在近视眼检测方面，基于眼底影像的人工智能(AI)技术可直接从眼底彩色图像中检测屈光不正度与眼轴长度，利用多任务学习技术则可进一步提升其联合预测精度。在近视眼病变识别方面，改进的U-Net与Transformer在视盘旁萎缩、脉络膜厚度及豹纹状眼底等病灶的识别分割上已接近甚至超过专家水平；当前，病理性近视眼Meta分析系统(META-PM)的自动化分级体系逐渐成熟，光学相干断层扫描(OCT)成像技术在牵引性和新生血管性黄斑病变检测中表现突出。在近视眼风险预测方面，结合眼底彩色图像中与基线屈光度、眼轴及年龄等临床信息的模型，可较准确预测儿童向高度近视眼进展的风险，为筛查与个体化管理提供支持。但因数据来源异质性、设备差异及合并眼病导致的泛化不足仍亟待解决。未来在多模态大数据融合、可解释性、不确定性量化以及跨设备的通用化平台等方面有望在短期内快速推动病理性近视眼进程全程智能管理。

Fundus image-based artificial intelligence (AI) technology has shown substantial promise for myopia detection. AI models can directly infer refractive status and axial length from color fundus photographs, and multi-task learning strategies may further improve the accuracy of joint prediction. For myopia-related lesion recognition, enhanced U-Net architectures and Transformer-based models have achieved near-expert performance-and in some tasks even surpassed expert graders-in detecting and segmenting findings such as peripapillary atrophy, choroidal thickness-related changes, and tessellated fundus. Currently, automated grading under the Meta-analysis for pathologic myopia (META-PM) classification framework is also becoming increasingly mature. In parallel, optical coherence tomography (OCT) demonstrates clear advantages in identifying tractional and neovascular macular complications. In terms of risk prediction, models that integrate color fundus photography with baseline clinical information-including refractive error, axial length, and age can more accurately estimate the risk of progression to high myopia in children, thereby supporting screening and individualized management. Nevertheless, insufficient generalizability remains a major barrier, driven by heterogeneity in data sources, inter-device variability, and coexisting ocular conditions. In the near future, rapid advances in multimodal large-scale data fusion, interpretability and uncertainty quantification, and device-agnostic platforms are expected to accelerate end-to-end intelligent management of pathologic myopia.