US Defense Department enlists Google for AI-powered cancer-spotting kit

Google has contributed software to an AI-powered microscope the US Department of Defense hopes will help pathologists spot cancerous cells in tissue samples much more quickly.

The device, named the Augmented Reality Microscope (ARM), is in many regards a conventional ‘scope – users place samples on a glass tray and peer at them through an eyepiece.

But the ARM also employs computer vision algorithms that guides medicos to focus on potentially problematic areas, and creates heatmaps labelling cells as benign or cancerous. Those images can be displayed on a monitor to allow more detailed study.

Researchers are already using ARMs to detect breast cancer, cervical cancer, prostate cancer and cell duplication using four different AI algorithms, CNBC first reported.

The DoD's Defense Innovation Unit (DIU) is investing in the technology to aid pathologists working at military hospitals that are often short-staffed. The Unit believes that the technology can speed the diagnosis process and provide experts with a second opinion when diagnosing diseases.

The Unit has awarded contracts to Jenoptik, an optics company to build the hardware. Google develops the software for ARMs.

In all, 13 prototypes have been built so far; one of them is at the MITRE lab near Washington DC, one of the US government's advanced science and technology R&D research groups.

A team of researchers have tested the device's abilities to detect breast cancer that has metastasized in lymph nodes. Results from early experiments published in the Journal of Pathology Informatics last year showed that it was able to classify cancerous and benign cells with an accuracy level of about 94 percent.

Lead author David Jin, who was a senior AI engineer at MITRE and is now the deputy director for AI assessment at the DOD's Chief Digital and Artificial Intelligence Office, said the technology is promising but not yet ready for real clinical use.

"Something like this has an extreme potential for benefit, but also there's a lot of risks," he commented. The AI models have to be accurate, and mistakes or errors could lead pathologists astray if they rely too heavily on the software. ARM is meant to augment their expertise, helping analyze tissue samples more quickly by highlighting potentially diseased cells.

• It's official – Google AI gives you cancer ...diagnosis in real time: Neural net can spot breast, prostate tumors
• Google Health licenses its AI breast cancer screening tool to a medical biz
• US task force to look into how military could use generative AI

The AI-powered microscopes also provide experts with an easier way to digitize images and minimizes data storage requirements, helping hospitals to collect and study data without having to bulk up their infrastructure too much. It was trained on private, encrypted, data from the DIU.

Officials have reportedly negotiated contracts with Google and Jenoptik to provide ARMs to the US military as early as this Autumn. Devices may be sold to hospitals and clinics for $90,000 to $100,000 apiece in the future.

Google has been working on ARMs for years: in 2019 the ads and search giant published a paper detailing a microscope connected to a camera system running AI algorithms on an old GPU. Researchers at the time said ARMs could be as a tool to help train pathologists and could be cheaper than conventional whole-slide scanners.

The Register has asked Google and the DIU for more comment. ®