Zhongshan Ophthalmic Center Licenses Gaze-Tracking-Based Visual Function Assessment Technology Co-Developed with Tencent

The team of Li Jinrong and Feng Lei at the Zhongshan Ophthalmic Center of Sun Yat-sen University has a research achievement currently undergoing technology transfer:Objective Detection and Rehabilitation Technology for Visual Function Based on Eye Tracking, Including 4 Patents, Transferred to Huishi Medical Technology (Guangzhou) Co., Ltd. for a Total Price of RMB 500,000.

What is contrast sensitivity?Visual function metrics that are more sensitive than conventional visual acuity charts can detect early signs of ophthalmic diseases such as retinopathy and glaucoma at an earlier stage. However, this examination has long faced an insurmountable hurdle: traditional testing relies on subjects pressing buttons or providing verbal responses—pressing a button when the target is clearly seen, or saying “I can’t see it clearly” when it is not. While this poses no problem for adults, what about children under three years of age? Or children with speech impairments?

Alternative approaches have indeed been attempted in clinical practice. These include the grating acuity chart method, the raster card method, and the preferential looking method. However, each of these techniques has its own limitations: some use fixed parameters and fail to capture the complete contrast sensitivity function curve; others rely on the examiner’s subjective judgment, allowing experienced clinicians to make accurate assessments at a glance, while less experienced practitioners are left to guess; and still others are too time-consuming to be widely adopted. For children aged 2 to 5 years, contrast sensitivity testing has long remained a significant blind spot.

The Zhongshan Ophthalmic Center team’s approach is to change a variable:Do not respond with behavior; use eye movements instead.

During testing, a light spot first appears on the screen to guide the child’s gaze to the center, after which a sinusoidal grating is presented on either the left or right side. Contrast and spatial frequency are adaptively determined by an algorithm. With each presentation, the eye tracker records where the child’s gaze lands: if the stimulus is seen, the gaze naturally shifts toward it; if not, the gaze remains stationary. Based on these gaze data, the system updates its estimates in real time using Bayesian methods, while the algorithm determines the parameters for the next trial, prioritizing areas of greatest uncertainty until the entire curve is fitted. Requiring no button presses or verbal feedback, patent literature indicates that completing binocular testing takes approximately 20 minutes.

To implement this testing protocol, two prerequisite issues must be addressed:The algorithm needs to determine "where the eye is looking" and "which region in the image seen by the eye is the corneal limbus."The Zhongshan Ophthalmic Center team also incorporated these two issues into this batch of patents.

Eye tracking itself relies on precise imaging of the eyes.Patent for VR Glasses Rehabilitation Equipment (ZL202322569257.6) designs an integrated solution:The fill light and camera are mounted on a bracket below the display device, with the camera tilted at an angle of 30° to 60° toward the eyes. Infrared fill lighting is used in conjunction with an infrared camera to ensure tracking accuracy in a closed VR environment. The distance-measuring device detects interpupillary distance (IPD) in real time, and the main control board drives the optical engine module to automatically adjust for different users, eliminating the need for manual calibration.

The patent for the limbus localization algorithm (ZL202211451710.7) addresses the issue of "where the limbus is" in images.The team designed a set of multi-resolution convolutional kernels, ranging from a minimum size of 15×15 pixels to a maximum of 127×127 pixels. Each kernel incorporates a three-layer weight structure: -1 for the periphery, +1 for the limbal region, and -1 for the interior. As the limbus is a grayish-white transitional zone, the region with a weight of +1 generates the strongest response in edge images. For localization, low-resolution kernels are first used to rapidly scan all candidate positions. Then, inspired by particle filtering, the resolution is progressively increased to filter for the optimal candidate. Finally, precise localization is achieved through iterative ellipse optimization based on nearest-point association. This method requires less computational power than the traditional Hough transform, offers an interpretable iterative process, and outputs the center coordinates of the limbus along with elliptical geometric parameters, making it directly applicable to surgical navigation.

Three patents point in the same direction:Transform subjective, coordination-dependent, experience-based examinations into an objective, automated, hardware-supported system.Li Jinrong’s team had previously collaborated with Tencent to embed similar algorithms into an intelligent visual function screening system for children. Over 800 clinical validations were conducted at the Zhongshan Ophthalmic Center, reducing the single examination time from approximately two hours to three to five minutes. The system is currently in clinical use.

Inventors: Li Jinrong, Feng Lei, et al.

Patent Nos.: ZL202211451710.7 (Limbal Localization), CN202410838468.1 (CSF Detection), ZL202322569257.6 (VR Rehabilitation Device), ZL202330613911.1 (Visual Training and Rehabilitation Instrument).

Source: Public Notice on the Transformation of Scientific and Technological Achievements by Zhongshan Ophthalmic Center, Sun Yat-sen University (2026-04-03);

Tencent Official Website, “Innovative Solutions for Efficient Identification of Eye Diseases” (2022).