A deep learning approach to predict visual field using optical coherence tomography

Abstract

We developed a deep learning architecture based on Inception V3 to predict visual field using optical coherence tomography (OCT) imaging and evaluated its performance. Two OCT images, macular ganglion cell-inner plexiform layer (mGCIPL) and peripapillary retinal nerve fibre layer (pRNFL) thicknesses, were acquired and combined. A convolutional neural network architecture was constructed to predict visual field using this combined OCT image. The root mean square error (RMSE) between the actual and predicted visual fields was calculated to evaluate the performance. Globally (the entire visual field area), the RMSE for all patients was 4.79 ± 2.56 dB, with 3.27 dB and 5.27 dB for the normal and glaucoma groups, respectively. The RMSE of the macular region (4.40 dB) was higher than that of the peripheral region (4.29 dB) for all subjects. In normal subjects, the RMSE of the macular region (2.45 dB) was significantly lower than that of the peripheral region (3.11 dB), whereas in glaucoma subjects, the RMSE was higher (5.62 dB versus 5.03 dB, respectively). The deep learning method effectively predicted the visual field 24-2 using the combined OCT image. This method may help clinicians determine visual fields, particularly for patients who are unable to undergo a physical visual field exam.

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

“…We propose RetiNerveNet, a deep fully convolutional neural architecture for obtaining estimates of SAP visual field values based on RNFL thickness values obtained from the more objective SDOCT tests. Unlike existing works with similar aim 9 , 10 , 13 – 15 , 17 , 18 , we postulate that building our network to mimic the arcuate structure of the axons of the retinal ganglion cells can help improve performance for this task. The fact that the proposed architecture performs better than a number of baselines in Table 2 seems to corroborate our hypothesis.…”

“…Other works 15 – 17 , more closely related to our own, have attempted to estimate pointwise sensitivities for all the locations tested by SAP, based on the SDOCT thickness values. Some studies 18 , 19 have even attempted to predict the SAP sensitivities based on the raw images obtained from SDOCT. However, these previous investigations have not made use of the known topographic characteristics of the RNFL when attempting to estimate SAP data 20 .…”

RetiNerveNet: Using Recursive Deep Learning to Estimate Pointwise 24-2 Visual Field Data based on Retinal Structure

“…We propose RetiNerveNet, a deep fully convolutional neural architecture for obtaining estimates of SAP visual field values based on RNFL thickness values obtained from the more objective SDOCT tests. Unlike existing works with similar aim 8,9,[11][12][13][14][15] , we postulate that building our network to mimic the arcuate structure of the axons of the retinal ganglion cells can help improve performance for this task. The fact that the proposed architecture performs better than a number of baselines in Table 2 seems to corroborate our hypothesis.…”

“…Some existing studies 11,12 , have used such deep learning techniques to estimate SAP summary metrics like Mean Deviation (MD, a weighted average of the age-corrected visual field values) using information acquired with SDOCT. Other works [13][14][15] , more closely related to our own, have attempted to estimate pointwise sensitivities for all the locations tested by SAP, based on SDOCT measurements. However, these previous investigations have not made use of the known topographic characteristics of the RNFL when attempting to estimate SAP data 16 .…”

The Role of Artificial Intelligence in the Diagnosis and Management of Glaucoma

“…In regards to global indices, the root-mean-square error was 3.27 dB for normal patients and 5.27 dB for glaucoma patients, reflecting that the prediction error in normal subjects was significantly lower than that in glaucoma patients (P < 0.001). 57 Hashimoto and colleagues developed a CNN algorithm to predict SAP VF changes within the central 10 degrees from visual fixation from SD-OCT images that included the RNFL, GCC, and outer segment-retinal pigment epithelium. The application demonstrated a wholefield analysis with an MAE of 2.84 dB, R 2 of 0.74 and outperformed 2 alternate classic ML algorithms.…”