Augmented eye clinic shows why AI-native systems work where standalone algorithms don't

New AI-native system, called AI-TEC, is revolutionizing eye care at Beijing Tsinghua Changgung Hospital in China. This framework, which combines two powerful AI engines - Qwen3.5-35B for text analysis and RETFound for retinal images - operates on a network of specialized AI agents that collaborate throughout a patient's care journey.

The interface has been optimized for speed, reducing the number of clicks needed, and allowing doctors to access key AI recommendations in just three clicks. However, initial results showed diagnostic accuracy scores of 0.7996, which were insufficient for real-world use. To improve accuracy, four eye specialists manually selected and verified 1,426 high-quality images to fine-tune the AI, raising the accuracy score to 0.9383.

With this enhancement, the system can now provide doctors with consistent, evidence-based recommendations, allowing for personalized care. The researchers note that AI-native systems, which integrate AI across the entire care process, offer a more holistic approach to healthcare, as opposed to standalone tools that focus on isolated tasks.

Initial adoption in the hospital was high, reaching 25.7% in December, but decreased to 3.8% in April due to slow performance and excessive manual data entry. Redesigning the system to load faster and require fewer clicks improved usage to 23.0% within a month. By incorporating feedback from doctors during actual consultations, the AI's accuracy score further improved to 0.9482 when tested on a second cohort of patients.

The researchers believe that AI-TEC has the potential to streamline administrative tasks and reduce cognitive fatigue for clinicians, ultimately saving time during patient consultations. However, further real-world trials are needed to evaluate its impact on patient outcomes and overall hospital efficiency.