How a Japanese AI Device Is Turning the Throat Exam Into a Faster, Less Awkward Test

A Japanese medical device called Nodoca is using artificial intelligence to diagnose influenza from throat images in a little over 10 seconds, reducing reliance on the uncomfortable nasal swab and helping doctors make faster decisions during routine exams. The system has already been approved in Japan, covered by national health insurance, and deployed at more than 2,000 medical institutions, with a new Covid-19 function added in 2025.

What began as an attempt to modernize one of medicine’s oldest routines has become a case study in how AI may reshape frontline care. Nodoca’s developer, Iris, argues that the biggest breakthrough is not just the algorithm, but the way data are captured: with a compact camera device that photographs the pharynx and pairs those images with a short patient interview.

In an era when many medical AI products are still limited to pilot programs or software demonstrations, Nodoca stands out for reaching clinical use at scale. Its success also highlights a broader shift in healthcare technology: the move from purely digital prediction tools toward systems that combine hardware, imaging and clinical workflow to make diagnosis easier, quicker and less unpleasant for patients.

What is Nodoca, and why does it matter?

Nodoca is an AI-assisted diagnostic device designed to assess influenza by analyzing images of the throat, or pharynx, alongside information gathered during a medical interview. The system matters because it offers an alternative to the standard nasal swab test, which can be uncomfortable and is often disliked by patients.

In practical terms, the device aims to answer a common clinical question faster and with less friction. It can also be used earlier after symptoms begin, giving doctors another option when traditional sampling may be inconvenient or poorly tolerated.

How the device works

The device uses a small camera to capture images inside the throat. An AI model then evaluates those images together with other patient data and returns an influenza assessment in just over 10 seconds.

That speed is important in busy outpatient settings, where clinicians often need to decide quickly whether to test, treat, isolate or send a patient home. Iris says the system is meant to support those decisions by making the exam easier to perform and the result faster to obtain.

• Images are collected from the pharynx using a compact device
• AI analyzes the visuals along with interview data
• The system produces an influenza assessment in seconds
• It reduces the need for a deep nasal swab

Why the throat became the focus

The throat exam may look ordinary, but it contains a large amount of clinical information. Different infections leave different patterns in the tissues, and physicians have long relied on visual inspection as part of the basic physical exam.

According to Iris founder Sho Okuyama, the real innovation is not the diagnosis layer alone. He says the differentiating factor in medical AI lies in data acquisition — in other words, how the information is sensed before software does anything with it.

Okuyama has said medical AI is often discussed as if the diagnosis itself were the main breakthrough, but he believes the real advantage comes from building the hardware and collecting the right kind of clinical data in the first place.

That view shaped Nodoca’s development from the start. Instead of adapting an existing imaging product, Iris built its own capture device around a clinical problem that had not previously been addressed with AI.

How Iris built a dataset from scratch

When Iris was founded in 2017, there was no substantial training dataset for AI focused on throat imagery. That created a major obstacle: the company could not simply train a model on an existing archive and launch a product.

To solve that problem, Okuyama and his team lent specialized cameras to about 100 medical institutions and collected data over roughly three years, with patient consent. That effort gave the company a foundation for training and validation that did not exist before.

The process tackled three separate uncertainties at once: whether the hardware could be built, whether enough usable data could be gathered, and whether AI-supported diagnosis could work reliably in the real world. Iris says overcoming those hurdles opened the door to practical deployment.

What changed after approval in Japan?

Nodoca became the first AI-equipped medical device in Japan to be approved as a new medical device and reimbursed through the national health insurance system. That combination of regulatory approval and insurance coverage is crucial, because it makes adoption more realistic for clinicians and institutions that must weigh cost, workflow and patient acceptance.

Since then, the device has been introduced at more than 2,000 medical institutions across Japan. That scale suggests it has moved beyond novelty and into routine clinical use, at least for one specific diagnostic application.

In October 2025, Iris added another approved function tied to Covid-19, broadening the system’s role within respiratory illness assessment. The company’s progress indicates that once the imaging platform is in place, it may be possible to extend its use beyond influenza alone.

Why reimbursement matters

Approval alone does not guarantee adoption. In many healthcare systems, tools remain underused when there is no clear reimbursement path.

By securing national insurance coverage, Iris made Nodoca easier for providers to justify financially. That may be one reason the system spread to thousands of facilities rather than remaining confined to research centers or a few high-profile hospitals.

Who is Sho Okuyama?

Sho Okuyama is a former emergency physician whose clinical background helped shape Iris’s approach to product design. He also worked in medicine on remote islands, where equipment was extremely limited and doctors had to rely heavily on observation and bedside judgment.

That experience appears to have influenced his thinking about diagnostic technology. Rather than imagining AI as a replacement for clinicians, he has focused on tools that support front-line assessment and reduce the burden of routine tasks.

Okuyama has described working on islands with only basic equipment, saying physicians often depended on what they could see and hear in order to treat patients.

That kind of practical experience is visible in Nodoca’s design philosophy: it does not attempt to reinvent medicine from a distance, but to improve one of the most common touchpoints between patient and doctor.

Why the system could become more accurate over time

Nodoca is designed to improve as more clinicians use it. After commercialization, Iris says the system’s accuracy began to rise more quickly because each real-world use generates anonymized processed data that can be fed back into the system.

Before launch, the company says it had already accumulated hundreds of thousands of throat images. That number has since grown into the millions, creating a much larger and more diverse dataset than the one available during development.

This matters because medical AI systems often become stronger when they are exposed to broader clinical conditions. More data can improve model performance, help the software handle edge cases and make the product more defensible against competitors.

A data network effect

The growing database may also create a competitive moat. If a device gets better with every clinical use, new entrants face a difficult challenge: they must not only match the current product, but also replicate the accumulated experience embedded in the training pipeline.

Iris says that no serious latecomers have yet emerged in the same niche, suggesting the combination of hardware, data and regulatory approval has created a difficult market to enter.

How could throat imaging expand beyond flu?

It could eventually become a broader screening tool for multiple diseases. Iris is researching whether AI can detect signs of lifestyle-related illnesses such as diabetes and hypertension by analyzing changes in the throat’s mucosal and vascular patterns.

The idea is based on the premise that the throat may reveal more systemic information than doctors have traditionally used. If validated, a single image could help screen for several conditions, not just respiratory infection.

Okuyama has suggested that within about a decade, a single throat photograph could support a much wider range of tests. He argues that because AI inference costs are very low, adding more analyses may not meaningfully raise the price of the service.

That vision, if realized, could turn the throat exam into a high-yield diagnostic gateway rather than a narrow infection check.

Why Iris is talking about redesigning health care

Okuyama sees Nodoca as one part of a larger effort to rethink the healthcare journey itself. In his view, AI could support care at multiple stages, from online medical interviews and triage to in-person visits with primary care doctors and onward referrals to specialist hospitals.

In other words, the company is not only trying to replace one test; it is trying to reduce fragmentation across the system. That includes proposing regulatory changes and working with government ministries to make AI-enabled workflows easier to deploy.

Okuyama has said the “future” he talks about is not about waiting for an entirely new invention, but about putting already available technology into everyday medical practice.

That framing is important. Many healthcare innovations fail not because they are technically impossible, but because regulation, reimbursement and clinical habits lag behind the technology. Iris is essentially arguing that the missing ingredient is implementation.

What makes Nodoca different from other medical AI tools?

Nodoca differs from many medical AI products in that it pairs software with custom hardware and clinical workflow redesign. Instead of analyzing data that already exists in electronic records, it creates a new kind of input at the point of care.

This approach may be one reason it has moved beyond demonstration status.

1. It solves a specific and common problem: flu testing in routine care.
2. It reduces patient discomfort compared with nasal swabs.
3. It creates its own proprietary data source through dedicated hardware.
4. It has regulatory approval and national insurance coverage in Japan.
5. It is already being used in thousands of medical institutions.

Could it work outside Japan?

Possibly, though not automatically. Okuyama believes the biological features being analyzed are consistent enough that the Japanese data and algorithms should transfer to other countries.

Even so, international rollout would likely require local validation, regulatory review and perhaps new clinical trials. Iris is already preparing for overseas expansion, and the United States is being considered as a future market.

Global expansion would be a major test of whether the system’s performance is robust across different patient populations and healthcare settings. It would also reveal whether the company’s model depends on uniquely Japanese clinical workflows or can be adapted more broadly.

Why this story matters for the future of medical AI

Nodoca offers a glimpse of where practical AI in medicine may be headed. The most successful tools may not be the ones that simply promise to make expert decisions faster. They may be the ones that change how the data are gathered, how the exam is experienced and how the result fits into a real clinical workflow.

For patients, that could mean less discomfort and faster answers. For doctors, it could mean more efficient care and earlier decision-making. For healthcare systems, it could mean new ways to scale screening without dramatically raising costs.

At the same time, the story is a reminder that medical AI is not just a software challenge. It is also a hardware, data, reimbursement and regulatory challenge — and Iris’s success suggests that companies willing to solve all of those pieces may be best positioned to bring AI into everyday medicine.

Timeline of Nodoca’s development

The bottom line

Nodoca is trying to modernize a basic physical exam that has changed little for centuries. By combining throat imaging, AI analysis and a purpose-built device, Iris has created a flu-testing system that is faster, less invasive and already embedded in real clinics.

Its longer-term ambition is even bigger: to make the throat a gateway for multiple diagnoses and to help redesign the way healthcare is delivered. Whether that vision scales internationally will depend on regulation, validation and clinician adoption, but in Japan, the company has already shown that medical AI can move from theory to everyday practice.