Nodoca: How Japan's AI Throat Camera Diagnoses Influenza in 10 Seconds

A Japanese medical technology startup has developed an AI-powered throat camera that can assess influenza in little more than 10 seconds—without the discomfort of a nasal swab. The device, called Nodoca, was created by Aillis Inc. (also reported as Iris in some media), founded by former emergency physician Sho Okiyama in 2017. Approved by Japan's Pharmaceuticals and Medical Devices Agency (PMDA) in February 2022 and covered by national health insurance since December 2022, Nodoca is now installed in over 2,000 clinics nationwide. By photographing the pharynx and using deep learning to detect influenza-specific patterns invisible to the human eye, the system is redefining how doctors screen for infectious diseases—and its creators believe the same technology could one day detect diabetes, hypertension, and even COVID-19 from a single throat image. This article explains how Nodoca works, the science behind it, its regulatory journey, and what it means for the future of AI-driven healthcare.

The Problem with Traditional Influenza Testing

Nasal Swabs: Uncomfortable and Time-Consuming

Conventional influenza diagnosis relies on rapid antigen tests that require inserting a swab deep into the nasal cavity to collect a mucosal specimen. The process is uncomfortable, often triggering gag reflexes or nosebleeds, and results typically take 10 to 20 minutes to process. For healthcare systems already under strain—particularly during Japan's annual flu season and potential "twindemics" with COVID-19—this bottleneck creates significant delays in patient flow.

Moreover, rapid antigen tests have a narrow detection window. Their sensitivity is highest 12–48 hours after fever onset, meaning early-stage infections can yield false negatives. The tests also require trained personnel to collect samples correctly, adding labor costs and infection risk during outbreaks.

The Hidden Information in Your Throat

The human pharynx contains a wealth of diagnostic information that traditional examinations often miss. When influenza infects the respiratory tract, it triggers visible changes in the throat—most notably the appearance of lymphoid follicles on the posterior pharyngeal wall. These clusters of immune cells are difficult for even experienced physicians to detect consistently, yet they represent one of the earliest and most specific signs of influenza infection.

Dr. Masaki Okamoto, an AI medicine expert at Harvard Medical School, described Nodoca's approach as "reconstructing the eye of a skilled doctor using AI technology"—a system that captures subtle pharyngeal patterns that human observation might overlook, especially in busy primary care settings.

How Nodoca Works: AI Meets Pharyngeal Imaging

The Hardware: A Compact Pharyngeal Camera

Nodoca consists of a dedicated camera device mounted on a wand-like instrument, similar to a tongue depressor with an imaging sensor at the tip. A medical professional inserts the camera into the patient's oral cavity to photograph the pharynx. The procedure takes only seconds and, unlike nasal swabs, does not require penetrating the nasal passage.

The system is available in two configurations:

• Console System: A standalone unit with a camera, stand, console, and accessories for on-site analysis.
• Cloud System: Images and data are transmitted to a cloud-based server for AI analysis, with results displayed on a tablet or PC.

The AI Analysis Pipeline

Once the pharyngeal image is captured, Nodoca's software performs a multi-step analysis:

1. Image Acquisition: The camera captures high-resolution images of the posterior pharyngeal wall, tonsils, and lymph tissue.
2. Data Integration: The system combines the image with a 26-item medical questionnaire completed by the patient or clinician, covering symptoms, body temperature, onset time, and vaccination history.
3. Deep Learning Analysis: A convolutional neural network (CNN) trained on over 500,000 pharyngeal images analyzes the photograph for influenza-specific patterns—primarily the presence, distribution, and morphology of lymph follicles.
4. Result Output: Within approximately 10 seconds, the system displays whether "characteristic pharyngeal signs or symptoms of influenza infection" are "Detected" or "Not detected."

The AI model was developed through a clinical research study involving 708 patients across 11 medical institutions in Japan. The training data was carefully curated to avoid overemphasis on PCR results, ensuring the model learned genuine visual patterns rather than statistical correlations.

Clinical Performance: How Accurate Is Nodoca?

According to PMDA review documents, Nodoca demonstrated the following performance in clinical validation:

Metric	Nodoca (Lower 95% CI)	Conventional Rapid Antigen Test
Sensitivity	70.7%	86.7%
Specificity	85.5%	Comparable
Result Time	~10 seconds	10–15 minutes
Patient Discomfort	Minimal (oral camera)	Moderate to high (nasal swab)
Early Detection	High (even within 6 hours of onset)	Lower (viral load dependent)

Notably, Nodoca's sensitivity is highest in the very early phase of illness (within 6 hours of fever onset), when rapid antigen tests often yield false negatives due to low viral loads. This complementary performance profile makes Nodoca particularly valuable for initial screening.

Important Limitations

PMDA and Aillis have been transparent about Nodoca's boundaries:

• Nodoca is an aid to diagnosis, not a definitive test. It does not directly detect viral proteins or genetic material.
• Physicians must make final diagnoses by integrating Nodoca results with clinical symptoms and other examinations.
• The system cannot distinguish between Influenza A and B subtypes.
• As lymph follicles heal over time, sensitivity decreases with illness duration—opposite to antigen tests, which improve as viral load increases.

From Startup to National Coverage: Nodoca's Regulatory Journey

Founding and Data Collection (2017–2020)

Sho Okiyama, a graduate of the University of Tokyo's Faculty of Medicine and practicing emergency physician, founded Aillis in 2017 after witnessing patients who might have been saved with faster, more accessible diagnostics. At the time, no training dataset existed for AI-based pharyngeal analysis—a significant hurdle since deep learning models require massive labeled datasets.

To solve this, Okiyama lent dedicated Nodoca cameras to approximately 100 medical institutions across Japan. Over three years, clinicians captured pharyngeal images with patient consent, building the foundational dataset that would train the AI. This grassroots data collection effort was essential, as publicly available pharyngeal image databases did not exist.

PMDA Approval: Japan's First AI Pharyngeal Device

On February 15, 2022, PMDA approved Nodoca as a "new medical device" under Japan's Pharmaceutical and Medical Device Act. The approval made Nodoca the first AI-equipped medical device in Japan specifically designed for pharyngeal imaging and influenza screening. The regulatory review emphasized that while Nodoca's sensitivity did not achieve non-inferiority to immunochromatography in head-to-head comparisons, its unique value lay in providing rapid, minimally invasive screening that complements rather than replaces existing tests.

National Health Insurance Coverage

In December 2022, the Central Social Insurance Medical Council granted Nodoca a C2 reimbursement classification—a category reserved for novel medical technologies with new functions. This marked the first time an AI-based medical device received C2 coverage in Japan. Under the reimbursement scheme, Nodoca testing costs 305 points (approximately ¥915 with 30% patient copay), nearly identical to conventional rapid antigen tests at 301 points (~¥903).

This pricing parity was crucial for adoption. Clinicians could adopt Nodoca without increasing patient costs or administrative burden, removing a major barrier to AI integration in primary care.

Beyond Influenza: The Future of Throat-Based AI Diagnostics

Detecting Lifestyle Diseases from the Pharynx

Aillis is actively researching whether pharyngeal images can reveal signs of diabetes and hypertension. The theory is based on observable changes in mucosal patterns and blood vessel morphology that correlate with systemic vascular and metabolic conditions. During internal discussions, the engineering team has even explored whether throat videos could estimate vital signs such as body temperature and heart rate.

"If we can extract vital data from throat videos," Okiyama has stated, "it could open up an entirely new approach to screening." The company's CTO, Jun Fukuda, noted that as the AI model improved, it began detecting patterns the team never explicitly trained it for—including the ability to infer gender and identify hypertension from throat images.

COVID-19 and Respiratory Infections

While Nodoca is currently approved only for influenza, the underlying technology is theoretically adaptable to other respiratory pathogens. The company has discussed expanding detection to COVID-19 and other infectious diseases that present with pharyngeal lymphoid changes. The key challenge is that different pathogens produce subtly different pharyngeal patterns, requiring additional training data and clinical validation.

Overseas Expansion and US Clinical Trials

Aillis is preparing for international expansion, with clinical trials in the United States under consideration. The company believes the data and algorithms accumulated in Japan are transferable to other populations, though regulatory pathways will require independent validation. Okiyama has also been actively lobbying for regulatory reform in Japan to accelerate approval timelines for AI medical devices, arguing that the current system—while rigorous—can delay life-saving innovations.

Real-World Impact: 2,000 Clinics and Counting

Adoption in Primary Care

As of late 2025, Nodoca has been installed in more than 2,000 clinics nationwide—an impressive penetration for a device launched just three years prior. Clinics report several operational benefits:

• Nurses can perform imaging while physicians review results, freeing doctor time for complex cases.
• Patient throughput increases during flu season, when clinics face surging demand.
• Patient satisfaction improves due to the painless, non-invasive procedure.
• Early screening identifies high-risk patients for confirmatory PCR testing, optimizing resource allocation.

Patient Experience

One patient quoted in Japanese media summarized the shift: "It used to involve swabbing the mouth, which made people gag. Now the nurse just takes a quick look with the camera—no scraping, no pain." Adverse events in clinical trials were minimal, with retching reported in only 1.7% of patients—and all cases resolved quickly without intervention.

Aillis' Engineering Talent

Aillis has recruited top-tier AI talent to advance its platform, including engineers recognized by Google as elite AI specialists. CTO Fukuda credits this talent density for the company's rapid progress: "At first, I tried to build the model myself, but it didn't go well. Then Yoshihara joined us, and within three weeks he identified the core problem. In just three months, he solved issues I couldn't fix in half a year." The company now employs approximately 100 people and maintains a team of five world-class AI engineers.

Nodoca in the Context of Japan's AI Healthcare Revolution

Nodoca is not an isolated innovation. Japan has emerged as a global leader in regulatory frameworks for AI medical devices, with 172 program medical devices holding valid marketing authorization as of September 2025, of which 51 carry an AI-utilization designation. These span imaging, endoscopy, cardiology, and infectious disease applications.

What makes Nodoca's regulatory-reimbursement pathway particularly significant is its completeness: prospective clinical validation → PMDA authorization → national insurance coverage. This sequence serves as a reference model for subsequent AI diagnostic tools in Japan and potentially worldwide.

However, challenges remain. AI systems for infectious diseases require more frequent recalibration than stable-disease models because circulating strains, vaccine components, and population immunity change seasonally. Aillis must ensure Nodoca's algorithm stays accurate across evolving influenza variants—a task that requires ongoing data collection and model updates.

Frequently Asked Questions (FAQ)

• What is Nodoca?
  • Nodoca is an AI-powered medical device developed by Japanese startup Aillis Inc. that photographs the pharynx and uses deep learning to detect characteristic signs of influenza infection in approximately 10 seconds. It is approved by Japan's PMDA and covered by national health insurance.
• How does Nodoca work?
  • Nodoca uses a compact camera-equipped wand to capture images of the patient's pharynx. The images are combined with a 26-item medical questionnaire and analyzed by AI trained on over 500,000 pharyngeal images. The system detects influenza-specific patterns—primarily lymphoid follicles—and returns a "Detected" or "Not detected" result in about 10 seconds.
• Is Nodoca a definitive diagnostic test for influenza?
  • No. Nodoca is classified as an "aid to diagnosis." It does not directly detect viral proteins or genetic material. Physicians must make final diagnoses by integrating Nodoca results with clinical symptoms and other examinations. It also cannot distinguish between Influenza A and B.
• How accurate is Nodoca compared to rapid antigen tests?
  • In clinical validation, Nodoca achieved a sensitivity of 70.7% and specificity of 85.5%. While its sensitivity is lower than rapid antigen tests (86.7%), Nodoca performs better in the very early phase of illness (within 6 hours of onset) and delivers results in ~10 seconds versus 10–15 minutes for antigen tests.
• Who founded Aillis and why?
  • Aillis was founded in 2017 by Sho Okiyama, a University of Tokyo medical graduate and practicing emergency physician. He started the company to strengthen diagnostic capabilities through AI after witnessing cases where faster diagnostics might have saved patients' lives.
• What diseases could Nodoca detect in the future?
  • Aillis is researching whether the same throat imaging technology can detect lifestyle-related diseases such as diabetes and hypertension by identifying changes in mucosal and blood vessel patterns. The company is also exploring COVID-19 detection and vital sign estimation (body temperature, heart rate) from pharyngeal videos.
• How much does a Nodoca test cost?
  • In Japan, Nodoca testing is covered by national health insurance at 305 reimbursement points—approximately ¥915 (about $6) with standard 30% patient copay. This is nearly identical to the cost of conventional rapid antigen tests.
• Is Nodoca available outside Japan?
  • Currently, Nodoca is only approved and available in Japan. Aillis is preparing for overseas expansion and considering clinical trials in the United States, but no international launch timeline has been confirmed.