Guibas Intelligence Announces IPO Filing for Deep Learning-Based AI Software for Diabetic Retinopathy Diagnosis, Cleared as China’s First Class III Medical Device in Ophthalmology

Led by Shenzhen SiBright Co., Ltd. (hereinafter referred to as “SiBright”), the research and development team successfully created an AI-assisted diagnostic product for medical imaging with independent intellectual property rights—the “Fundus Image-Assisted Diagnostic Software for Diabetic Retinopathy” (hereinafter referred to as “DR-Assisted Diagnostic Software”)—after five years of relentless effort. On August 7, 2020, it was approved by the National Medical Products Administration as a Class III medical device, becoming one of the first fundus lesion-assisted diagnostic software products based on deep learning technology to receive such approval. This milestone also marks another significant achievement in the National Key R&D Program’s special project on “Digital Diagnosis and Treatment Equipment.”

In terms of underlying technology, the product employs a self-designed convolutional neural network architecture. The model was trained and validated for diabetic retinopathy using a dataset of over 103,000 fundus images from Chinese patients, with cross-annotations provided by dozens of ophthalmologists. Nine domestic invention patents and two international (PCT) invention patents have been filed for the core technologies, four of which have already been granted.

In terms of product registration and filing, SiBright collaborated with the entire industry chain encompassing academia, research, production, healthcare, and testing to jointly promote the development of industry standards and regulations. In early 2018, SiBright deeply participated in the construction of China’s first AI standard test dataset (fundus portion), led by the National Institutes for Food and Drug Control (NIFDC) and jointly undertaken by nine enterprises. By the end of April, it obtained the first AI-assisted diagnostic product registration inspection report issued by the NIFDC in China, with key performance indicators of 86.8% sensitivity and 91.9% specificity. Meanwhile, through extensive communication with the Artificial Intelligence Medical Device Innovation Cooperation Platform, SiBright conducted product cybersecurity testing at the China Academy of Information and Communications Technology (CAICT), with all test items meeting regulatory requirements. In April 2019, SiBright completed a multi-center prospective clinical trial involving 1,000 samples, led by the Zhongshan Ophthalmic Center and participating institutions including Peking University People’s Hospital and Wenzhou Eye Hospital. The key performance indicators were 87.29% sensitivity and 95.51% specificity, placing the product’s performance at a world-leading level (the sensitivity and specificity of IDx-DR, the first FDA-approved AI software for diabetic retinopathy diagnosis globally, were 87.4% and 89.5%, respectively).

In terms of product application value, the diabetic retinopathy diagnostic software has learned from the extensive professional experience of fundus specialists. Through the widespread application of 5G networks, it is deployed in endocrinology departments and primary healthcare institutions to assist endocrinologists and general practitioners in diagnosing and managing ocular complications of diabetes. This effectively helps diabetic patients receive timely diagnosis or referral and prompt intervention, thereby avoiding the risk of blindness. Furthermore, it significantly increases the opportunity for diabetic patients to undergo diabetic retinopathy screening, promotes tiered diagnosis and treatment, empowers primary healthcare institutions, and makes high-quality medical resources accessible to the grassroots population. Upon regulatory approval, the product will be initially implemented in Hainan Province, Shandong Province, and Shenzhen.