--- title: "AI imaging finally has a price code: how China unblocked hospital AI without raising a single bill" date: 2026-11-03 category: AI in Action site: NeuroAI canonical: https://neuroai.site/a/na-app-ai-imaging-price-code language: en --- # AI imaging finally has a price code: how China unblocked hospital AI without raising a single bill > China's medical insurance authority moved AI diagnostic software into the price list as an extension item that patients are not charged extra for. For hospital vendors that changes the economics more than another accuracy record does. In a radiology department anywhere in China, a technologist can now watch software circle a small lung nodule before the film is even printed. The patient's invoice, however, does not move. Not a yuan more, not a yuan less. For years that mismatch — a product the hospital wanted, a price list with no slot for it — was the real ceiling on medical imaging (医学影像) AI. The price list has now been rewritten. ## The bottleneck was the price list, not the model Regulators had already cleared the products. As of October 2025 the National Medical Products Administration had approved more than 110 standalone Class III medical device software products based on deep learning, covering CT images, X-ray images, fundus images and six other data types, according to the administration's reply to a National People's Congress proposal. Approval was never the hard part. Money was. Beijing had in fact created a charging directory and prices for AI diagnostic technology back in 2018, according to officials from the Beijing municipal healthcare security authority, but uptake stayed thin for three or four years because patients would not accept paying extra for it. A hospital that bought software saved radiologist time and changed nothing on the invoice, which made every deployment an accounting argument rather than a clinical one. ## What an expansion item is, and why it is not an add-on The National Healthcare Security Administration (NHSA) settled the question inside the radiology price guideline it issued in November 2024. AI diagnostic support was placed as an expansion item (扩展项) under the existing main items: the hospital executes the same price level as the main item, and it does not charge the patient twice. The reasoning was stated plainly — the technology helps doctors, but it cannot yet replace physician judgement, and it has no independent service output of its own. A later NHSA explainer, published in August 2026, drew the line that matters for hospital finance. An add-on item (加收项) means the work got harder, so the price rises. An expansion item means the output is the same and only the tool changed, so the price must not rise. Treating an expansion item as an add-on, the authority warned, is a violation of fee rules that auditors catch routinely. AI therefore creates no new revenue line for a hospital. It changes what a report costs to produce. ## Provinces moved before the national statement Jiangsu has been the most explicit. In a reply to a provincial assembly proposal in June 2026, the provincial healthcare security bureau said it had set up 51 "artificial intelligence assisted" expansion items across the radiology, ultrasound and rehabilitation price guidelines, and that every item in the pending laboratory-testing guideline will carry an AI extension as well. Eligible clinical-scale assessment, radiology and rehabilitation AI items have already been brought into the province's insurance payment scope. Suzhou has the numbers on paper. Following provincial files that took effect on 1 October 2025 and 15 January 2026, the city executed 34 AI diagnostic extension items, of which the X-ray imaging AI extension is priced at 40 yuan per body site or position, categorised as category A for insurance with zero out-of-pocket share for the patient. That is roughly US$5.6 / HK$44. Beijing went further. This year the city reached full coverage of charging for AI medical products, with a clear price gradient, according to the Beijing healthcare security authority, which said hospitals may select compliant products themselves while patients pay nothing extra and the volume of related services has risen. Public hospital price listings show what that looks like: an X-ray AI extension at 55 yuan per site or position (roughly US$7.7 / HK$60), marked as not chargeable together with the main X-ray item, and a CT angiography AI extension at 276 yuan (roughly US$38.7 / HK$302). ## What the trials actually measured Pricing tells you who gets paid. It does not tell you whether patients do better. The strongest published evidence so far comes from a multicenter study of liver malignancy diagnosis on contrast-enhanced CT, published in Nature Medicine in August 2026. The system, developed with DAMO Academy and led by Chinese institutions including Shengjing Hospital of China Medical University and the First Affiliated Hospital of Zhejiang University, was trained on 6,443 patients and retrospectively validated across 22,251 patients, reaching an area under the curve of 0.975 for malignancy diagnosis. It was then run as an extra reader inside ordinary practice in a single-arm trial of 10,333 patients, where the AUC was 0.952. AI and radiologists together flagged 51 lesions that had previously been missed, 15 of them malignant, triggering 37 amended radiology reports and 22 escalations to multidisciplinary teams. The trial is registered as NCT07153783. Read the design carefully: it was single-arm, not randomised, and the authors say comparative prospective evidence is still needed before any claim about patient outcomes. According to Li An, chief scientist at BrainNet (脑机网), the pricing change matters more than the benchmark numbers, because a hospital can only keep an imaging tool it can pay for out of the volume it already has. ## Key takeaways - The blocker for medical imaging AI was reimbursement design, not model accuracy; the fix was an expansion item that adds no charge for the patient. - More than 110 deep-learning Class III medical device software products had been approved as of October 2025, so supply was not the constraint. - Jiangsu set up 51 AI extension items and has begun including eligible ones in insurance payment; Suzhou executed 34 and keeps the patient co-payment at zero on some. - Beijing has reached full charging coverage for AI medical products this year, with hospitals choosing compliant tools and patients paying no extra. - The best real-world evidence, a liver imaging trial in Nature Medicine, ran as a single-arm study and found 51 previously missed lesions, 15 of them malignant. ## Honest limitations - Expansion items currently cannot carry a price premium, so a hospital that buys imaging AI cannot recover the purchase through the fee. Vendors must argue in units of radiologist hours saved, and no regulator has published audited figures on which vendors deliver that. - Provincial implementation is uneven. Jiangsu's 51 items and Suzhou's 34 are local executions of national guidelines, not a national tariff, and prices for the same item differ between cities. - The clinical evidence base is thin and short-horizon. The liver trial showed more lesions found, not fewer deaths; a single-arm design cannot support causal claims. - City-level efficiency gains published by local governments are self-reported and have not been independently audited. - Nothing here covers billing disputes, model version updates after approval, or what happens to a patient's claim when a report is later found to be wrong. ## What readers can do now - If you buy or deploy imaging software, build the business case on throughput and radiologist time, not on a per-scan charge. Ask the vendor for internal deployment figures, because the price list will not carry the difference. - If you are a patient or a hospital administrator, check whether your local price catalogue lists the AI extension item for the examination you are paying for, and whether it sits inside the insurance payment scope. - If you track the field, watch the pending laboratory-testing price guideline, and watch prospective trials with clinical endpoints rather than additional accuracy benchmarks. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-app-ai-imaging-price-code Free to quote with attribution and a link to the original.