Medical AI deepens roots across hospitals

Artificial intelligence is moving deeper into the day-to-day operations of hospitals in China as more medical institutions develop their own large language models and embed AI to improve the quality and efficiency of healthcare services.

At East Hospital Affiliated to Tongji University in Shanghai, physician Duan Tao now works with what he describes as a "third participant" in the consulting room.

As Duan speaks with patients, Med-Go — a medical LLM spearheaded by doctors at the hospital — can retrieve medical histories, identify relevant information and offer suggestions for complicated cases.

For Duan, who heads the hospital's obstetrics and gynecology department as well as its medical AI innovation center, the bigger opportunity lies in developing AI agents for specific clinical tasks rather than simply building another general-purpose model.

Zhang Haitao, director of the hospital's emergency and critical care department and a leading developer of Med-Go, said the model's defining feature is that it was "developed by doctors for doctors".

"By handling preliminary information gathering and reference checks, Med-Go is designed to free doctors to spend more time communicating with patients and making clinical judgments," he said. The system is intended to support specialists in handling complex cases while also helping grassroots doctors strengthen their diagnostic capabilities.

The hospital is developing more than 40 such agents for tasks ranging from outpatient consultation and medical-record review to research support.

Med-Go has also expanded outside the hospital's walls, having been deployed across 46 community health service centers in Shanghai's Pudong New Area and at more than 30 major hospitals elsewhere nationwide, according to the hospital team and local government information.

A similar push is underway in other leading hospitals. In Chengdu, Sichuan province, West China Hospital of Sichuan University has developed its own Huaxi Hongyi medical AI model, integrating more than 10 general models and over 50 specialized models. Its medical-documentation agent generated more than 1.1 million medical records in 2025, saving more than 300,000 hours of doctors' paperwork time, the hospital said.

Such applications show how hospital-developed medical AI is moving from model testing into real clinical workflows.

Market research firm IDC estimates China's AI-powered healthcare application software market reached 3.54 billion yuan ($528 million) in 2025 and could expand to 14 billion yuan by 2030, as hospitals increasingly adopt AI agents for more complex operational and clinical tasks.

Policy support is reinforcing that shift. National guidelines issued in November call for widespread use of AI-assisted primary care, specialist decision support and patient services by 2027, with intelligent assistance at grassroots medical institutions expected to achieve near-universal coverage by 2030. Shanghai has separately set a target for "AI plus healthcare" application scenarios to cover all medical institutions by 2030.

Nevertheless, transitioning AI from demonstration projects into routine clinical use remains challenging at times.

At Shanghai's East Hospital, some clinicians initially hesitated to use Med-Go because system responses could take 20 to 40 seconds. Adoption also varied by department, depending partly on whether department heads embraced the technology. More broadly, hospital-grade AI systems require substantial computing resources, integration with existing information systems and continuous clinical validation.

Data fragmentation also poses a major obstacle to wider deployment.

Zhang Qi, associate dean of the School of Artificial Intelligence and Data Science at the University of International Business and Economics, said in an interview with China Daily that the lack of unified data standards, privacy concerns, and the high cost of data labeling and storage are among the biggest constraints for medical AI.

She said hospitals hold large troves of valuable clinical data, but concerns over patient privacy and data security can make institutions reluctant to share them. She expects general foundation models and specialized medical models to coexist, making common interfaces and communication protocols increasingly important if different systems are to exchange data and work together.

Overcoming those barriers could determine how far AI spreads beyond leading urban hospitals.

Zhang of UIBE said AI-enabled remote-care platforms and specialist AI assistants could become one of the most practical applications over the next three to five years, helping grassroots institutions improve diagnostic capabilities while giving patients greater access to expertise from leading hospitals.

But wider adoption will also raise pressing questions over safety and accountability.

She said faster innovation should be accompanied by clear requirements for safety validation and ethical review, as well as continued monitoring after AI systems enter clinical use. National health authorities have similarly stressed that AI should support and complement — rather than replace — medical professionals.

For now, when Med-Go and a physician reach different conclusions, the doctor retains the final say. Duan said the aim at this stage is not to remove doctors from the decision-making chain, pointing out that medical errors existed even before the arrival of AI.

As the technology becomes more capable, however, he expects that balance to change. Human involvement currently accounts for nearly the entire medical decision-making process, but could eventually fall below 10 percent in some workflows, Duan said, with doctors concentrating on critical decisions and oversight.

lijing2009@chinadaily.com.cn