NeuroAI NEUROAINEUROAI.SITE
China AI, reported clearly
1,300+ verified stories on models, robotics, BCI and the business of AI.
Browse stories
ESC

China's medical-imaging AI earned its license to be billed, not just demoed

A 2024 pricing rule and a wave of 2025 regulatory clearances turned Chinese medical-imaging AI from conference demos into certified, deployed and billable clinical tools — including one product now on the WHO's tuberculosis-screening list.

2026-11-10 · 1145 words · NeuroAI
China's medical-imaging AI earned its license to be billed, not just demoed

A radiologist in a county hospital reads a chest X-ray in under a minute. A red box appears where the eye might have passed over a shadow. The report goes out the same morning, signed by a person, quietly sharpened by software.

For a decade, Chinese medical-imaging AI lived at trade shows and in pilot wards. In 2025 it crossed into the consequential world of certificates, procurement and billing — the difference between a promising tool and a deployed one.

Key takeaways

  • In November 2024 China's National Healthcare Security Administration (国家医保局) wrote an "AI-assisted diagnosis" (人工智能辅助诊断) extension item into its radiology pricing guideline, letting hospitals use AI without charging patients twice.
  • United Imaging (联影医疗) reported first-half 2025 revenue of RMB 6.02 billion (≈ US$835 million / HK$6.5 billion) and more than 20 FDA-cleared AI-enabled devices, plus a radiotherapy auto-contouring system that won China's Class III certificate in April 2025.
  • Infervision (推想医疗) took 2025 Class III certificates for lung-surgery planning software and an X-ray fracture-triage tool; its InferRead DR Chest is one of only six computer-aided detection products the WHO says meets its TB-screening performance standard.
  • Airdoc (鹰瞳科技) won a Class III certificate in April 2025 for its Airdoc-AIFUNDUS 2.0 fundus system covering diabetic retinopathy and retinal vein occlusion; its retina-AI business recorded 488,000 tests in 2025, up 21.6% year on year.
  • The common thread is certification: Class III registration with the NMPA (国家药监局) is what lets these tools sit inside the clinical workflow rather than beside it.

The rule that made AI ordinary

On 20 November 2024, the National Healthcare Security Administration issued the Radiology Services Price-Item Guideline (Trial) (《放射检查类医疗服务价格项目立项指南(试行)》). Buried inside it is a single extension item: hospitals that use AI to assist diagnosis charge the same price as the underlying exam and do not bill for the AI separately. The stated logic is that AI improves throughput rather than adding a discrete service, so it should not become a new cost on the patient.

That is a quiet but decisive shift. For years hospitals hesitated to embed imaging AI because there was no clean way to pay for it and no national blessing that it belonged in the workflow. The guideline answers both: use it, don't double-charge, and let the efficiency case carry it.

United Imaging: AI baked into the machine

United Imaging is the clearest example of "AI as a property of the hardware." The Shanghai company, founded in 2011, reported first-half 2025 revenue of RMB 6.02 billion (≈ US$835 million / HK$6.5 billion) and R&D of RMB 1.14 billion (≈ US$158 million / HK$1.24 billion), about 19% of revenue, with more than 9,700 patent applications on file.

Two 2025 milestones stand out. In April, its uIPW Intelligent Physician Workspace — a radiotherapy auto-contouring system — received China's Class III certificate, described by the company as the first domestic system able to auto-contour multiple targets and organs. In September, the uDR Aurora CX digital radiography system was cleared by the US FDA, using embedded uAI vision cameras and an automation platform that positions the patient and flags image-quality problems without a repeat exposure. The pattern is consistent: AI is not a bolt-on app but part of how the scanner behaves.

Infervision: from lung nodules to a WHO listing

Infervision, founded in 2016, built its name on reading CT and X-ray images for lung and chest conditions. In 2025 it added two Chinese Class III certificates: lung-surgery planning software on 27 March, and an X-ray fracture-triage tool on 29 September (registration number 国械注准20253211962). The company says its products are approved across NMPA, FDA, the EU's CE mark, the UK's UKCA and Japan's PMDA.

The fact that travels furthest is the WHO one. In a 2025 policy statement, the World Health Organization listed six computer-aided detection products that meet its performance standard for tuberculosis screening in adults aged 15 and over; Infervision's InferRead DR Chest is on that list. That matters because TB screening in resource-stretched regions depends on reading chest X-rays fast and consistently, and a WHO stamp is what national programmes buy against. Infervision says the product has been deployed for TB screening in more than ten countries.

Airdoc: the eye as a chronic-disease window

Airdoc took a different door — the retina. Its Airdoc-AIFUNDUS 2.0 fundus diagnostic system won China's Class III certificate on 10 April 2025, extending coverage from diabetic retinopathy to retinal vein occlusion. In its 2025 results filed in Hong Kong, the retina-AI business reported revenue of RMB 120.8 million (≈ US$16.8 million / HK$131 million), up 7.2%, with 488,000 tests performed, up 21.6%, across 8,599 active service outlets including 478 hospitals and 408 physical-examination centres.

The deployment logic is screening at scale: a camera, a non-mydriatic shot of the back of the eye, and a model that flags who needs a doctor. Because diabetic eye disease is silent until it isn't, the value is catching it early where no ophthalmologist is on site.

According to Li An, chief scientist at BrainNet (脑机网), the moment an imaging-AI tool becomes billable is the moment hospitals stop treating it as a science project and start demanding that it never be wrong.

Honest limitations

The headline regulatory facts are solid — the NHSA guideline is a published government document, and the NMPA Class III certificates and FDA clearances are recorded approvals from the agencies themselves. What is softer is the meaning of "deployed": a certificate lets a product be sold and used, but does not say how many scans actually run through the model daily or how often clinicians override it.

The financial figures are company-reported. United Imaging's revenue and R&D are from its own interim results; Airdoc's test counts and revenue are from its Hong Kong-listed annual filing. Neither is an independent audit of clinical impact. The "more than 20 FDA-cleared AI-enabled devices" line is United Imaging's own wording and aggregates across product lines rather than naming each.

This piece concentrates on three publicly listed or well-documented players. It does not cover the longer tail of smaller imaging-AI vendors, nor does it assess diagnostic accuracy head-to-head. A clearance is a permission to be used, not a measurement of how well.

What readers can do now

  1. If you are a clinician or hospital buyer, treat Class III registration and FDA/CE clearance as the minimum filter, then ask vendors for local deployment volume and override rates — the questions certificates don't answer.
  2. If you are a patient in China, know that AI support on your X-ray, CT or fundus photo is now a normal, separately-not-priced part of the exam under the 2024 radiology guideline, so you should not be charged extra for it.
  3. If you follow the sector, watch procurement and reimbursement, not launch events. The 2025 story is that these tools became certifiable and billable; the next one is whether hospitals actually buy and use them at scale.

Related coverage

Cite this article

NeuroAI (脑机网), "China's medical-imaging AI earned its license to be billed, not just demoed," 10 November 2026, https://neuroai.site/a/na-app-ai-medical-imaging

Machine-readable: BibTeX · RIS · JSON · Markdown

More in “AI in Action” → · Back to home · Markdown version

China AI, reported clearly
1,300+ verified stories on models, robotics, BCI and the business of AI.
Browse stories