Google Health all but invited the blowback when its AI developer-researchers suggested their breast-cancer model may be superior to radiologists’ eyes and generalizable across differing demographics.

That was in January. The team is still championing the implications of its work.

This month around 25 researchers from around the world dinged the Google Health study hard, laying bare its shortcomings regarding methodological transparency and scientific reproducibility.

Both papers ran in Nature. And now Team Google is back in that publication, responding to the response.

“Led by Dr. Scott McKinney, Google Health did not mince words,” as Shelly Fan, PhD, describes in a showdown synopsis running in SingularityHub. “Their general argument: ‘No doubt the commenters are motivated by protecting future patients as much as scientific principle. We share that sentiment.’”

However, under current regulatory frameworks, Fan writes, paraphrasing Google Health’s gist, “our hands are tied when it comes to open sharing.”

Many healthcare AI watchers will find Fan’s coverage helpful, as all three papers are behind Nature’s paywall. For her full take on the kerfuffle, click here.

Last fall CMS whittled a field of 300-something entrant teams in its AI Health Outcomes Challenge to 25 semifinalists. This week the agency revealed seven finalist entities, one of which will claim the grand prize of up to $1 million next spring.

The runner-up will fare well too. The payout for second place will be as much as $230,000.

In an Oct. 29 announcement, CMS says this final stage will push those still standing to further refine their already impressive algorithms.

Just getting to the finals puts $60,000 in an entrant entity’s pocket. Here’s the field:

• Ann Arbor Algorithms (Sterling Heights, MI)
• ClosedLoop.ai (Austin, TX)
• Deloitte Consulting, LLP (Arlington, VA)
• Geisinger (Danville, PA)
• Jefferson Health (Philadelphia, PA)
• Mathematica Policy Research, Inc. (Princeton, NJ)
• University of Virginia Health System (Charlottesville, VA)

The top finishers from among these will excel at demonstrating the utility of AI tools for predicting unplanned admissions and adverse events at hospitals and skilled nursing facilities.

In addition, the winner will have done the best job of developing predictive algorithms for a standard target to be selected by CMS, as the agency puts it in the announcement.

Commenting on the announcement, CMS Administrator Seema Verma says that President Trump “recognizes that the complexity and growth of data in healthcare holds the potential to transform healthcare, and AI is the key to manage and analyze that data.”

The competition launched in March of 2019. Project leaders named the 25 finalists last November.

For an updated challenge overview from CMS, click here.

A little more than a year after inking a 10-year pact on digital healthcare development, Mayo Clinic and Google Health have launched their first major research project together. Interestingly, it directly involves AI.

The two announced Oct. 28 they’ll be building AI and machine learning tools to optimize radiation treatment plans for cancer patients.

Specifically, Mayo medical clinical experts and medical physicists will work with Google healthcare AI specialists to train and test an algorithm for automatically distinguishing tumor margins from adjacent healthy tissue.

The fast precision will allow radiation oncologists to confidently fine-tune frequency and dosage planning of radiation therapy for optimal outcomes.

The initial phase of this initial project will focus on cancers of the head and neck. In these areas, contouring is especially challenging because many delicate anatomic structures are squeezed into tight spaces.

“If organs are not properly identified, the radiation plan may not protect these critical structures or adequately treat the tumor,” Chris Beltran, PhD, tells Mayo Clinic News Network. Beltran, one of the lead investigators on the project, chairs the division of medical physics at Mayo Clinic Florida.

Because radiation treatment planning is one of the most labor-intensive and data-rich disciplines in medicine, Beltran adds, radiation oncology is “ripe if not overdue for application of AI-augmented methods, particularly deep-learning-based approaches.”

Cían Hughes, MB, ChB, a Google Health informaticist and research scientist, comments that radiation oncologists today “painstakingly draw lines around sensitive organs like eyes, salivary glands and the spinal cord to make sure radiation beams avoid these areas.”

The standard protocol works well, he adds, but it can take a “really long time”—six hours or more, in cases—to get it precisely right.

Hughes notes AI’s potential to streamline parts of the contouring workflow, which should get patients to radiation therapy sooner and finished with it faster, improving the overall experience.

What’s more, remarks Nadia Laack, MD, chair of the radiation oncology at Mayo Clinic in Rochester and a principal investigator on the project, applying AI and machine learning “may also help to alleviate radiation oncology workforce shortages throughout the world and decrease the variability in the quality of care patients receive.”

Fifteen of 47 healthcare providers and payers considered Jvion as a potential supplier of AI software over the past year and a half. Of the 15, eight either made the buy or were leaning hard toward it when they were asked.

In baseball terms—this is a World Series week, after all—Jvion batted a cool .533.

By comparison, Epic had fewer looks but closed the sale on all of them, hitting a perfect 1.000 by going 12 for 12.

The findings are from market research conducted by KLAS in the 18 months prior to mid-September and published in an October report.

The report’s subhead encapsulates its upshot: “Investment Continues but Results Slower Than Expected.”

The benefits of integrating AI into the healthcare ecosystem “are undeniable, and investment from healthcare organizations is increasing at a fast pace,” the authors observe in the introduction. “Yet few organizations have settled on any one AI vendor as their go-forward choice—most use a hodgepodge of solutions to meet their needs.”

Here’s how the other vendors fared on the purchase-decision scorecard, ranked by considerations (i.e., at bats): 

10—DataRobot (5 hits, good for .500)

9—Health Catalyst (4 hits, .444)

8—IBM (0 hits, .000)

8—KenSci (5 hits, .625)

6—ClosedLoop.ai (5 hits, .833)

5—Amazon (1 hit, .200)

5—Cerner (3 hits, .600)

4—Medial EarlySign (3 hits, .750)

4—Microsoft (2 hits, .500)

Meanwhile the top reasons for choosing any given vendor were expertise (named by 19% of respondents), partnership (13%), functionality (12%) and technology (10%). Named by less than 10% were integration, prior experience, price, references and reputation.

As for who’s leading the purchase decisions around AI for healthcare, CIOs and other technology O’s lead the way in 34% of the surveyed settings. Clinical leadership is next (28%), followed by data science leadership (21%), CXO (21%) and analytics leadership (17%).

KLAS’s crib notes for reporters synopsize some themes that emerged in the vendor comparisons, including interview answers offered by the respondents:

• Jvion has high market visibility and purchase consideration, struggles to deliver; Epic growing fast, with embedded workflows an advantage
• KenSci client satisfaction remains high; Health Catalyst and DataRobot see decreases
• Cross-industry AI giants perceived as average overall—Microsoft seen as the strongest, while IBM trails

One of the interview respondents may have spoken for many healthcare AI shoppers when he or she observed: “Nobody has a great offering yet.”

For more on the KLAS report looking at the healthcare AI market, click here.

The novel coronavirus has shown a nasty penchant for targeting the kidneys, and physicians can’t always tell which patients will need dialysis until they do. By then it’s often too late to save a life.

A new machine learning model has proven accurate at making the call early enough to allow for advance planning, preparation and patient routing.

Lili Chan, MD, and colleagues at the Icahn School of Medicine at Mount Sinai described their work developing and testing the algorithm at a virtual national meeting of the American Society of Nephrology, which concluded over the weekend.

The team trained their model on data from more than 3,000 patients who were both hospitalized and COVID-positive. The researchers only included information gathered withing 48 hours of admission, challenging the AI to predict which patients with acute kidney injury would require dialysis.

In the test phase, the model delivered high accuracy (AUC of 0.79). The results showed the most valuable at-admission predictors to be blood levels of creatinine and potassium, age, and vital signs of heart rate and oxygen saturation.

“A machine learning model using admission features had good performance for prediction of dialysis need,” Chan says in a news release. “Models like this are potentially useful for resource allocation and planning during future COVID-19 surges. We are in the process of deploying this model into our healthcare systems to help clinicians better care for their patients.”

In another recent research project, Chan and colleagues found that, of nearly 4,000 COVID patients hospitalized in New York City, some 46% had acute kidney injury. Among these, 19% needed dialysis, and half of them died in the hospital.  

*The list also includes virtual reality, augmented reality and 3D printing. Links lead to announcements as sent by the companies.

1. Ezra: AI assistance for radiologists interpreting prostate MRI

2. Aidoc: Algorithm for triaging patients with incidental pulmonary embolism

3. ImmersiveTouch: Virtual reality for planning oral and maxillofacial surgery

4. 3D Systems: New 3D printing materials for planning maxillofacial surgery

A hearing aid manufacturer founded in the 1950s has introduced what it says is the first use of AI to optimize auditory assistance for the hearing impaired in real time.

In announcing the application, Denmark-based Widex describes its system as capable of pinpointing user preferences for sound settings using only 12 or so comparisons. Without the algorithms, the company says, the process would take more than 2 million tests.

“When applied, the settings create a personalized hearing experience based on context, content and intent,” Widex claims. “Users can store the settings as programs in their smartphones and activate them throughout the day, such as when they’re at work, at the supermarket or in their kitchen.”

Further, these programs can be securely stored in the cloud to help “enhance the hearing experience of other users.”

As it happens, the Widex product release coincides with the publication of a paper in Ear and Hearing in which researchers at Facebook Reality Labs suggest augmented reality platforms used with hearing aids “may offer ideal affordances to compensate for hearing loss.”

Such AR platforms, Lunner et al. write, could also help medical science better understand the role of hearing in everyday life. The paper is available in full for free.

The FDA has greenlit a European company’s AI software package designed to streamline workflow for radiologists reading prostate MRIs.

Netherlands-based Quantib announced the U.S. go-ahead Wednesday, remarking that this is the sixth product in its portfolio to earn an FDA nod.

The company says the prostate offering comes with tools for lifting quality in radiology reports and efficiency in related processes.

Quantib also notes that new guidelines took effect in numerous countries this year, widely standardizing prostate MRI for patients suspected of having prostate cancer.

“Triple digit growth in MRI prostate scans is the result, but there are not enough expert radiologists to report all these extra scans,” the company comments, adding that prostate cancer represents one in four of all male cancer cases.

Three machine learning algorithms have identified patients likely to suffer extreme pain following surgery with about 80% accuracy each. The predictive assistance may help physicians appropriately prescribe alternate pain-management plans over dangerous and addictive opioids.

The research was presented at the 2020 annual meeting of the American Society of Anesthesiologists, held virtually Oct. 2 to 5.

Lead study author Mieke Soens, MD, of Harvard told attendees his team plans to integrate the models with the EHR at Brigham and Women’s Hospital to “provide a prediction of post-surgical pain for each patient.”

To build their models, Soens and colleagues reviewed data from almost 6,000 postsurgical patients spanning a variety of procedure categories. They found some 22% of these patients had been given high doses of morphine milligram equivalent in the first 24 hours after their operation.

Next they consulted pain-care experts and searched the literature to come up with 163 factors potentially predictive of severe postsurgical pain.  

Armed with these insights, the team constructed three models—logistical regression, random forest and artificial neural networks—capable of poring through the patients’ medical records and pruning the 163 factors to only the most strongly predictive.

Comparing the models’ predictions with actual opioid use in the same patients, Soens and colleagues found all three had around 80% accuracy at pinpointing which patients suffered the greatest pain and needed the higher doses of opioids.

“Electronic medical records are a valuable and underused source of patient data and can be employed effectively to enhance patients’ lives,” Soens says in prepared remarks issued after the presentation. “Selectively identifying patients who typically need high doses of opioids after surgery is important to help reduce opioid misuse.”

A Silicon Valley maker of surgical robots is launching a venture capital arm to invest in future leaders of minimally invasive care. Inaugural ante: $100 million.

Intuitive Surgical—the outfit best known for its da Vinci system—says Intuitive Ventures will concentrate on accelerating R&D for digital tools, precision diagnostics, focal therapeutics and platform technologies.

Its monies will also support independent innovators working to broaden the field of minimally invasive care.

“The future of minimally invasive care spans the patient journey from early diagnosis to treatment and beyond,” Julian Nikolchev, president of the new investment firm, says in an announcement.

Oliver Keown, MD, joined Intuitive to co-found the fund last year and is now serving as its director.  

“We are value-add investors who leverage access to Intuitive’s unique industry expertise and customer connections,” says Keown, who previously worked as an investor at GE Ventures. “Our nimble structure and alignment with the startups we will back empower us to invest early and support our portfolio companies as they pioneer markets.’’