Multimodal Machine Learning Using Visual Fields and Peripapillary Circular OCT Scans in Detection of Glaucomatous Optic Neuropathy.

To develop and validate a multimodal artificial intelligence algorithm, FusionNet, using the pattern deviation probability plots from visual field (VF) reports and circular peripapillary OCT scans to detect glaucomatous optic neuropathy (GON). Cross-sectional study. Two thousand four hundred sixty-three pairs of VF and OCT images from 1083 patients. FusionNet based on bimodal input of VF and OCT paired data was developed to detect GON. Visual field data were collected using the Humphrey Field Analyzer (HFA). OCT images were collected from 3 types of devices (DRI-OCT, Cirrus OCT, and Spectralis). Two thousand four hundred sixty-three pairs of VF and OCT images were divided into 4 datasets: 1567 for training (HFA and DRI-OCT), 441 for primary validation (HFA and DRI-OCT), 255 for the internal test (HFA and Cirrus OCT), and 200 for the external test set (HFA and Spectralis). GON was defined as retinal nerve fiber layer thinning with corresponding VF defects. Diagnostic performance of FusionNet compared with that of VFNet (with VF data as input) and OCTNet (with OCT data as input). FusionNet achieved an area under the receiver operating characteristic curve (AUC) of 0.950 (0.931-0.968) and outperformed VFNet (AUC, 0.868 [95% confidence interval (CI), 0.834-0.902]), OCTNet (AUC, 0.809 [95% CI, 0.768-0.850]), and 2 glaucoma specialists (glaucoma specialist 1: AUC, 0.882 [95% CI, 0.847-0.917]; glaucoma specialist 2: AUC, 0.883 [95% CI, 0.849-0.918]) in the primary validation set. In the internal and external test sets, the performances of FusionNet were also superior to VFNet and OCTNet (FusionNet vs VFNet vs OCTNet: internal test set 0.917 vs 0.854 vs 0.811; external test set 0.873 vs 0.772 vs 0.785). No significant difference was found between the 2 glaucoma specialists and FusionNet in the internal and external test sets, except for glaucoma specialist 2 (AUC, 0.858 [95% CI, 0.805-0.912]) in the internal test set. FusionNet, developed using paired VF and OCT data, demonstrated superior performance to both VFNet and OCTNet in detecting GON, suggesting that multimodal machine learning models are valuable in detecting GON.

Xiong J ，Li F ，Song D ，Tang G ，He J ，Gao K ，Zhang H ，Cheng W ，Song Y ，Lin F ，Hu K ，Wang P ，Olivia Li JP ，Aung T ，Qiao Y ，Zhang X ，Ting D ... -  《-》

From Machine to Machine: An OCT-Trained Deep Learning Algorithm for Objective Quantification of Glaucomatous Damage in Fundus Photographs.

Previous approaches using deep learning (DL) algorithms to classify glaucomatous damage on fundus photographs have been limited by the requirement for human labeling of a reference training set. We propose a new approach using quantitative spectral-domain (SD) OCT data to train a DL algorithm to quantify glaucomatous structural damage on optic disc photographs. Cross-sectional study. A total of 32 820 pairs of optic disc photographs and SD OCT retinal nerve fiber layer (RNFL) scans from 2312 eyes of 1198 participants. The sample was divided randomly into validation plus training (80%) and test (20%) sets, with randomization performed at the patient level. A DL convolutional neural network was trained to assess optic disc photographs and predict SD OCT average RNFL thickness. The DL algorithm performance was evaluated in the test sample by evaluating correlation and agreement between the predictions and actual SD OCT measurements. We also assessed the ability to discriminate eyes with glaucomatous visual field loss from healthy eyes with area under the receiver operating characteristic (ROC) curves. The mean prediction of average RNFL thickness from all 6292 optic disc photographs in the test set was 83.3±14.5 μm, whereas the mean average RNFL thickness from all corresponding SD OCT scans was 82.5±16.8 μm (P = 0.164). There was a very strong correlation between predicted and observed RNFL thickness values (Pearson r = 0.832; R2 = 69.3%; P < 0.001), with mean absolute error of the predictions of 7.39 μm. The area under the ROC curves for discriminating glaucomatous from healthy eyes with the DL predictions and actual SD OCT average RNFL thickness measurements were 0.944 (95% confidence interval [CI], 0.912-0.966) and 0.940 (95% CI, 0.902-0.966), respectively (P = 0.724). We introduced a novel DL approach to assess fundus photographs and provide quantitative information about the amount of neural damage that can be used to diagnose and stage glaucoma. In addition, training neural networks to predict SD OCT data objectively represents a new approach that overcomes limitations of human labeling and could be useful in other areas of ophthalmology.

Medeiros FA ，Jammal AA ，Thompson AC 《-》

Human Versus Machine: Comparing a Deep Learning Algorithm to Human Gradings for Detecting Glaucoma on Fundus Photographs.

To compare the diagnostic performance of human gradings vs predictions provided by a machine-to-machine (M2M) deep learning (DL) algorithm trained to quantify retinal nerve fiber layer (RNFL) damage on fundus photographs. Evaluation of a machine learning algorithm. An M2M DL algorithm trained with RNFL thickness parameters from spectral-domain optical coherence tomography was applied to a subset of 490 fundus photos of 490 eyes of 370 subjects graded by 2 glaucoma specialists for the probability of glaucomatous optical neuropathy (GON), and estimates of cup-to-disc (C/D) ratios. Spearman correlations with standard automated perimetry (SAP) global indices were compared between the human gradings vs the M2M DL-predicted RNFL thickness values. The area under the receiver operating characteristic curves (AUC) and partial AUC for the region of clinically meaningful specificity (85%-100%) were used to compare the ability of each output to discriminate eyes with repeatable glaucomatous SAP defects vs eyes with normal fields. The M2M DL-predicted RNFL thickness had a significantly stronger absolute correlation with SAP mean deviation (rho=0.54) than the probability of GON given by human graders (rho=0.48; P < .001). The partial AUC for the M2M DL algorithm was significantly higher than that for the probability of GON by human graders (partial AUC = 0.529 vs 0.411, respectively; P = .016). An M2M DL algorithm performed as well as, if not better than, human graders at detecting eyes with repeatable glaucomatous visual field loss. This DL algorithm could potentially replace human graders in population screening efforts for glaucoma.

Jammal AA ，Thompson AC ，Mariottoni EB ，Berchuck SI ，Urata CN ，Estrela T ，Wakil SM ，Costa VP ，Medeiros FA ... -  《-》

Correlation of localized glaucomatous visual field defects and spectral domain optical coherence tomography retinal nerve fiber layer thinning using a modified structure-function map for OCT.

Wu H ，de Boer JF ，Chen L ，Chen TC ... -  《-》

A Comparison between the Compass Fundus Perimeter and the Humphrey Field Analyzer.

To evaluate relative diagnostic precision and test-retest variability of 2 devices, the Compass (CMP, CenterVue, Padova, Italy) fundus perimeter and the Humphrey Field Analyzer (HFA, Zeiss, Dublin, CA), in detecting glaucomatous optic neuropathy (GON). Multicenter, cross-sectional, case-control study. We sequentially enrolled 499 patients with glaucoma and 444 normal subjects to analyze relative precision. A separate group of 44 patients with glaucoma and 54 normal subjects was analyzed to assess test-retest variability. One eye of recruited subjects was tested with the index tests: HFA (Swedish interactive thresholding algorithm [SITA] standard strategy) and CMP (Zippy Estimation by Sequential Testing [ZEST] strategy), 24-2 grid. The reference test for GON was specialist evaluation of fundus photographs or OCT, independent of the visual field (VF). For both devices, linear regression was used to calculate the sensitivity decrease with age in the normal group to compute pointwise total deviation (TD) values and mean deviation (MD). We derived 5% and 1% pointwise normative limits. The MD and the total number of TD values below 5% (TD 5%) or 1% (TD 1%) limits per field were used as classifiers. We used partial receiver operating characteristic (pROC) curves and partial area under the curve (pAUC) to compare the diagnostic precision of the devices. Pointwise mean absolute deviation and Bland-Altman plots for the mean sensitivity (MS) were computed to assess test-retest variability. Retinal sensitivity was generally lower with CMP, with an average mean difference of 1.85±0.06 decibels (dB) (mean ± standard error, P < 0.001) in healthy subjects and 1.46±0.05 dB (mean ± standard error, P < 0.001) in patients with glaucoma. Both devices showed similar discriminative power. The MD metric had marginally better discrimination with CMP (pAUC difference ± standard error, 0.019±0.009, P = 0.035). The 95% limits of agreement for the MS were reduced by 13% in CMP compared with HFA in participants with glaucoma and by 49% in normal participants. Mean absolute deviation was similar, with no significant differences. Relative diagnostic precision of the 2 devices is equivalent. Test-retest variability of MS for CMP was better than for HFA.

Montesano G ，Bryan SR ，Crabb DP ，Fogagnolo P ，Oddone F ，McKendrick AM ，Turpin A ，Lanzetta P ，Perdicchi A ，Johnson CA ，Garway-Heath DF ，Brusini P ，Rossetti LM ... -  《-》