Want to let an AI-powered doctor loose on patients? Try slapping a food-label-like sticker on it, says Uncle Sam

Action plan to regulate machine-learning software in 2021 revealed

The US Food and Drug Administration wants manufacturers to specifically label AI-powered medical devices to help patients understand the technology, and to test them on real-world data to see how they perform in the wild beyond clinical environments.

The watchdog has published a five-point action plan on regulating standalone devices that deploy machine-learning algorithms. AI in healthcare is a massive market projected to reach a whopping $6.6bn in 2021 in America.

Companies hoping to get a slice of the pie will have to secure FDA approval if they want to get their AI-based products onto the market. The process is tricky for the regulator to formulate, however, as it has to grapple with how quickly the technology is evolving.

“This action plan outlines the FDA’s next steps towards furthering oversight for AI/ML-based SaMD,” said Bakul Patel, director of FDA’s Digital Health Center of Excellence in the Center for Devices and Radiological Health, responsible for approving premarket medical devices. That's Software as a Medical Device – SaMD – in case you were wondering.

“The plan outlines a holistic approach based on total product lifecycle oversight to further the enormous potential that these technologies have to improve patient care while delivering safe and effective software functionality that improves the quality of care that patients receive. To stay current and address patient safety and improve access to these promising technologies, we anticipate that this action plan will continue to evolve over time.”


Top doctors slam Google for not backing up incredible claims of super-human cancer-spotting AI


First, the action plan outlines the need to update the current regulatory framework to scrutinize AI software that changes over time. Algorithms are improved and updated by training on new data, and it’s important to make sure that despite these changes, devices remain safe for medical use.

Manufacturers will have to provide a “predetermined change control plan,” a document that outlines what is being altered exactly, and explains how the algorithm will learn those changes. The updated regulatory framework is expected to be detailed in a separate report to be published later this year.

The second point in the plan tackles what the regulator calls “Good Machine Learning Practice, or GMLP, for short.” It has to establish a series of standards to guide companies developing and manufacturing AI medical devices.

To ensure uniformity, the FDA said it was partnering with other communities that are working on similar benchmarking problems, such as the British Standards Institution and the Institute of Electrical and Electronics Engineers (IEEE).

Biases in machine-learning software are especially important in healthcare, where the quality of care can vary across people’s ages, genders, wealth, and ethnicity. If devices are trained on historical data that reflects these biases, the software can be problematic and perform worse on different groups of people, the FDA warned.

“The Agency recognizes the crucial importance for medical devices to be well suited for a racially and ethnically diverse intended patient population and the need for improved methodologies for the identification and improvement of machine learning algorithms,” the action plan said.

It is working closely with academics to devise algorithms capable of identifying biases and even potentially eliminating them to examine future software.

Who reads the label?

If food carries labels listing ingredients and nutritional value, why not medical devices? The FDA wants manufacturers to provide clear labels to describe what goes in and comes out of the algorithm.


Study: AI designed to detect diabetic eye disease blinks in the real world, makes more work for doctors


The labels should state the data used to train the software, what the device measures or takes in as input to give a result, and its overall performance in terms of accuracy. Greater transparency, it argued, will mean patients and users can trust the technology and are more likely to adopt it. The FDA will hold a public workshop later this year to see how to implement such labels.

Finally, the agency is asking businesses to enroll in a pilot study that focuses on testing AI medical devices with real-world data. The study will be voluntary, however, and it requires members of the public willing to be guinea pigs.

“Evaluations performed as part of these efforts could be used to determine thresholds and performance evaluations for the metrics most critical to the RWP of AI/ML-based SaMD, including those that could be used to proactively respond to safety and/or usability concerns, and for eliciting feedback from end users,” the watchdog stated.

“We acknowledge that AI/ML-based SaMD is a rapidly progressing field, and we anticipate that this Action Plan will continue to evolve as we pursue these activities and seek to provide additional clarity in this space,” the FDA concluded. A spokesperson for the agency was not available for immediate further comment. ®

Other stories you might like

  • Assange can go to UK Supreme Court (again) to fend off US extradition bid

    Top Brit judges may consider whether an American prison is just too much

    Julian Assange has won a technical victory in his ongoing battle against extradition from the UK to the United States, buying him a few more months in the relative safety of Her Majesty's Prison Belmarsh.

    Today at London's High Court, the Lord Chief Justice of England and Wales Lord Burnett approved a question on a technical point of law, having refused Assange immediate permission to appeal to the UK Supreme Court. The WikiLeaker's lawyers had asked for formal permission to pose this legal conundrum about Assange's likely treatment in US prisons to the Supreme Court:

    Continue reading
  • They see us Cinnamon Rolling, they're rating: GeckoLinux incorporates kernel 5.16 with familiar installation experience

    A nice, clean community distro that works well

    Most distros haven't got to 5.15 yet, but openSUSE's downstream project GeckoLinux boasts 5.16 of the Linux kernel and the latest Cinnamon desktop environment.

    Some of the big-name distros have lots of downstream projects. Debian has been around for decades so has umpteen, including Ubuntu, which has dozens of its own, including Linux Mint, which is arguably more popular a desktop than its parent. Some have only a few, such as Fedora. As far as we know, openSUSE has just the one – GeckoLinux.

    The SUSE-sponsored community distro has two main editions, the stable Leap, which has a slow-moving release cycle synched with the commercial SUSE Linux Enterprise; and Tumbleweed, its rolling-release distro, which gets substantial updates pretty much every day. GeckoLinux does its own editions of both: its remix of Leap is called "GeckoLinux Static", and its remix of Tumbleweed is called "GeckoLinux Rolling".

    Continue reading
  • Running Windows 10? Microsoft is preparing to fire up the update engines

    Winter Windows Is Coming

    It's coming. Microsoft is preparing to start shoveling the latest version of Windows 10 down the throats of refuseniks still clinging to older incarnations.

    The Windows Update team gave the heads-up through its Twitter orifice last week. Windows 10 2004 was already on its last gasp, have had support terminated in December. 20H2, on the other hand, should be good to go until May this year.

    Continue reading
  • Throw away your Ethernet cables* because MediaTek says Wi-Fi 7 will replace them

    *Don't do this

    MediaTek claims to have given the world's first live demo of Wi-Fi 7, and said that the upcoming wireless technology will be able to challenge wired Ethernet for high-bandwidth applications, once available.

    The fabless Taiwanese chip firm said it is currently showcasing two Wi-Fi 7 demos to key customers and industry collaborators, in order to demonstrate the technology's super-fast speeds and low latency transmission.

    Based on the IEEE 802.11be standard, the draft version of which was published last year, Wi-Fi 7 is expected to provide speeds several times faster than Wi-Fi 6 kit, offering connections of at least 30Gbps and possibly up to 40Gbps.

    Continue reading
  • Windows box won't boot? SystemRescue 9 may help

    An ISO image you can burn or drop onto a USB key

    The latest version of an old friend of the jobbing support bod has delivered a new kernel to help with fixing Microsoft's finest.

    It used to be called the System Rescue CD, but who uses CDs any more? Enter SystemRescue, an ISO image that you can burn, or just drop onto your Ventoy USB key, and which may help you to fix a borked Windows box. Or a borked Linux box, come to that.

    SystemRescue 9 includes Linux kernel 5.15 and a minimal Xfce 4.16 desktop (which isn't loaded by default). There is a modest selection of GUI tools: Firefox, VNC and RDP clients and servers, and various connectivity tools – SSH, FTP, IRC. There's also some security-related stuff such as Yubikey setup, KeePass, token management, and so on. The main course is a bunch of the usual Linux tools for partitioning, formatting, copying, and imaging disks. You can check SMART status, mount LVM volumes, rsync files, and other handy stuff.

    Continue reading

Biting the hand that feeds IT © 1998–2022