Figure 5 | Scientific Reports

Figure 5

From: RetiNerveNet: using recursive deep learning to estimate pointwise 24-2 visual field data based on retinal structure

Figure 5

Overview of the RetiNerveNet structure: A 768-dimensional RNFL thickness vector obtained from the SDOCT test, arranged in the Temporal-Superior-Nasal-Inferior-Temporal (TSNIT) order, is split into two halves having size 384. As the retinal ganglion fibers in the eye do not cross into the opposite hemiretina, we use separate sub-networks to estimate the superior and inferior visual fields. The superior (inferior) half of the RNFL thickness values proceeds through the superior (inferior) sub-network of the RetiNerveNet to yield an estimate of the inferior (superior) half of the visual field. For the multi-task version of RetiNerveNet, a weight mask is applied to the estimated visual field to obtain a scalar estimate of Mean Deviation.

Back to article page