Predicting 10-2 Visual Field From Optical Coherence Tomography in Glaucoma Using Deep Learning Corrected With 24-2/30-2 Visual Field.

Purpose: To investigate whether a correction based on a Humphrey field analyzer (HFA) 24-2/30-2 visual field (VF) can improve the prediction performance of a deep learning model to predict the HFA 10-2 VF test from macular optical coherence tomography (OCT) measurements. Methods: This is a multicenter, cross-sectional study. The training dataset comprised 493 eyes of 285 subjects (407, open-angle glaucoma [OAG]; 86, normative) who underwent HFA 10-2 testing and macular OCT. The independent testing dataset comprised 104 OAG eyes of 82 subjects who had undergone HFA 10-2 test, HFA 24-2/30-2 test, and macular OCT. A convolutional neural network (CNN) DL model was trained to predict threshold sensitivity (TH) values in HFA 10-2 from retinal thickness measured by macular OCT. The predicted TH values was modified by pattern-based regularization (PBR) and corrected with HFA 24-2/30-2. Absolute error (AE) of mean TH values and mean absolute error (MAE) of TH values were compared between the CNN-PBR alone model and the CNN-PBR corrected with HFA 24-2/30-2. Results: AE of mean TH values was lower in the CNN-PBR with HFA 24-2/30-2 correction than in the CNN-PBR alone (1.9dB vs. 2.6dB; P = 0.006). MAE of TH values was lower in the CNN-PBR with correction compared to the CNN-PBR alone (4.2dB vs. 5.3 dB; P < 0.001). The inferior temporal quadrant showed lower prediction errors compared with other quadrants. Conclusions: The performance of a DL model to predict 10-2 VF from macular OCT was improved by the correction with HFA 24-2/30-2. Translational Relevance: This model can reduce the burden of additional HFA 10-2 by making the best use of routinely performed HFA 24-2/30-2 and macular OCT.

Predicting the central 10 degrees visual field in glaucoma by applying a deep learning algorithm to optical coherence tomography images.

We aimed to develop a model to predict visual field (VF) in the central 10 degrees in patients with glaucoma, by training a convolutional neural network (CNN) with optical coherence tomography (OCT) images and adjusting the values with Humphrey Field Analyzer (HFA) 24-2 test. The training dataset included 558 eyes from 312 glaucoma patients and 90 eyes from 46 normal subjects. The testing dataset included 105 eyes from 72 glaucoma patients. All eyes were analyzed by the HFA 10-2 test and OCT; eyes in the testing dataset were additionally analyzed by the HFA 24-2 test. During CNN model training, the total deviation (TD) values of the HFA 10-2 test point were predicted from the combined OCT-measured macular retinal layers' thicknesses. Then, the predicted TD values were corrected using the TD values of the innermost four points from the HFA 24-2 test. Mean absolute error derived from the CNN models ranged between 9.4 and 9.5 B. These values reduced to 5.5 dB on average, when the data were corrected using the HFA 24-2 test. In conclusion, HFA 10-2 test results can be predicted with a OCT images using a trained CNN model with adjustment using HFA 24-2 test.

Deep learning model to predict visual field in central 10° from optical coherence tomography measurement in glaucoma.

BACKGROUND/AIM: To train and validate the prediction performance of the deep learning (DL) model to predict visual field (VF) in central 10° from spectral domain optical coherence tomography (SD-OCT). METHODS: This multicentre, cross-sectional study included paired Humphrey field analyser (HFA) 10-2 VF and SD-OCT measurements from 591 eyes of 347 patients with open-angle glaucoma (OAG) or normal subjects for the training data set. We trained a convolutional neural network (CNN) for predicting VF threshold (TH) sensitivity values from the thickness of the three macular layers: retinal nerve fibre layer, ganglion cell layer+inner plexiform layer and outer segment+retinal pigment epithelium. We implemented pattern-based regularisation on top of CNN to avoid overfitting. Using an external testing data set of 160 eyes of 131 patients with OAG, the prediction performance (absolute error (AE) and R2 between predicted and actual TH values) was calculated for (1) mean TH in whole VF and (2) each TH of 68 points. For comparison, we trained support vector machine (SVM) and multiple linear regression (MLR). RESULTS: AE of whole VF with CNN was 2.84±2.98 (mean±SD) dB, significantly smaller than those with SVM (5.65±5.12 dB) and MLR (6.96±5.38 dB) (all, p<0.001). Mean of point-wise mean AE with CNN was 5.47±3.05 dB, significantly smaller than those with SVM (7.96±4.63 dB) and MLR (11.71±4.15 dB) (all, p<0.001). R2 with CNN was 0.74 for the mean TH of whole VF, and 0.44±0.24 for the overall 68 points. CONCLUSION: DL model showed considerably accurate prediction of HFA 10-2 VF from SD-OCT.

Evaluation of the relationship between glaucomatous disc subtypes and occurrence of disc hemorrhage and glaucoma progression in open angle glaucoma.

To compare the occurrence of disc hemorrhages (DH) and glaucoma progression in open-angle glaucoma (OAG) patients with different glaucomatous disc types. Prospective, hospital-based, observational cohort study. OAG patients examined between 2000 and 2005, whose discs were classified as typical myopic glaucomatous (MG), generalized enlargement of cup (GE), or focal glaucomatous (FG) disc type were included and followed for 5 years. The first occurrence of DH during follow-up was analyzed using Kaplan-Meier analysis and difference in DH occurrence based on glaucomatous disc type using the Cox proportional-hazards model to adjust for effects of confounding factors. For inter-group comparison of glaucoma progression, the change rate of the mean deviation, Collaborative Initial Glaucoma Treatment Study scores, and fundus photographs were used. Thirty-nine patients with MG-, 18 with FG-, and 17 with GE-disc types were included. No significant inter-group difference was seen in the rate of glaucoma progression. The five-year probability of DH occurrence was much lower with MG- than with FG- or GE-disc types (P < 0.0220). The central corneal thickness (P = 0.0024) and mean intraocular pressure and its variations (P = 0.0450, 0.0219) contributed to DH occurrence. The MG-disc type demonstrated a much lower DH occurrence during follow-up than other disc types.

Relationship Between Macular Ganglion Cell Thickness and Ocular Elongation as Measured by Axial Length and Retinal Artery Position.

Purpose: We recently reported on the usefulness of retinal artery trajectory in estimating the magnitude of retinal stretch due to myopia. The purpose of the present study was to elucidate the relationship between the peripapillary retinal artery angle (PRAA) and thickness of the macular ganglion cell-inner plexiform layer (GCIPL). Methods: This r included 138 healthy eyes of 79 subjects older than 20 years of age without any known eye disease. GCIPL thickness was separated into eight sectors according to quadrant and eccentricity from the fovea. The PRAA was calculated as the angle between the superior and inferior retinal arteries. Relationships between whole GCIPL thickness (average and sectorial) and the values of PRAA and axial length (AL) were investigated using a linear mixed model. Results: Average GCIPL thickness in the whole scanned area decreased significantly with narrowing of the PRAA with and without adjusting for AL. Sectorized macular GCIPL thickness also decreased significantly, with narrowing of the PRAA in seven out of the eight with the adjustment of AL, the exception being the inferior peripheral temporal sector. Conclusions: Macular GCIPL thickness decreased significantly with narrowing of the PRAA on average and in seven out of eight sectors.

Predicting the Glaucomatous Central 10-Degree Visual Field From Optical Coherence Tomography Using Deep Learning and Tensor Regression.

PURPOSE: To predict the visual field (VF) of glaucoma patients within the central 10° from optical coherence tomography (OCT) measurements using deep learning and tensor regression. DESIGN: Cross-sectional study. METHODS: Humphrey 10-2 VFs and OCT measurements were carried out in 505 eyes of 304 glaucoma patients and 86 eyes of 43 normal subjects. VF sensitivity at each test point was predicted from OCT-measured thicknesses of macular ganglion cell layer + inner plexiform layer, retinal nerve fiber layer, and outer segment + retinal pigment epithelium. Two convolutional neural network (CNN) models were generated: (1) CNN-PR, which simply connects the output of the CNN to each VF test point; and (2) CNN-TR, which connects the output of the CNN to each VF test point using tensor regression. Prediction performance was assessed using 5-fold cross-validation through the root mean squared error (RMSE). For comparison, RMSE values were also calculated using multiple linear regression (MLR) and support vector regression (SVR). In addition, the absolute prediction error for predicting mean sensitivity in the whole VF was analyzed. RESULTS: RMSE with the CNN-TR model averaged 6.32 ± 3.76 (mean ± standard deviation) dB. Significantly (P < .05) larger RMSEs were obtained with other models: CNN-PR (6.76 ± 3.86 dB), SVR (7.18 ± 3.87 dB), and MLR (8.56 ± 3.69 dB). The absolute mean prediction error for the whole VF was 2.72 ± 2.60 dB with the CNN-TR model. CONCLUSION: The Humphrey 10-2 VF can be predicted from OCT-measured retinal layer thicknesses using deep learning and tensor regression.

Comparing Structure-Function Relationships Based on Drasdo's and Sjöstrand's Retinal Ganglion Cell Displacement Models.

Purpose: To compare structure-function relationships based on the Drasdo and Sjöstrand retinal ganglion cell displacement models. Methods: Single eyes from 305 patients with glaucoma and 55 heathy participants were included in this multicenter, cross-sectional study. The ganglion cell and inner plexiform layer (GCIPL) thickness was measured using spectral domain optical coherence tomography. Visual field measurements were performed using the Humphrey 10-2 test. All A-scan pixels (128 × 512 pixels) were allocated to the closest 10-2 location with both displacement models using degree and millimeter scales. Structure-function relationships were investigated between GCIPL thickness and corresponding visual sensitivity in nonlong (160 eyes) and long (200 eyes) axial length (AL) groups. Results: In both the nonlong and long AL groups, compared with the no-displacement model, both the Drasdo and the Sjöstrand models showed that the structure-function relationship around the fovea improved (P < 0.05). The magnitude of improvement in the area was either comparable between the model or was larger for the Drasdo model than the Sjöstrand model (P < 0.05). Meanwhile, structure-function relationships outside the innermost retinal region that were based on the Drasdo and Sjöstrand models were comparable to or were even worse than (in the case of the Drasdo model) those obtained using the no-displacement model. Conclusions: Structure-function relationships evaluated based on both the Drasdo and Sjöstrand models significantly improved around the fovea, particularly when using the Drasdo model. This was not the case in other areas.

Using Deep Learning and Transfer Learning to Accurately Diagnose Early-Onset Glaucoma From Macular Optical Coherence Tomography Images.

PURPOSE: We sought to construct and evaluate a deep learning (DL) model to diagnose early glaucoma from spectral-domain optical coherence tomography (OCT) images. DESIGN: Artificial intelligence diagnostic tool development, evaluation, and comparison. METHODS: This multi-institution study included pretraining data of 4316 OCT images (RS3000) from 1371 eyes with open angle glaucoma (OAG) regardless of the stage of glaucoma and 193 normal eyes. Training data included OCT-1000/2000 images from 94 eyes of 94 patients with early OAG (mean deviation > -5.0 dB) and 84 eyes of 84 normal subjects. Testing data included OCT-1000/2000 from 114 eyes of 114 patients with early OAG (mean deviation > -5.0 dB) and 82 eyes of 82 normal subjects. A DL (convolutional neural network) classifier was trained using a pretraining dataset, followed by a second round of training using an independent training dataset. The DL model input features were the 8 × 8 grid macular retinal nerve fiber layer thickness and ganglion cell complex layer thickness from spectral-domain OCT. Diagnostic accuracy was investigated in the testing dataset. For comparison, diagnostic accuracy was also evaluated using the random forests and support vector machine models. The primary outcome measure was the area under the receiver operating characteristic curve (AROC). RESULTS: The AROC with the DL model was 93.7%. The AROC significantly decreased to between 76.6% and 78.8% without the pretraining process. Significantly smaller AROCs were obtained with random forests and support vector machine models (82.0% and 67.4%, respectively). CONCLUSION: A DL model for glaucoma using spectral-domain OCT offers a substantive increase in diagnostic performance.

Improving the structure-function relationship in glaucomatous and normative eyes by incorporating photoreceptor layer thickness.

The purpose of the study was to investigate whether the structure-function relationship in glaucomatous and normative eyes is improved by considering photoreceptor layer thickness. Humphrey 10-2 visual fields (VF) and optical coherence tomography were carried out in 615 eyes of 391 subjects, including 100 eyes of 53 healthy controls and 515 eyes of 338 glaucoma patients. The relationship between mean VF sensitivity and the thickness of the retinal nerve fiber layer (RNFL) and ganglion cell layer and inner plexiform layer (GCL + IPL) was analyzed using linear mixed models, by glaucoma status and degree of myopia. The structure-function relationship was also analyzed by supplementing the RNFL and GCL + IPL thickness with the thicknesses of: (i) the inner nuclear layer and outer plexiform layer (INL + OPL); (ii) the outer nuclear layer and inner segment of photoreceptor layer (ONL + ISL); (iii) the outer segment layer of photoreceptor and retinal pigment epithelium (OSL + RPE). The model included total thickness of RNFL, GCL + IPL and OSL + RPE was highly more optimal than the model that only included the total thickness of RNFL and GCL + IPL, in all subsets of eyes by glaucoma status and degree of myopia.

Mapping the Central 10° Visual Field to the Optic Nerve Head Using the Structure-Function Relationship.

Purpose: To investigate the structure-function mapping in the central 10° by relating Humphrey field analyzer (HFA) 10-2 visual field (VF) and circumpapillary retinal nerve fiber layer (cpRNFL) thickness from spectral-domain optical coherence tomography (SD-OCT). We also compared the obtained results with a previously reported mapping between 10-2 VF and the optic disc. Methods: In 151 eyes of 151 POAG patients and 35 eyes from 35 healthy participants, cpRNFL thickness measurements were obtained using SD-OCT and the 10-2 VF was measured with the HFA. The relationship between visual sensitivity and cpRNL thickness values in the temporal 180° was analyzed using least absolute shrinkage and selection operator (LASSO) regression. The optic disc angle corresponding to each VF test point was then derived using the coefficients from the optimal LASSO regression. Results: The structure-function map obtained was largely consistent with the mapping reported previously; superior central VF test points correspond to a more vulnerable area of the optic disc, more distant toward the inferior pole from the center of the temporal quadrant (9:00 o'clock for the right eye) while inferior VF test points correspond closer to the center of the temporal quadrant. The prediction error tended to be large in the 'more vulnerable area' in the map reported previously. Conclusions: The structure-function map obtained largely confirms the previously reported map; however, some important differences were observed.

The association between photoreceptor layer thickness measured by optical coherence tomography and visual sensitivity in glaucomatous eyes.

PURPOSE: To assess the thickness of the photoreceptor layer in the macular region in glaucomatous eyes. METHOD: Humphrey 10-2 visual field (VF) testing was carried out and mean threshold (mTH) was calculated in 118 eyes from 118 patients with open angle glaucoma. Macular optical coherence tomography (OCT) measurements (RS 3000, Nidek Co.ltd., Aichi, Japan) were also carried out in all eyes. Thickness measurements were recorded in the outer segment and retinal pigment epithelium (OS+RPE), the nerve fiber layer (NFL), the ganglion cell layer and inner plexiform layer (GCL+IPL), the inner nuclear layer and outer plexiform layer (INL+OPL) and the outer nuclear layer and inner segment (ONL+IS). The relationship between mTH and the thickness of these five different layers was investigated. Additionally, the influence of OS+RPE on mTH was investigated using partial correlation eliminating the effect of other variables of NFL, GCL+IPL, INL+OPL, ONL+IS, age, gender and axial length. RESULTS: The thickness of the OS+RPE layer was significantly decreased with the decrease of mTH (coefficient = 0.63 p <0.001). Partial correlation analysis suggested OS+RPE thickness is significantly (coefficient = 0.31, p <0.001) related to mTH, independent from NFL, GCL+IPL, INL+OPL, ONL+IS, age, gender and axial length. CONCLUSIONS: The thickness of the RPE+OS layer appears to be related to visual sensitivity in glaucoma.

Visual field damage in normal-tension glaucoma patients with or without ischemic changes in cerebral magnetic resonance imaging.

PURPOSE: To compare the pattern of visual field damage between normal-tension glaucoma (NTG) patients with signs indicative of ischemic changes and those NTG patients without signs of ischemic changes, using brain magnetic resonance imaging (MRI), in a single center, cross-sectional study. METHODS: In 94 consecutive NTG patients who were younger than 61 years old, brain MRI images were obtained using fluid-attenuated inversion recovery pulse sequences. The presence of signs indicative of ischemic changes in brain MRI images was decided separately by two neuroradiologists masked to the diagnosis and stage of glaucoma. Visual field testing was performed using the 30-2 program of the Humphrey Visual Field Analyzer. Between the patients with signs indicative of ischemic changes in brain MRI (ischemic group) and those without MRI signs (nonischemic group), total deviation (TD) at each test point less the average of TDs of the 30-2 program ([TD - TD(mean)])--was compared at each test point. RESULTS: Signs indicative of ischemic changes in brain MRI were found in 32 of the 94 patients (34.0%). Age, blood pressure, refraction, intraocular pressure, the average of TDs, mean deviation, and corrected pattern standard deviation were not significantly different between the ischemic (N = 32) and nonischemic (N = 62) groups (P > 0.2). [TD - TD(mean)] in the ischemic group was significantly smaller than that in the nonischemic group at 6 nonedge contiguous test points in the inferior pericentral to nasal field (P = 0.005-0.047). CONCLUSION: NTG patients with signs indicative of ischemic changes in brain MRI had a relatively deeper depression in the inferior pericentral visual field.