“Everyone is in the dark – It’s not just individual patients who are in the dark. Employers are in the dark. Governments are in the dark. It’s just astonishing how deeply ignorant we are. . .”
Carnegie Mellon economist, Martin Gaynor, on how far we are from actual healthcare transparency.
I’m happy to share the latest Imaging Wire Show, featuring Riverain Technologies’ Chief Science Officer, Jason Knapp. We discuss everything from the evolution of imaging AI, how to get generalizability right, AI’s path forward, and Ohio State football.
If you’ve been needing an imaging AI conversation without an ounce of spin, this is the one for you.
Imaging Wire Sponsors
Arterys | Bayer Radiology | Blackford Analysis
Canon Medical Systems | Fujifilm Healthcare Americas
GE Healthcare | Novarad | Nuance
Riverain Technologies | Siemens Healthineers
United Imaging | Zebra Medical Vision
The Imaging Wire
“The Price for an MRI at Mass General is: $1,019 with a Cigna plan, $3,101 with an Aetna plan, $3,809 with a Humana plan.” That’s the first thing readers see in the New York Times’ recent exposé detailing how widely healthcare costs can vary and how poorly this part of the U.S. healthcare system seems to run.
The Investigation – NYT journalists and a pair of UMD-Baltimore researchers compiled data from 60 hospitals and analyzed the negotiated rates that hospitals are now required to disclose.
Consistently Inconsistent – The authors found wide variations and very few areas of consistency (except that smaller plans often get worse rates than big ones). There were even major inconsistencies when comparing plans from the same payor at the same hospital, such as United-covered MRIs at Aurora St. Luke’s ($1,093 w/ United HMO, $4,029 w/ PPO).
Imaging Emphasis – The authors arguably placed the most emphasis on imaging variations, including an remarkable comparison of knee MRIs at Baptist Memorial in Memphis ($210 to $2,800), Beaumont Hospital near Detroit ($250 to $3,100), and Mass General Hospital ($830 to $4,200). Many of these rates were >10x more than Medicare (particularly at Baptist & MGH) and many were well above rates for patients without any coverage (particularly at Baptist & Beaumont).
Passive Payors – The fact that covered procedures are sometimes 2x more than procedures paid out-of-pocket also calls into question how fiercely insurers are actually negotiating, or whether the current system even gives them reason to negotiate.
The Takeaway – This story won’t come as a major surprise to many Imaging Wire readers, but it’s getting a lot of attention from patients and healthcare advocates, and it definitely casts a negative light on hospitals and payors. It’s going to take a lot for this situation to change, but public scrutiny like this definitely helps.
GE & AI-Enabled Imaging Centers
Some small imaging centers haven’t been getting a fair chance with AI – here’s how GE Healthcare plans to change that.
Nuance & UVM’s Enterprise Imaging Vision
Join Nuance and The University of Vermont Health Network to learn what shaped UVM’s enterprise imaging vision and how they are executing the strategy to achieve it.
- Prostate MRI Protocol AI: Swiss researchers developed a deep learning model that can help clinicians decide whether patients with suspected prostate cancer should receive biparametric MRI (no-contrast, faster) or multiparametric MRI (contrast, slower). The DL model would have correctly assigned 78% of patients to a biparametric or multiparametric protocol, with just 2% of patients requiring re-examination with contrast-enhanced sequences.
- Telerobotic Ultrasound: Scientists from India’s IIT-Delhi and AIIMS developed a Telerobotic Ultrasound System that uses a WIFI-connected robotic arm to remotely perform ultrasound exams. The researchers positioned the Telerobotic Ultrasound System as a way to allow socially distanced bedside exams during the COVID pandemic and expand imaging in rural areas.
- Agent Amnesia: The vast majority of patients who’ve experienced allergic reactions to CT contrast agents don’t know what agent they’re allergic to, creating challenges with their future scans. That’s from a Clinical Imaging study (n = 307 w/ prior CT agent reactions) that found just 1.6% of these patients could name the agent that caused their previous reaction. Less than half could remember where or when the reaction occurred, limiting clinicians’ ability to look up the previous event or confirm if the agent is still in use.
- Covera’s Series C: Covera Health completed a $25m Series C round ($57m total funding) that it will use to scale its Radiology Centers of Excellence (CoE) program, expand into new specialties, and add new analytics and integration capabilities. Covera has made solid progress over the last two years, including a high-profile alliance with Walmart, landing its first insurer client, becoming a patient safety organization, and integrating its first imaging AI tools – making this commercialization push worth watching.
- IR Consultants: A new JACR study found that US interventional radiologists saw their industry consultation fees increase by 65% between 2014 and 2018 ($2.8m to $4.6m), driven by more IRs charging consulting fees (156 to 219) and rising annual IR consulting earnings ($4,327 to $5,419).
- TI-RADS’ High Standards: ACR TI-RADS recommends far fewer thyroid biopsies than other thyroid risk categorization systems, avoiding unnecessary exams but potentially missing some malignant nodules (at least until the next follow-up). A JACR study analyzed thyroid ultrasound reports for 27,933 nodules (12,208 patients), finding that TI-RADS recommended fine needle aspiration for 29.1% of nodules, far lower than the other systems (ATA = 58.7%, EU-TIRADS = 38.9%, K-TIRADS = 57%, and AI-TIRADS = 26.3%), largely due to TI-RADS’ higher risk criteria for level 3 & 4 nodules.
- Racial AI Goes Mainstream: News that imaging AI can accurately detect patients’ race quickly expanded beyond radiology and AI circles, after a number of mainstream publications covered this sensational study (Wired, Vice News, The Register, Daily Mail, more). It’s not surprising that this story made it to mainstream media outlets, and it also shouldn’t be a surprise if more patients start asking how undetectable AI bias might influence their diagnosis.
- Physician Mistrust: New research from patient engagement company SymphonyRM revealed patients’ growing distrust for their doctors since COVID-19. The survey (n = 1,192) found that 41% of patients lost trust in their doctor during the pandemic, due to infrequent communication about COVID-19 (53%), slow adoption of virtual care (29%), and under-utilization of digital communication tools (24%). These patients probably didn’t have radiologists in mind when surveyed, but their desire for proactive digital / virtual communication likely applies to all specialties.
- DBT AI Progress: A Duke University team unveiled a solid effort to support DBT AI development, which has been hindered so far (few publicly-available images, low prevalence of breast cancer, sharing/privacy rules, laborious annotations). To address this, the team annotated a 22k reconstructed/anonymized DBT dataset (5,060 patients) and used it to train and test a breast cancer detection deep learning model, making both publicly available to help other groups expand their DBT training data and improve their own models.
- The Patient-Centered MPI Report: Myocardial perfusion imaging reports are historically complex, but they will have to become a lot more patient-friendly as more patients access their own MPI reports. To help with this change, a Kansas City-based team performed focus groups and a pilot test (n = 36 & 123 patients) and concluded that MPI reports should include: 1) Written info; 2) Simple language and graphics; 3) Comparisons versus normal results; 4) Personalized risk estimates; 5) Direct verbal explanations from a clinician.
- Vysioneer Validation: A new study from Stanford and Vysioneer found that the startup’s VBrain solution was able to detect and automatically contour metastatic brain tumors with relatively high accuracy and generalizability. The researchers used head CT and T1 MRI scans from 60 patients with 321 extremely small brain metastases (median size: 0.1 cc). VBrain contoured the tumors comparably to the patients’ physicians (mean false positives: 0.657 tumors / case, lesion-wise sensitivity: 84.5%).
The Potential of Population Health Cardiac AI
Cardiovascular disease is the number one global cause of death, but it’s also preventable, which is one of the reasons why Zebra-Med views AI-powered cardiovascular screening as the next frontier in population health.
The Resource Wire
- Learn how leveraging the right cardiology image and reporting platform drives performance and outcomes in this Fujifilm Healthcare white paper.
- Dynamic imaging on Dual Source CT can help you provide more confident, precise, and accurate answers. Download Siemens Healthineers’ clinical case collection to see powerful examples where functional imaging with DSCT is making a difference.
- Did you know 80% to 90% of sonographers experience pain while performing scans at some stage in their career? Check out this Canon Medical Systems video detailing its latest innovations that improve sonographer comfort and help reduce risk of injury.
- Here’s a quick introduction to Blackford Analysis’ dedicated AI platform and its service for the selection, deployment, orchestration, and use of imaging applications and AI.
- Check out this Imaging Wire Q&A, where Bayer Radiology’s Dennis Durmis and MITA’s Peter Weems discuss the medical device service debate and how ongoing legislation and regulation efforts could impact patients, clinicians, and OEMs.
- Tune in to Riverain Technologies’ on-demand webinar demonstrating how its AI solutions integrate into LucidHealth’s radiology workflow and sharing best practices on how to combine AI with radiologist expertise.
- See how ED physicians at France’s Hospital of Maubeuge reduced emergency imaging error risk by 75% using Arterys Chest MSK.
- United Imaging’s “all-in” approach means that every system ships with its entire suite of features and capabilities (no options), giving its clients more clinical flexibility and predictability.
- Learn how Novarad’s Nova RIS system helps radiology departments streamline front office workflows, ensuring they get all the reimbursements they’ve earned while also providing a great patient experience.