Background/aimsTo identify objective glaucoma-related structural features based on peripapillary (p) and macular (m) spectral domain optical coherence tomography (SD-OCT) parameters and assess their discriminative ability between healthy and glaucoma patients.MethodsTwo hundred and sixty eyes (91 controls and 169 glaucoma) were included in this prospective study. After a complete examination, all participants underwent the posterior pole and the peripapillary retinal nerve fibre layer (pRNFL) protocols of the Spectralis SD-OCT. Principal component analysis (PCA), a data reduction method, was applied to identify and characterise the main information provided by the ganglion cell complex (GCC). The discriminative ability between healthy and glaucomatous eyes of the first principal components (PCs) was compared with that of conventional SD-OCT parameters (pRNFL, macular RNFL (mRNFL), macular ganglion cell layer (mGCL)and macular inner plexiform layer (mIPL)) using 10-fold cross-validated areas under the curve (AUC).ResultsThe first PC explained 58% of the total information contained in the GCC and the pRNFL parameters and was the result of a general combination of almost all variables studied (diffuse distribution). Other PCs were driven mainly by pRNFL and mRNFL measurements. PCs and pRNFL had similar AUC (0.95 vs 0.96, p=0.88), and outperformed the other structural measurements: mRNFL (0.91, p=0.002), mGCL (0.92, p=0.02) and mIPL (0.92, p=0.0001).ConclusionsPCA identified a diffuse representation of the papillary and macular SD-OCT parameters as the most important PC to summarise structural data in healthy and glaucomatous eyes. PCs and pRNFL parameters showed the greatest discriminative ability between healthy and glaucoma cases.

The mean VFI regression slope in our cohort of eyes without perimetric progression showed a statistically significant difference compared with those with suspected and definite progression. VFI analysis and GPA II both had similarly high specificity but low sensitivity when compared with expert consensus opinion.

The most effective indices were maximum contour elevation, reference height and PULSAR-sLV, although the inclusion of the optic nerve head assessment in the selection of the GS sample may have favored the HRT-II results.

The combination of MD and PSD regression analysis in the Quasar program showed better agreement with the experts and higher sensitivity than GPA II.

No abstract

PurposeTo determine the diagnostic generalizability of two deep learning models when trained only with images of the ganglion cell layer (GCL) of mild glaucoma.MethodsWe have collected a sample from patients with primary and secondary open‐angle glaucoma and normal patients. The sample was divided into mild glaucoma (MD≤6 dB), and moderate‐advanced (MD > 6 dB). The GCL images were recorded with a spectral‐domain Optical Coherence Tomography. Two pre‐trained models were used, the ResNet101 and the Shufflenet. The sensitivity, specificity, diagnostic precision in training and test, and the ROC area were calculated for the two models with three different training conditions according to how the images were partitioned into training and test. In the first partition, mild glaucomas were used for training and moderate‐advanced for test. In the second, moderate‐advanced glaucomas were used for training and mild for test. In the third, the whole sample was used without classifying by severity. Gradient‐weighted Class Activation Mapping (GradCAM) was used to obtain saliency maps which highlight the more important components in the images for the model prediction. The correlation coefficient between the maps of the glaucoma and normal images of the two models was calculated.Results561 eyes were collected from 337 patients, 356 are glaucomatous and 200 are normal. The precision of the models in the test set in partition 1, was 90.9% (ResNet101) and 94.2% (Shufflenet). In partition 2, was 74.4% (ResNet101) and 73.5% (Shufflenet), and in partition 3 an accuracy of 94.6% was found with both models. The correlation coefficient between the GradCAM saliency maps of the models was 0.46 for glaucoma images and 0.83 for normal images.ConclusionsThe two deep learning models are able to generalize and have high diagnostic precision if they are trained only with images of the GCL of mild glaucoma. Both models show high correlation in the GradCAM saliency maps with normal images.

No abstract

No abstract