军事知识服务平台
从临床经验到人工智能:斜视数字化诊断与智能手术规划系统的研究进展
From clinical experience to artificial intelligence:research progress in digital diagnosis and intelligent surgical planning systems for strabismus
摘要 斜视是常见眼科疾病,传统诊断和手术规划高度依赖临床经验,易受检查者经验、患者配合程度和医疗资源差异影响.近年来,人工智能(artificial intelligence,AI)、深度学习、眼动追踪和虚拟现实(virtual reality,VR)等数字化技术为斜视筛查、偏斜角量化、动态功能评估和手术辅助规划提供了新的工具.本文定位为叙述性综述,围绕近年代表性研究,归纳斜视数字化诊断与智能手术规划的主要技术路线、应用场景和证据边界.在数字化诊断方面,基于静态图像/视频的深度学习方法可实现斜视筛查、分型和偏斜角估计;眼动追踪方法有助于捕获间歇性斜视和双眼协调异常;移动端与嵌入式系统则提高了基层和居家筛查的可及性.在智能手术规划方面,机器学习回归、多任务学习、角膜光反射照片分析、生物力学建模和VR模拟等方法正在探索对手术肌肉选择、手术量设计和术后效果预测的辅助价值.现有研究提示,部分AI模型在特定数据集和受控条件下显示出较高性能,但多数证据仍来自单中心、回顾性或内部验证研究,复杂斜视类型、外部验证、多模态整合、长期预后预测及临床责任边界仍不足.未来研究应重视标准化数据采集、多中心前瞻性验证、可解释性、安全监管和临床工作流融合,使AI更稳妥地服务于斜视精准诊疗.
Abstract:
Strabismus is a common ocular disorder characterized by horizontal,vertical,and/or torsional misalignment of the visual axes.Accurate assessment of the type and magnitude of ocular deviation is essential for treatment selection and surgical planning.However,conventional diagnosis and surgical decision-making still rely heavily on clinical experience,manual observation,patient cooperation,and local availability of trained specialists.These factors may lead to variability in measurement and limit access to timely screening and referral in primary care settings.In recent years,artificial intelligence,deep learning,eye tracking,virtual reality,and other digital health technologies have provided new opportunities for objective and quantitative assessment of strabismus.This narrative review summarizes representative advances in digital diagnosis and intelligent surgical planning for strabismus,with an emphasis on clinical tasks,data modalities,algorithmic approaches,validation status,and application boundaries.For digital diagnosis,image-and video-based deep learning models have been explored for screening,subtype classification,and estimation of ocular deviation.Wearable or infrared eye-tracking systems can capture dynamic binocular coordination and may be particularly useful for intermittent strabismus or subtle abnormalities that are difficult to identify from a single static image.Smartphone-based and embedded systems may improve accessibility in community screening and remote follow-up,although their roles are currently better defined as risk stratification tools rather than replacements for specialist diagnosis.For surgical planning,machine learning regression,multi-task learning,artificial intelligence platforms based on corneal light-reflection photographs,biomechanical modeling,and virtual reality simulation have been investigated to support muscle selection,dosage prediction,and surgical training.These studies suggest that artificial intelligence may help standardize parts of the diagnostic and planning workflow.Nevertheless,current evidence remains limited by small or single-center datasets,retrospective design,internal validation,heterogeneous reference standards,and underrepresentation of complex strabismus,reoperations,vertical or torsional deviations,and patients with combined ocular or neurological conditions.Therefore,the performance reported in controlled studies should not be directly interpreted as readiness for autonomous clinical decision-making.Future research should prioritize standardized multimodal data acquisition,multicenter external validation,prospective real-world evaluation,explainable models,integration with clinical workflows,long-term outcome prediction,and clear ethical,regulatory,and responsibility frameworks.Artificial intelligence is more likely to become a safe and useful adjunct to strabismus care when it is developed as part of a clinician-centered,interpretable,and verifiable decision-support system.
刊 名 科学通报 ISTICEIPKU
年,卷(期): 2026, 71(15)
机标分类号: TP391.41 G350 TP181
在线出版日期: 2026年7月23日
基金项目: 31427801:国家自然科学基金,