Focal nerve fiber layer reflectance analysis for glaucoma evaluation

项目总结 青光眼是导致失明的主要原因，有效的青光眼治疗需要及早发现。 利用光学相干断层扫描(OCT)测量神经纤维层(NFL)厚度 确认青光眼的诊断，但其诊断敏感性不足以单独用于 基于人群的筛查。 青光眼患者的NFL反射率降低，可能是由于轴突和轴突微管的丢失所致 内容。但其诊断价值因其对OCT光束的入射角的依赖而减弱， 这在常规的临床成像中是高度可变的。我们假设诊断的准确率可以是 通过对NFL反射率剖面进行方位滤波来降低入射角效应，并通过分析 而不是平均反射率的变化。初步结果，基于100名正常和青光眼 诊断灵敏度从平均NFL厚度的71%显著提高到 PG眼局灶性NFL反射损失97%，特异度99%。我们建议在以下方面验证这一结果 大型青光眼高级成像(AIG)研究数据集，包括249例周围性青光眼(PG)， 252例周围性青光眼(PPG)患者和145例正常受试者。AIG的这项研究的平均后续行动是 4年以上，这也可以评估预测青光眼进展的准确性。 1.利用大型AIG再现焦域NFL反射损耗分析的高诊断精度 数据集。如果我们能再次在AIG数据集中证明高诊断准确性，特别是在PPG中 和早期PG亚组，这可能将OCT青光眼评估纳入人群筛查领域。 主要的性能指标将是在固定的99%特异度临界点的诊断灵敏度。 2.用焦域NFL反射损耗预测视野转换和进展。在美国国际集团 黄斑神经节细胞复合体(GCC)和乳头周围神经纤维层局灶性变薄的研究 被发现是室颤转换的最好预测因子(青光眼性室颤异常的发生 基础室颤正常的眼)和进展(室颤显著恶化)。我们假设焦点是 NFL的反射损耗将具有更好的预测准确性。将对预测准确性进行评估 使用受试者操作曲线下面积(AROC)和Logistic回归(优势比)。 3.利用机器学习将OCT反射率和结构图相结合来改善青光眼 诊断准确率。视盘、乳头周围和黄斑厚度参数的组合 以前被证明是协同作用的，产生比任何单一参数更高的AROC。我们 假设将新的NFL反射损耗图添加到输入参数集将 进一步提高了机器学习算法的诊断精度。

PROJECT SUMMARY Glaucoma is a leading cause of blindness, and effective glaucoma management requires early detection. Nerve fiber layer (NFL) thickness measurement by optical coherence tomography (OCT) is useful for confirming the diagnosis of glaucoma, but its diagnostic sensitivity is not sufficient to be used alone for population-based screening. NFL reflectivity is reduced in glaucoma subjects, presumably due to loss of axons and axonal microtubule content. But its diagnostic value is diminished by its dependence on the incident angle of the OCT beam, which is highly variable in routine clinical imaging. We hypothesize that the diagnostic accuracy can be boosted by reducing incidence angle effects with azimuthal filtering of NFL reflectance profile, and by analysis of focal rather than average reflectance changes. The preliminary result, bases on 100 normal and glaucoma eyes, showed that the diagnostic sensitivity was significantly improved from 71% for average NFL thickness to 97% for focal NFL reflectance loss in PG eyes, at a 99% specificity cutoff. We propose to validate this result in the large Advanced Imaging for Glaucoma (AIG) study dataset that comprises 249 perimetric glaucoma (PG), 252 pre-perimetric glaucoma (PPG), and 145 normal participants. The AIG study has an average follow-up of more than 4 years, which also allows assessment of the accuracy in predicting glaucoma progression. 1. Reproduce the high diagnostic accuracy of focal NFL reflectance loss analysis using the large AIG dataset. If we could again demonstrate high diagnostic accuracy in the AIG dataset, especially in the PPG and early PG subgroups, this could bring OCT glaucoma evaluation into the realm of population screening. The primary performance metric will be the diagnostic sensitivity at a fixed 99% specificity cut point. 2. Use focal NFL reflectance loss to predict visual field (VF) conversion and progression. In the AIG study, focal thinning of the macular ganglion cell complex (GCC) and peripapillary nerve fiber layer (NFL) were found to be the best predictors of VF conversion (development of glaucomatous VF abnormality in an eye with normal baseline VF) and progression (significant worsening of VF). We hypothesize that focal NFL reflectance loss would have even better predictive accuracy. Predictive accuracy will be assessed using the area under the receiver operating curve (AROC) and logistic regression (odds ratio). 3. Combine OCT reflectance and structural maps using machine learning to improve glaucoma diagnostic accuracy. A combination of disc, peripapillary, and macular thickness parameters had previously been shown to be synergistic, producing higher AROC than any single parameter. We hypothesize that the addition of the novel NFL reflectance loss map to the set of input parameters will further enhance the diagnostic accuracy of a machine learning algorithm.