人工智能用于泪道引流障碍：诊断、管理与临床转化进展
Artificial Intelligence for Lacrimal Drainage Disorders: Advances in Diagnosis, Management, and Clinical Translation
泪道引流障碍在眼科整形中常见，但现有检查多依赖主观判断和有创操作。这篇叙述性综述梳理了人工智能（包括传统机器学习、深度学习、影像组学和大语言模型）在该领域的应用证据。已有研究显示，机器学习模型可从泪道闪烁显像区分解剖性与功能性溢泪，可从眼前段光学相干断层扫描的泪河图像筛查泪道阻塞，可从泪囊造影自动定位阻塞部位，还可分割骨性鼻泪管以辅助手术规划；大语言模型在泪道知识测试中表现中等但进步很快。作者指出，这些证据多为单中心、早期阶段，下一步需要外部验证、前瞻性评估和影像数据标准化才能真正进入临床。
为什么推荐给您：系统梳理人工智能在泪道疾病中的新应用方向，属值得关注的方法学进展，但证据仍处早期。
不需要生物学背景，多打比方
正在获取全文并生成讲解（拿不到全文就依据摘要），大约需要 30–60 秒…
已等待 0 秒
这篇还没有动画
动画会把研究的流程、作用机制和关键结果一步一步演示出来，每一步都标明出自原文哪里。制作大约需要 30–60 秒。
摘要Abstract
PURPOSE: Lacrimal drainage disorders (LDDs) are common in oculoplastic practice, but their diagnostic evaluation continues to depend on tests that are variably subjective, operator-dependent, and at times invasive. In recent years, artificial intelligence (AI), spanning conventional machine learning (ML), deep learning (DL), radiomics, and large language models (LLMs), has shown increasing potential to enhance screening, diagnostic classification, lesion localization, and decision support in LDDs. This narrative review summarizes the current evidence on AI applications in the lacrimal field, places these studies within the broader context of AI in oculoplastic practice, and highlights the principal barriers to clinical translation.
METHODS: We conducted a narrative review of studies applying AI methods to lacrimal drainage disorders and related oculoplastic tasks, with emphasis on ML, DL, radiomics, image segmentation, and LLM-based approaches. Eligible evidence was synthesized according to major clinical use cases, including noninvasive screening, functional and anatomic diagnostic classification, obstruction localization, surgical planning, and patient-facing or clinician-facing decision support.
RESULTS: Dedicated studies have demonstrated the feasibility of ML and DL models for classifying anatomic versus functional epiphora from lacrimal scintigraphy, screening for lacrimal duct obstruction from anterior segment optical coherence tomography (AS-OCT) tear meniscus images, noninvasively predicting dacryocystitis from ocular surface indicators using conventional ML and explainable deep stacked networks, automatically detecting and localizing obstruction on dacryocystography (DCG), and segmenting the bony nasolacrimal canal on cone-beam computed tomography (CBCT) using the nnU-Net v2 architecture. In parallel, LLMs such as ChatGPT and DeepSeek have been benchmarked on lacrimal knowledge, showing average but rapidly improving performance.
CONCLUSIONS: AI in LDDs remains at an early, largely single-center stage, but the trajectory is clear: image-based models already approach clinician-level accuracy for well-defined tasks, explainable tabular models make noninvasive screening plausible, and segmentation networks are becoming precise enough to support surgical planning. External validation, prospective evaluation, standardization of lacrimal imaging data sets, and careful integration with clinical workflows are the next steps needed to translate these proofs of concept into routine lacrimal practice.