Iris's Nodoca AI device transforms throat exams into 10-second diagnostic tool

AI-powered throat imaging reduces discomfort and time-to-diagnosis

Iris, a Japanese medtech startup led by former emergency physician Sho Okuyama, has deployed Nodoca—an artificial intelligence system that diagnoses influenza from pharyngeal images in just over 10 seconds. According to Wired AI, the device has been installed at more than 2,000 medical institutions across Japan since becoming the country’s first AI-equipped medical device to receive regulatory approval in 2022, and it gained COVID-19 diagnostic approval in October 2025. Unlike conventional nasopharyngeal swab tests, which cause discomfort and require insertion deep into the nasal cavity, Nodoca uses a compact camera to capture throat images and analyzes them with AI, enabling earlier diagnosis after symptom onset.

Hardware-first strategy and dataset moat

The foundation of Nodoca’s competitive advantage lies not in the AI alone, but in the integration of proprietary sensing hardware and years of accumulated training data. Okuyama emphasizes that “the real differentiating factor lies in sensing—how the data is acquired,” according to Wired AI. When Iris was founded in 2017, no labeled dataset of pharyngeal images existed for AI training. To address this, the company distributed dedicated cameras to approximately 100 medical institutions over three years, collecting thousands of patient consent-validated images. This painstaking data-collection phase solved three major challenges: hardware development, training-data assembly, and clinical validation of AI-assisted diagnosis.

Once Nodoca entered clinical practice, the dataset grew exponentially. Wired AI reports that the company’s throat-image repository expanded from hundreds of thousands of images before launch to several million images today. Each clinical use contributes anonymized, processed data, creating a self-reinforcing cycle—the more the system is used, the more it improves. This network effect has effectively prevented competitors from entering the same market segment, since building a comparable dataset now requires years of infrastructure investment.

Why This Matters

Nodoca represents a shift in how medtech startups should think about moat-building: hardware and data integration, not software alone. For clinicians and hospital administrators evaluating AI diagnostic tools, this case shows that regulatory approval (Japan’s “new medical device” designation) combined with real-world deployment at scale creates defensible value. For competing vendors, the 4-year head start in data collection for throat imaging is now a near-insurmountable barrier.

The broader implication is that point-of-care AI devices—those that combine custom sensors, immediate feedback, and minimal discomfort—may see faster adoption than lab-dependent tests, especially in undersupported regions. Okuyama’s background practicing on remote islands with minimal equipment illuminates why: when infrastructure is sparse, devices that replace invasive procedures have outsized clinical value. As Nodoca’s COVID-19 expansion shows, the platform can be extended to new conditions without rebuilding the hardware or the underlying dataset strategy, making incremental regulatory approvals faster and cheaper than greenfield product launches.