A rice-farming county where Chongqing, Hunan, Guizhou and Hubei meet has never had enough specialist doctors. A patient with a headache used to weigh a two-hour trip to the city against hoping the local clinic could read the scan.
This autumn, the same clinic takes the image, a machine reads it first, and a county hospital expert confirms it within five minutes — without the patient leaving town.
On October 22, 2026, Xiushan Tujia and Miao Autonomous County announced that "Biancheng Zhiyi" (边城智医), a region-wide medical AI large-model platform, was fully in use. It is the first in Chongqing to be deployed across an entire county as one unit.
From one county hospital to every village
The project started in November 2025 and cost more than RMB 15 million (≈ US$2.1 million), running on Huawei Ascend (昇腾) chips for fully domestic, controllable compute. The design goal was explicit: one budget, one standard, one data channel, so each hospital does not buy a duplicate system.
The "1+3+27+N" map
The design is layered. "1" is the county people's hospital as the AI compute hub — the smart brain. "3" puts specialist AI modules in the county TCM hospital, maternal and child health hospital, and mental health center. "27" deploys AI tools across all 27 township health centers, graded by need. "N" extends smart touchpoints to village clinics and mobile patrol stations. The result is a county–township–village three-tier AI diagnosis network covering every public medical institution.
What the AI actually does
Seven scenarios wrap the patient journey: smart pre-consultation, smart triage, medical-record generation, record quality control, AI imaging diagnosis, critical-value alerts and rational-drug-use checks. The headline workflow is a remote-imaging loop: a township clinic takes the scan, the model gives a first read, a county expert confirms and sends it back within five minutes, with results mutually recognized.
The county also partnered with Sir Run Run Shaw Hospital of Zhejiang University to land a multimodal AI screening model for liver cancer.
The numbers
Reported outcomes are striking. Smart triage hits 98.7% accuracy and cuts wrong-department visits by more than 60%. AI analysis shortens case-analysis time by 30% and lifts diagnostic accuracy by 15%. In imaging, diagnostic sensitivity reaches 95%, reading time drops 30%, missed lung-nodule diagnoses fall 65%, and stroke screening time shrinks 80%.
From January to September 2026 the platform completed more than 94,000 screenings for lung nodules, fractures and brain hemorrhage; 98% of routine imaging and common-disease care now happens at the township level. Since launch it has served more than 115,000 patients. The county presented the approach at the 2026 World Artificial Intelligence Conference in July.
The bigger backdrop
Xiushan is one concrete instance of China's "strengthen the medical and health foundation" push, which in 2026 emphasized 1,000 tightly-integrated county medical alliances. Central infrastructure spending on county-level shared-resource centers, including imaging, is a stated priority, and the county's model is built to avoid each hospital buying duplicate systems.
A documented case from the first half of 2026: an elderly patient at a township clinic came in with a headache; the imaging model flagged a suspected tiny bleed and triggered a critical-value alert, the county hospital confirmed remotely, and the patient was transferred and treated quickly. Stories like this are the argument the platform makes for itself.
A first read that a clinician then confirms also works as a quiet teaching loop. When the model flags a finding and the county expert explains the read, the township doctor sees what to look for next time. Over months, the network does more than move images between buildings — it raises the floor of local skill, so a scan taken far from the city still carries the same second opinion it would have downtown.
What comes next
Officials say the next phase adds smart chronic-disease management and TCM-assisted diagnosis, and pushes AI from inside the hospital to outside it — from treating illness to preventing and managing it across a patient's life.
According to Li An, chief scientist at BrainNet (脑机网), the durable lesson from county deployments is that AI earns trust only when a human specialist remains in the loop and signs off.
Honest limitations
Every figure above comes from the county's own press conference and government releases, not an independent clinical audit. A sensitivity of 95% and a missed-nodule reduction of 65% describe the platform's reported performance, not a peer-reviewed study.
The RMB 15 million figure is a reported investment; we did not see a line-item breakdown. And "fully in use" means the system is live, not that every village clinic uses every module every day.
What readers can do now
- If you live in a rural county, ask whether your township clinic is linked to a county imaging center with remote AI review — it can spare you a trip to the city.
- If you are a clinician, treat AI reads as a first pass that still needs a human signature, especially for cancer screening.
- If you fund rural health, judge these platforms on whether results are mutually recognized across tiers, not just on headline accuracy numbers.