Perimetry, the mapping of the sensitivity of different visual field locations, is an essential procedure in ophthalmology. Unfortunately, standard automated perimetry (SAP), suffers from some practical issues: it can be tedious, requires manual feedback and a high level of patient compliance. These factors limit the effectiveness of perimetry in some clinical populations. In an attempt to remove some of these limitations, alternatives to SAP have been tried based on tracking eye movements. These new approaches have attempted to mimic SAP, thus presenting stimuli on a fixed grid, and replacing manual by ocular responses. While this solves some issues of SAP, these approaches hardly exploit the high spatial and temporal resolution facilitated by eye-tracking. In this study, we present two novel computational methods that do tap into this potential: (1) an analytic method based on the spatio-temporal integration of positional deviations by means of Threshold Free Cluster Enhancement (TFCE) and (2) a method based on training a recursive deep artificial neural network (RNN). We demonstrate that it is possible to reconstruct visual field maps based on continuous gaze-tracking data acquired in a relatively short amount of time.

中文翻译：

---

基于时空积分和深度递归神经网络的连续眼动视野测量方法

视野测定法是不同视野位置的灵敏度的映射，是眼科的基本程序。不幸的是，标准自动视野检查（SAP）遇到一些实际问题：这可能很乏味，需要人工反馈和高度的患者依从性。这些因素限制了视野计在某些临床人群中的有效性。为了消除这些限制中的一些限制，已经基于跟踪眼睛的运动尝试了SAP的替代方法。这些新方法试图模仿SAP，从而在固定的网格上显示刺激，并通过眼部反应代替人工。虽然这解决了SAP的某些问题，但是这些方法几乎没有利用眼动追踪所带来的高时空分辨率。在这项研究中，我们提出了两种确实可以发挥这种潜力的新颖计算方法：（1）一种基于阈值时空积分的阈值自由聚类增强（TFCE）的解析方法，（2）一种基于递归深度人工神经网络（RNN）训练的方法。我们证明，有可能基于在相对较短的时间内获取的连续凝视跟踪数据来重建视野图。