“Meaningful machine learning for health is not just about turning a crank, but it requires the careful and thoughtful application of analytical techniques.”

An editorial led by MIT’s Tom J. Pollard in response to a recent study suggesting that physicians might be able to develop their own algorithms without working with AI development experts.

---

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
• Nuance – AI and cloud-powered technology solutions to help radiologists stay focused, move quickly, and work smarter
• Pocus Systems – A new Point of Care Ultrasound startup, combining a team of POCUS veterans with next-generation genuine AI technology to disrupt the industry
• Qure.ai – Making healthcare more accessible by applying deep learning to radiology imaging

---

The Imaging Wire

Physicians Replace Developers

A new study in The Lancet Digital Health demonstrated that physicians without coding experience can create algorithms that are able to classify medical images comparable to the leading platforms. As you might imagine, this prompted some debates.

• Study – A pair of physicians without AI experience studied AI programming for 10 hours before developing and testing an image classification algorithm using five open-source datasets and Google Cloud AutoML. The algorithm performed binary classification tasks relatively successfully (sensitivity 73.3–97%; specificity 67–100%; AUPRC 0.87–1.00) but was unsuccessful validating external data (sensitivity 49%; AUPRC 0.47, positive predictive value 52%).

• Ramifications – The authors suggested that automated deep learning might lead to the “democratization of sophisticated algorithmic modelling in health care,” allowing physicians to develop AI in-house without expertise in the mathematical, statistical, and programming components of AI.

• The Debates – The study reignited online discussions about the pitfalls of AI developed by technical experts who don’t have medical expertise and whether the physician-trained algorithms may have advantages (if you had to choose between tech or domain expertise). Others pointed out the ethical issues (and safety concerns) of allowing healthcare AI tools to be programmed by folks without the appropriate expertise. Both of these arguments make sense, although few would argue that automating the AI development process is a good thing.

An Epidemiologic Transition

New research published in The Lancet revealed a major “epidemiologic transition” taking place, with cancer surpassing cardiovascular disease as the leading cause of adult mortality in high-income countries and some middle-income countries.

• Study – The researchers analyzed 162.5k patients between 2005 and 2016 (35-70yr-old, 21 countries, 5 continents), with 11,307 participants dying during the study.

• Results – Cardiovascular disease was the leading cause of death globally (40%), but only caused 23% of deaths in high-income countries (vs. 41% in mid-income countries and 43% in low-income.). Meanwhile, cancer was responsible for twice as many deaths than cardiovascular disease in higher-income countries, despite its position as the distant second-leading cause of death in middle and low-come countries.

• Implications – Noting that cardiovascular disease mortality is expected to fall in poorer countries, the researchers believe that cancer will become the leading cause of death globally within a few decades.

Walmart Health

Walmart took another step into healthcare last week, opening a standalone Walmart Health clinic in Dallas, Georgia (population 13.2k, 32mi from Atlanta) that will lead with “low, transparent pricing for key health services.” This is a pretty traditional value proposition for Walmart (key products/services at low costs) but it’s part of a non-traditional shift for the retail giant.

• Walmart’s Key Health Services – The new Walmart Health pilot location will provide a range of services including primary care, dental, mental health counseling, and vision, and will have X-ray systems and labs on-site.

• Walmart’s Healthcare History – The new standalone clinic follows 19 other in-store Walmart Care Clinics across Texas, South Carolina, and Georgia, suggesting that Walmart is diversifying its clinic strategy. However, these clinics are just one component of Walmart’s healthcare strategy that began with building one of the largest retail pharmacy businesses in the US (nearly all 4.7k locations) and became more ambitious as Walmart built out a healthcare leadership team and was increasingly rumored to be acquiring a major healthcare player (particularly Humana).

• Walmart’s Healthcare Future – These clinics are still only a test and it is way too early to tell whether more locations are coming, but Walmart’s focus on bringing its value proposition into healthcare seems quite certain at this point. Given Walmart’s massive presence, funding, and focus on customer value, traditional healthcare players should keep an eye on Walmart as a potential competitor, partner, and/or customer going forward.

Algorithmic Audit

With the growing volume of radiology AI tools coming out each month and rising calls for “auditable AI” from the industry (especially in the absence of explainable AI), an Indian team introduced a new Algorithmic Audit framework for working with developers to test their AI tools’ clinical utility and risks.

• Framework – The framework outlines how to perform independent validation using data that the algorithm “has not seen before,” curating datasets for these algorithmic audits, examining false positives and false negatives (focused on the implications of these errors), and real-world deployment and testing.

• Reaction – The framework was welcomed by some in the radiology AI community, who suggested that such audits should be adopted across the ecosystem (for regulatory approval, monitoring, developer use).

---

The Wire

• The American Hospital Association published new research suggesting that hospital M&A helps reduce healthcare costs (2.3% lower hospital operating expenses, 3.5% lower per-patient billings) and improve care (reduced readmissions and mortality). Safe to say, not everyone agreed with this hospital lobby-funded research, as others in the industry quickly pointed out the numerous existing studies with the opposite findings (like this study and this study).

• Despite the buzz around the ultrasound’s emerging segments, IHS Markit reported that traditional ultrasound equipment (cardiology, OB/GYN, and radiology) continued to dominate in 2018, when it controlled 70% of global ultrasound shipments and revenue. The traditional ultrasound segment’s 6.6% revenue growth in 2018 to $4.9 billion (out of $7b total) also outpaced all other segments. However, non-traditional US segments continued to grow in 2018, with particularly strong years for breast ultrasound, POCUS (shipments up 7.9%), and primary care ultrasound (shipments up 14.2%).

• Bombay’s high court found two hospital staff members negligent in a fatal MRI room accident that occurred when a family member of a patient was sucked into an MRI machine after they let him enter the MRI room while carrying the patient’s oxygen tank.

• French researchers found that ultrasound is 1.7 times more effective than X-ray in diagnosing erosive rheumatoid arthritis (RA), adding to US’ cost and radiation advantages. The study assessed the joint erosion of 168 patients (32 w/ early RA, 90 w/ late RA, 46 w/ osteoarthritis) with ultrasound diagnosing rheumatoid arthritis patients with at least two worn joints with greater sensitivity (68% vs. 34.4%) and specificity (100% vs. 89.1%) than X-ray.

• Analysis of Stryker’s recent acquisition of Mobius Imaging and its GYS/Cardan Robotics subsidiary is calling this more than one of the company’s standard tuck-in acquisitions, suggesting that the GYS/Cardan piece would bolster Stryker’s spinal robotics business and may be more significant than Mobius’ CT business.

• A team of radiology leaders (Sogania, Allen, Dreyer, McGinty) shared a positive vision for radiology’s AI-integrated future, while reinforcing the need to create a radiology AI ecosystem and outlining how it would help clinical practices. The article argued that 1) AI’s efficiency and information advantages will help drive radiology’s value-based transition; 2) The importance of an interconnected AI ecosystem that involves all parties (providers, developers, regulators, etc.); and 3) How an AI ecosystem would mitigate future clinical AI implementation challenges (data availability, legal, interpretability).

• A coalition of cardiology and nuclear imaging groups (ACC, ACNM, ASNC, CAA and SNMMI) are taking a stand against proposed changes to the 2020 Medicare Physician Fee Schedule that might reduce myocardial PET reimbursements by as much as 80%. The groups will push for CMS to reconsider its decision to assume 90% utilization for PET Cameras and pricing for other equipment.

• A team of Swedish researchers found that DBT detects cancers with a similar biologic profile as digital mammography, suggesting that DBT would detect the same types of cancers as DM (but still maintaining its range of screening advantages). Using data from Sweden’s Malmö trial (n=14,848 women, 139 cancers, 118 invasive), the team found that of the 37 invasive cancers detected by DBT and 81 detected by DM there were no major differences in tumor characteristics (size, node-negative status, luminal A–like subtype).

• medPhoton’s ImagingRing CBCT imaging system (2D & 3D, 102cm bore, ceiling rail mounted) received FDA clearance and is now available in the U.S. configured with Mevion’s S250 Proton Therapy System for precision radiation therapy treatments. The ImagingRing CBCT and Mevion S250 were first integrated in the Netherlands earlier this year and the configuration is now available to all current and future Mevion clients in the U.S.

• New research from Harvard reveals that payers in Massachusetts could have saved significantly by steering patients to low-cost providers (~12.8% savings) or establishing a cost ceiling below the 76th percentile (9% savings). The study analyzed payments for 291 usually outpatient services, identifying significant unexplainable cost variations (e.g. medical services were 76% higher at acute hospitals) and the widest variations for ambulance/transportation services, physical/occupational therapy, and laboratory/pathology testing.

• Life Image launched its new Real World Imaging (RWI) solution, providing researchers with ‘living’ data sets to support life science studies and train AI algorithms. RWI’s heterogeneous and continuously growing ‘living’ data sets include millions of images and hundreds of thousands of associated reports and studies categorized by anatomical parts (e.g. head, lung, etc.), potentially streamlining the difficult task of accessing, de-identifying, and normalizing imaging data at-scale for use as Real World Evidence.

---

The Resource Wire

– This is sponsored content.

• POCUS Systems’ forthcoming ultrasounds will combine ease of use, durability, and reliability, allowing clinicians to focus on their patients.

• Profound Medical announced yet another Focused Ultrasound Foundation-related advancement with the FDA clearance of its TULSA-PRO transurethral ultrasound device for the MRI-guided ablation of prostate tissues.

• This Carestream case study compares images of foot trauma captured using the OnSight 3D Extremity System to images captured on 2D X-rays.

• Qure.ai got solution-oriented in its latest blog, analyzing how acquisition artifacts are responsible for AI performance degradation, and introducing two methods to solve this problem.

• Nuance’s Karen Holzberger and a team of AI and clinical leaders will take the C-MIMI 2019 stage on September 22 to discuss Driving AI Adoption in Clinical Practice.

• Did you know that imaging patients are most likely to no-show for their procedures on Mondays and Saturdays? By partnering with Medmo, imaging centers can keep their schedules full, despite the inevitable Monday no-shows.

---