Anton Nicoleta, Lisa Cătălin, Doroftei Bogdan, Pîrvulescu Ruxandra Angela, Barac Ramona Ileana, Lungu Ionuț Iulian, Bogdănici Camelia Margareta
Department of Ophtalmology, "Grigore T. Popa" University of Medicine and Pharmacy, 16 Universității Street, 700115 Iasi, Romania.
Ophthalmology Clinic, Sf. Spiridon Emergency Clinical Hospital, 700111 Iasi, Romania.
Life (Basel). 2025 May 27;15(6):865. doi: 10.3390/life15060865.
AI systems, especially artificial neural networks (ANNs), are increasingly involved in the diagnosis and personalized management of ophthalmologic disorders.
This study shows the practical applications of artificial intelligence for predicting the progression of intraocular hypertension (IOH) to glaucoma.
This study involved two groups of patients with IOH and a control group, analyzed using the commercial Neurosolution simulator. The findings were compared with experimental data. The performance of the neural models was evaluated using several metrics: Mean Squared Error (MSE), Normalized Mean Squared Error (NMSE), correlation coefficient (r), and percentage error (Ep).
For all three patient groups, the best performance was achieved with neural networks featuring two hidden layers: MLP(9:18:9:3) for group 1, MLP(10:20:10:3) for group 2, and MLP(10:30:20:3) for group 3. The MSE values during validation were 0.39 for groups 1 and 2, and 0.34 for group 3. For these neural networks, the probability of producing correct outputs during validation was 75% (i.e., 9 correct responses out of a possible 12). The findings in this study are in line with those reported by other researchers in the field.
The neural network models developed in this study demonstrated their potential for predicting the progression of intraocular hypertension to glaucoma.
人工智能系统，尤其是人工神经网络（ANNs），越来越多地参与到眼科疾病的诊断和个性化管理中。
本研究展示了人工智能在预测高眼压症（IOH）向青光眼进展方面的实际应用。
本研究纳入了两组高眼压症患者和一组对照组，使用商业神经解决方案模拟器进行分析。研究结果与实验数据进行了比较。使用多种指标评估神经模型的性能：均方误差（MSE）、归一化均方误差（NMSE）、相关系数（r）和百分比误差（Ep）。
对于所有三组患者，具有两个隐藏层的神经网络表现最佳：第一组为MLP(9:18:9:3)，第二组为MLP(10:20:10:3)，第三组为MLP(10:30:20:3)。验证期间第一组和第二组的MSE值为0.39，第三组为0.34。对于这些神经网络，验证期间产生正确输出的概率为75%（即，在可能的12个响应中有9个正确响应）。本研究的结果与该领域其他研究人员报告的结果一致。
本研究开发的神经网络模型展示了其在预测高眼压症向青光眼进展方面的潜力。
