Page 123 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

3) what type of foundation model is used? e.g. DeepSeek or GPT, etc

Response: The language model component of ChatZOC was independently developed by
the ZOC algorithm team. While it currently utilizes the Qwen 2.5 open-source architecture
as its foundation, the overall system is designed for flexibility and compatibility with multiple 4.1-Healthcare
large language model bases. This architecture ensures ChatZOC is not solely dependent on a
single foundation like Qwen 2.5 and allows for potential migration to other bases as needed.
Crucially, regardless of the base model, the system is heavily fine-tuned using ZOC’s proprietary
medical knowledge and extensive ophthalmic case data, creating a highly specialized tool
tailored for eye care.

4) are there references or articles or blogs or academic papers that can be referenced?

Response: The key underlying techniques like the Retrieval-Augmented Large Language
Model (RAG-LLM)[2] and Visionome[3] have been documented in publications such as JAMA
Ophthalmology and Nature Biomedical Engineering respectively.

5) are there open datasets or models?

Response: At present, the specific datasets compiled and the ChatZOC models developed are
not publicly available. This is due to strict patient data privacy laws and prevailing regulatory
constraints governing medical data.

6) could you add feedback or learnings from the deployments or pilots?

Response: Through previous pilot deployments, we have also identified areas for further model
optimization, such as designing it to be multilingual to meet the needs of users speaking
different regional dialects; the system needs to understand patients’ anxiety related to their
illness and enhance its capacity for humanistic care; and for users without a medical background,
the language used in responses needs to be easily understandable.

Use Case Status: The use case is part of a larger research project.

Partners: Huawei Technologies Co., Ltd.

2�2 Benefit of use case

ChatZOC leverages artificial intelligence technologies to deliver intelligent consultations,
precise screening, and diagnostic assistance, significantly improving the rate of early diagnosis
and treatment efficiency of ophthalmic diseases while reducing misdiagnosis rates. It helps
cover the shortage of medical resources and raises global standards for eye health. Additionally,
with the support of telemedicine technologies, patients can also access high-quality ophthalmic
medical services remotely, ensuring better health outcomes for more individuals.
AI large models can provide medical knowledge Q&A, virtual case-based teaching, and clinical
decision support, helping medical students and young doctors efficiently acquire ophthalmic
diagnostic and treatment skills, thus advancing the digital and intelligent development of
medical education.

Integrating cutting-edge artificial intelligence technologies with medicine, promote the
construction of smart medical infrastructure, and optimizing ophthalmic service processes,

87