CAS Hong Kong Center Releases CARES 4.0 Clinical Agent at CREATE Symposium

CARES 4.0 is built on the Harness agent framework and uses self-developed multimodal medical foundation models covering CT, MRI, ultrasound, endoscopy and EEG. Unlike single-point medical AI models, the system is designed to work inside hospital clinical workflows as a physician’s intelligent partner across diagnosis and treatment, department management and medical research. In clinical use, it can assist the full diagnostic process from patient chief complaint and imaging examinations to pathology assessment and final diagnosis; in department management, it supports natural-language queries, automatically summarizes equipment loads and disease distribution, and generates department operations reports; in research, it can perform multidimensional statistical analysis on clinical disease data sets, train simple machine-learning models and produce analysis reports that can be edited for paper submission. A clinical high-quality data platform automatically cleans, quality-checks and structures multi-source hospital data. CAIR said CARES 4.0 has been validated in several top-tier hospitals. Its website is open, the framework and general toolkit have been released, and medical toolkits are being rolled out; open-source ultrasound and surgical video models developed by the center will also be integrated and made freely available.

At the opening ceremony, CAIR Director Liu Hongbin, Hong Kong Science and Technology Parks Corporation’s chief executive, China Resources Health General Manager and China Resources Medical President Zhang Chuang, Executive Dean of the CAS Hong Kong Institute of Science and Innovation Hao Yinxing, Deputy Director of the CAS Institute of Automation Lü Pin and MGI Tech CEO Liu Jian spoke. A roundtable of hospital leaders from Hong Kong and Shenzhen included representatives from Gleneagles Hong Kong, the University of Hong Kong-Shenzhen Hospital, CUHK Medical Centre, Guangdong 999 Brain Hospital, Shenzhen Luohu People’s Hospital, Lingnan Hospital of the Third Affiliated Hospital of Sun Yat-sen University and the University of Hong Kong Faculty of Dentistry. The participants said private institutions introducing high-value AI medical systems need to assess investment returns while considering operational sustainability and public health value; cross-border medical AI applications should strictly comply with legal boundaries and explore data exchange under legal frameworks; smart hospital construction should focus on business process reengineering; clinical adoption depends on high-quality data and clinician-engineer collaboration, and the core challenge is defining risks and allocating responsibility in diagnosis and treatment scenarios; primary care should adopt AI step by step, starting with experience-optimization applications; rehabilitation AI can support assessment and treatment planning; and dental care can use AI for precise diagnosis and home screening.

Academic sessions featured Sebastien Ourselin, Nassir Navab, Liu Yunhui, Liu Hongbin, Liao Hongen and Ni Dong, who discussed medical devices and digital health ecosystems, human-robot collaboration in medical robotics, AI-driven surgical robots, clinical agent AI systems, imaging decision-making for diagnostic robots and intelligent ultrasound research and translation. Clinical experts from the First Affiliated Hospital of Sun Yat-sen University, Qilu Hospital of Shandong University, China-Japan Friendship Hospital, Tongji Hospital, the Chinese University of Hong Kong, Nanfang Hospital, the First Affiliated Hospital of Jinan University, Xuanwu Hospital and Queen Elizabeth Hospital shared experience in medical imaging, ultrasound diagnosis, intelligent nursing, cardiothoracic surgery, sleep medicine, vascular ultrasound and neurosurgery. Researchers from Shenzhen, Hong Kong and the National University of Singapore also presented work on AI algorithms, medical robot systems and clinical translation.

The forum held a venture roadshow with projects covering optical multidimensional force sensors, laparoscopic surgical robots, minimally invasive surgical devices and auxiliary equipment, a sleep large model, ultrasound robots, surgical robots, surgical navigation systems, ASIC chips and active implantable medical devices, brain-computer interfaces and bionic muscles, surgical planning and simulation, and personalized orthopedics and reconstruction. Hong Kong Science and Technology Parks Corporation hosted a session on entrepreneurship from idea to commercialization, detailing its life and health technology ecosystem and startup community. A medical technology exhibition featured companies and more than ten startups nurtured by the Incu-Bio platform. At CAIR’s embodied AI operating room, guests observed CARES, laryngoscope and bronchoscope diagnostic agents, the AI sleep system, the “Lingyin” ultrasound large model, a handheld ultrasound device, an ultrasound scanning robot, a narrow-space minimally invasive surgery solution, a real-time medical physics simulation platform, interventional low-field MRI and the MicroNeuro neurosurgical embodied AI robot.

In an interview with Leiphone.com, Liu Hongbin said CARES 1.0 to 3.0 was a large model that could answer multimodal questions about images or videos but remained a tool requiring doctors to operate. CARES 4.0 is an agent that can do work for doctors. For example, if a clinical department director wants a report on last month’s surgical volume, consumables and patients for hospital leadership or internal discussion, the agent system can complete it autonomously, interacting with the director to confirm requirements and report type. On surgical risk warning, Liu said warnings include spatial warnings when AI identifies an operation near important blood vessels, arteries or nerves, and temporal warnings related to surgical sequence, such as whether a bone window is large enough for complete tumor removal; spatial warnings are currently the main application. At the First Affiliated Hospital of Sun Yat-sen University, the system has been used with Dr Liao Huai during bronchoscopy, warning that a bulge appearing to be a bronchus may actually be part of a blood vessel, where biopsy could rupture an artery. On hallucination and misdiagnosis risk, Liu said CARES acts as an assistant and gives suggestions rather than conclusions. The team seeks to minimize hallucinations, especially for decision-related advice, by having the AI follow clinical experts’ reasoning logic, including guidelines and the latest clinical results from top international journals, and by structuring the information the AI can use in a knowledge-graph-like form so that recommendations come with their reasoning path. The interview transcript was edited by Leiphone.com without changing its meaning.