR study attributed ‘faulty reasoning’ for 34.3% of the discrepancies (misidentifying an abnormality) and ‘satisfaction of search’ (ending interpretation upon finding first abnormality) for 15% of the variances, while encouraging tertiary hospitals to use body MRI subspecialists for secondary reads.
- BrainKey DTC: A new profile on Auntminnie.com introduced us to direct-to-consumer brain MRI startup, BrainKey, which is provides curious/concerned consumers with brain scans, radiologist reports (and referrals, if necessary), and 3D models of their scans to review on their own. BrainKey comes with a solid pedigree (Stanford and Y Combinator alumni) and it appears to be part of a new consumerization trend in advanced imaging, joining DTC cancer screening startup Ezra.
- Fujifilm & Volpara: Fujifilm and Volpara Solutions expanded upon their 6-year partnership, giving centers that use Fujifilm’s ASPIRE Cristalle DBT system access to Volpara’s Breast Health Platform. The Volpara platform gives mammographers live feedback, analyzes mammogram quality, provides density assessments.
- BWH’s Follow Up Score: A new JACR study detailed how Brigham and Women’s Hospital’s creation of a new follow up program helped the hospital cut its unscheduled diagnostic imaging orders by 49% (from 52k of 607k to 21k of 475k). BWH’s System for Coordinating Orders for Radiology Exams (SCORE) monitors all scheduled scans, detects unresolved/expired imaging orders, notifies providers and patients of their unresolved orders, and either schedules the patients or cancels the orders.
- Imaging Confidence Still Rebounding: After a very ‘unconfident’ Q2, AHRA’s Q4 Medical Imaging Confidence Index (MICI) revealed that U.S. imaging managers and directors (n = 151) are cautiously optimistic for the second straight quarter. The new report detailed on Auntminnie.com revealed a “neutral” 109 overall confidence score (vs. 103 in Q3, 72 in Q2), a “high confidence” 124 score for scan/IR volume growth (vs. 125 in Q3, 59 in Q2), and a “high confidence” 129 score for their role as a profit center (vs. 131 in Q3, 76 in Q2). The imaging leaders still aren’t all that confident about their access to capital (86 in Q4, vs. 67 Q3 and 65 Q2) or whether they’ll receive adequate reimbursements (90 in Q4, vs. 88 Q3 and 80 Q2), but both are improving.
- Bayer’s Density Campaign: Bayer is collaborating with HealthyWomen on their new ‘Understand Your Density’ campaign intended to educate women with dense breasts about their supplemental screening options. Through the initiative, Bayer will launch a quiz on UnderstandYourDensity.com to help women test their breast density knowledge, learn about their breast cancer screening needs, and outline how the women can speak with their doctor about supplemental screening.
- Ultrasound for Fibrosis: Researchers from North Carolina State University and UNC teamed up to develop a new ultrasound technique to assess pulmonary fibrosis and pulmonary edema treatments, potentially serving as an alternative to CT scans and pulmonary function tests (lower cost, more accessible, no radiation, doesn’t require radiologist reads). The technique leverages transducer elements from conventional ultrasound probes to direct multiple ultrasound waves at lung tissue. The ultrasound data is then fed into a computational model that determines the density of healthy alveoli in the lung, which can be used to assess fibrosis levels.
- How to Code for COVID: Healthcare Administrative Partners shared some guidance on how imaging centers and non-hospital faculty can make the most of a new CPT code (99072) that’s intended to help recover COVID-related operational costs. Here are some of HAP’s details and suggestions: 1) The code can only be reported once per patient visit, per day; 2) The cost must be higher than a typical visit (e.g. additional clinical labor, pre-visit screenings, cleaning, etc.); 3) Filing this code will require some math (e.g. dividing the cost of PPE by the number of patients, multiplying cleaning time per patient by staff hourly pay rates); 4) It will also require documentation in every imaging report; 5) The code is not yet supported by Medicare, but is reimbursed by some payers (e.g. Aetna, Cigna, Humana); 6) Some practices might not be able bill for COVID-related costs if they received HHS funds.
- Touro’s Vave Ultrasounds: Touro University Nevada medical students will soon list personal handheld ultrasounds among their school supplies, after Nevada’s largest medical school signed a partnership with ultrasound startup Vave Health. The partnership will also provide the students with ongoing Vave EduPlus memberships, which Vave suggests makes this partnership unique, even as handheld ultrasound med school partnerships like this become increasingly common.
- Telerad Growth & Challenges: A new AJR paper detailed teleradiology’s considerable adoption growth and efficiency contributions, while addressing the challenges limiting teleradiology’s potential impact. Here are some of the key challenges: 1) Complicated state-specific licensure processes and oversight regulations; 2) Telerads’ limited access to patients’ clinical information and limited ability to communicate with on-site care team; 3) Overnight shifts might be typical for the practice of teleradiology, but they aren’t ideal for many and telerad reimbursements don’t account for this after hours work.
- Inland Imaging Adds Tri-City: Northwest U.S. radiology practice, Inland Imaging (100 rads, 700 staff), announced its acquisition of Kennewick, WA-based Tri-City Radiology (15 rads, 1 imaging center). The acquisition adds to Inland Imaging’s presence in Washington’s Columbia Basin after expanding there via its acquisition of CBI Radiology in early 2019 and it follows similar acquisitions in 2018 (Richland, Washington 2018) and 2017 (Missoula, Montana).
- OCT-FFR: The combination of fractional flow reserve (FFR) and OCT imaging can help improve clinicians’ accuracy in identifying high-risk cardiac lesions in patients with diabetes. The recent COMBINE (OCT-FFR) trial performed FFR on patients with diabetes and coronary syndrome, and then performed OCT on the FFR-negative patients to identify lipid-rich and TCFA lesions. They found that the TCFA-negative patients had much lower rates of major cardiac events within 18 months (3.1%) than TCFA-positive patients (13.3%).
The Resource Wire
– This is sponsored content.
- In its latest Q&A, Nuance sat down with Infervision North America Director, Matt Deng, Ph.D., to discuss the company’s journey and the impact of its InferRead Lung CT solution.
- What can your radiology practice do to improve patient satisfaction? This Healthcare Administrative Partners blog post outlines some steps to improve patient experience including issuing surveys and applying feedback, supporting a range of payment methods, and helping patients understand their coverage.
- This Riverain Technologies case study details how Einstein Medical Center adopted ClearRead CT enterprise-wide (all 13 CT scanners) and how the solution allowed Einstein radiologists to identify small nodules faster and more reliably.
- Did you tune into the Siemens Healthineers MRI Summer Webinar Series? If not, you still have the chance. The series featured 5 leading MRI experts that discussed hot topics within the magnetic resonance imaging world. Register for access to watch the webinar recordings at your own convenience.
- Learn how Scott County Hospital’s new CT helped them increase their imaging bandwidth in this upcoming Hitachi webinar.
- These two independent Phase III trials provided new evidence supporting the accuracy and effectiveness of using Bayer’s Gadavist contrast agent for cardiac MRI.
- This GE Healthcare post details how its Edison Developer Program creates a development and distribution ecosystem for independent software vendors and developers.
