A Camera Pointed at Your Throat Could Replace the Dreaded Nose Swab
A Japanese AI device diagnoses influenza by photographing the back of your throat in ten seconds flat, no swab required. It's already in use at more than 2,000 clinics, and its maker thinks the technology could eventually screen for diabetes and high blood pressure the same way.
Key points
- Iris, a Japanese medical company, received national health insurance approval in Japan for its AI throat-scanning device, Nodoca, in 2022.
- Nodoca delivers an influenza assessment in just over ten seconds by analysing a photograph of the back of the throat, replacing the nasal swab most patients dread.
- More than 2,000 medical clinics across Japan had adopted the device as of 2025.
- In October 2025, Nodoca received Japanese regulatory approval to assess Covid-19 as well as influenza.
- The company is exploring clinical trials in the United States and believes its AI models will transfer internationally because human mucous membranes are biologically consistent across populations.
Anyone who has had a flu test knows the moment: a nurse slides a long cotton swab deep into your nostril, you flinch, your eyes water, and the result takes fifteen minutes. A Japanese startup wants to end that routine entirely.
The company is called Iris. Its device, Nodoca, works like this: a doctor holds a small handheld camera up to a patient's open mouth, snaps an image of the pharynx (the back of the throat), and feeds that image into an AI model alongside basic information from the patient's initial questionnaire. Within roughly ten seconds, the system returns an influenza assessment. No swab. No waiting room purgatory.
How does photographing your throat tell a doctor anything?
The patterns of redness, swelling, and blood vessel changes visible on the throat's lining shift in specific ways depending on which pathogen is causing an infection. Iris founder Sho Okuyama, a former emergency physician who once worked on remote Japanese islands with little more than a stethoscope, noticed that experienced doctors were reading these patterns intuitively. He wanted to make that reading automatic.
Building the system from scratch was slow. When Iris was founded in 2017, no database of labelled throat images existed anywhere. Okuyama lent dedicated cameras to around 100 clinics and spent three years collecting images with patient consent. That initial dataset ran to hundreds of thousands of images. Today, with Nodoca in routine clinical use, the figure has grown to several million. Each anonymised image makes the model a little sharper, which in turn makes Nodoca harder for a rival to match without starting that same multi-year data collection process all over again.
What does this mean for patients right now?
If you visit one of the 2,000-plus Japanese clinics that have adopted Nodoca, you skip the swab and get a result in the time it takes to pull on your coat. The device can also be used earlier in an illness, before viral loads in the nose are high enough for a conventional swab to catch. As first reported by Wired AI's Japanese edition, Japanese regulators extended the device's approval in October 2025 to cover Covid-19 assessment as well.
For patients outside Japan, the practical benefit is not here yet. Iris is weighing clinical trials in the United States, which would be required before any American regulatory submission.
| Milestone | Date |
|---|---|
| Iris founded | 2017 |
| Training data collection completed | approx. 2020 |
| Japanese national health insurance approval | 2022 |
| Clinics using Nodoca in Japan | 2,000+ (2025) |
| Covid-19 function approved in Japan | October 2025 |
Could a throat photo ever screen for diabetes?
Okuyama thinks so, and Iris is actively researching it. The argument is that lifestyle diseases like diabetes and high blood pressure quietly change the tiny blood vessels visible in the throat lining. If an AI model can learn to spot those changes reliably, a single photograph taken during a routine visit could flag a patient for follow-up blood tests before they ever report symptoms.
"The cost of AI inference is extremely low," Okuyama told the outlet, "so even if we increase the number of tests, the cost of providing them hardly changes." That economic logic matters: adding a diabetes screen to a flu visit would cost the clinic almost nothing in compute costs once the model exists.
The technology required, Okuyama insists, is not futuristic. Every component already exists. The gap is regulatory clarity and the political will to close it.
