Free papers detail | 11th Asia Pacific Tele Ophthalmology Society Symposium

Title
Development and Clinical Application of an AI-Assisted Diagnosis and Treatment System for High Myopia Fundus Diseases

Authors
Yajun Wu

Presenting
Yajun Wu

PURPOSE:
To solve clinical dilemmas in high myopia fundus disease management (experience-dependent diagnosis, low image analysis efficiency, insufficient dynamic monitoring), develop a full-process AI-assisted diagnosis and treatment system. Achieve precise, personalized and intelligent clinical decision support, improve whole-life cycle management of patients, and reduce high myopia-induced blindness risk.

PURPOSE:

To solve clinical dilemmas in high myopia fundus disease management (experience-dependent diagnosis, low image analysis efficiency, insufficient dynamic monitoring), develop a full-process AI-assisted diagnosis and treatment system. Achieve precise, personalized and intelligent clinical decision support, improve whole-life cycle management of patients, and reduce high myopia-induced blindness risk.

METHODS:
Construct a multimodal database with 15,545 patients’ clinical/examination data (OCT, UWF, A/B ultrasound). Develop three AI models (diagnosis/classification via CNN/Transformer, treatment recommendation via LLM+RAG, risk prediction via time series analysis). Verify models through multi-center clinical trials (Nanchang/Shanghai/Changsha) and deploy the system on the Dify platform for clinical application.

METHODS:

Construct a multimodal database with 15,545 patients’ clinical/examination data (OCT, UWF, A/B ultrasound). Develop three AI models (diagnosis/classification via CNN/Transformer, treatment recommendation via LLM+RAG, risk prediction via time series analysis). Verify models through multi-center clinical trials (Nanchang/Shanghai/Changsha) and deploy the system on the Dify platform for clinical application.

RESULTS:
The diagnostic model achieved AUC≥0.82, sensitivity≥90%, specificity≥85%, with 85% consistency with senior physicians. The system provided ≥3 personalized schemes for complex cases (85% consistency with experts), saved 50% of physicians’ documentation time, increased expert daily consultations by 15-20%, advanced complication detection by 2-3 years, and raised follow-up completion rate by ≥20%.

RESULTS:

The diagnostic model achieved AUC≥0.82, sensitivity≥90%, specificity≥85%, with 85% consistency with senior physicians. The system provided ≥3 personalized schemes for complex cases (85% consistency with experts), saved 50% of physicians’ documentation time, increased expert daily consultations by 15-20%, advanced complication detection by 2-3 years, and raised follow-up completion rate by ≥20%.

CONCLUSIONS:
This AI system realizes closed-loop intelligent diagnosis and treatment of high myopia fundus diseases via multimodal data fusion and multi-AI technology application. It effectively improves clinical diagnosis and treatment efficiency, optimizes ophthalmic resource allocation, and provides a reliable intelligent tool for standardized management of high myopia patients, with significant clinical application value.

CONCLUSIONS:

This AI system realizes closed-loop intelligent diagnosis and treatment of high myopia fundus diseases via multimodal data fusion and multi-AI technology application. It effectively improves clinical diagnosis and treatment efficiency, optimizes ophthalmic resource allocation, and provides a reliable intelligent tool for standardized management of high myopia patients, with significant clinical application value.