人工智慧與青光眼

Artificial Intelligence and Glaucoma

本綜述闡述了人工智慧（artificial intelligence, AI）在青光眼診斷、疾病進展預測丶以及手術結果評估等方面的最新應用與臨床意義。傳統的青光眼診斷方式雖然完整，但常受操作經驗或主觀判斷影響。近年來，AI技術透過整合多種檢查方式，及結合病人的臨床資料，能自動偵測視網膜神經纖維層（retinal nerve fiber layer, RNFL）變薄、視神經結構異常等早期症狀，並減少人為誤差，幫助醫師診斷。疾病進展預測方面，多項研究證實，無論是深度學習，如卷積神經網路（convolutional neural network, CNN）、PointNet分析三維影像，或利用轉換器模型（transformer）和長短期記憶網路（long short-term memory, LSTM）進行時間序列分析，都能大褔提升預測精準度，在早期偵測及進展追蹤上具有明顯優勢。而在結構與功能並重的檢查，如光學相干斷層掃描（optical coherence tomography, OCT）、視神經乳頭攝影（optic-disc photographs, ODP）、視野數據，可顯著提高AI模型對快速惡化族群（MD每年減少大於1.0dB）的辨識能力；可視化儀表板（dashboard）更提供了直觀的臨床操作介面。手術評估領域，結合電子病歷中人口學特徵、全身健康資訊以及眼科手術史等多維資料，可幫助醫師更準確地預測手術成功機率或失敗風險，利於制定個人化治療計畫與評估術後併發症風險。整體而言，AI模型不僅能節省診療時間，也能為醫師和病人提供更客觀丶系統化的決策依據，未來透過持續優化演算法與多模態資料的導入，AI在青光眼整合式管理中將扮演更重要的輔助角色。

This review explains the latest applications and clinical significance of artificial intelligence (AI) in glaucoma diagnosis, disease progression prediction, and surgical outcome assessment. Traditional methods of glaucoma diagnosis, though comprehensive, often rely heavily on operator experience and subjective judgment. In recent years, AI technologies have been integrated with various diagnostic modalities and clinical data to automatically detect early signs such as thinning of the retinal nerve fiber layer(RNFL) and structural abnormalities of the optic nerve, thereby reducing human error and assisting clinicians in diagnosis. Regarding disease progression, multiple studies have demonstrated that the use of deep learning (e.g., convolutional neural networks, pointnet) for three-dimensional data analysis, or time-series analysis via transformers and long short-term memory (LSTM) networks, significantly increases predictive accuracy—particularly for early detection and monitoring of disease progression. Furthermore, combining structural and functional examinations [e.g., optical coherence tomography (OCT), optic-disc photographs (ODP), and visual field data] has been shown to markedly enhance AI models’ability to identify patients at high risk for rapid deterioration (Mean Deviation >1.0 dB loss per year), while dashboard visualization tools provide an intuitive clinical interface. In the field of surgical evaluation, incorporating demographic factors, systemic health information, and prior ophthalmic surgical history from electronic health records improves predictions of surgical success or failure, facilitating individualized treatment planning and perioperative complication risk assessment. Overall, AI models not only save clinical time but also offer more objective, systematic decision support for physicians and patients. With continued algorithm optimization and the integration of multimodal data, AI is poised to play an increasingly pivotal role in comprehensive glaucoma management.