Deep learning classification of early normal-tension glaucoma and glaucoma suspects using Bruch’s membrane opening-minimum rim width and RNFL

Abstract

We aimed to classify early normal-tension glaucoma (NTG) and glaucoma suspect (GS) using Bruch’s membrane opening-minimum rim width (BMO-MRW), peripapillary retinal nerve fiber layer (RNFL), and the color classification of RNFL based on a deep-learning model. Discriminating early-stage glaucoma and GS is challenging and a deep-learning model may be helpful to clinicians. NTG accounts for an average 77% of open-angle glaucoma in Asians. BMO-MRW is a new structural parameter that has advantages in assessing neuroretinal rim tissue more accurately than conventional parameters. A dataset consisted of 229 eyes out of 277 GS and 168 eyes of 285 patients with early NTG. A deep-learning algorithm was developed to discriminate between GS and early NTG using a training set, and its accuracy was validated in the testing dataset using the area under the curve (AUC) of the receiver operating characteristic curve (ROC). The deep neural network model (DNN) achieved highest diagnostic performance, with an AUC of 0.966 (95%confidence interval 0.929–1.000) in classifying either GS or early NTG, while AUCs of 0.927–0.947 were obtained by other machine-learning models. The performance of the DNN model considering all three OCT-based parameters was the highest (AUC 0.966) compared to the combinations of just two parameters. As a single parameter, BMO-MRW (0.959) performed better than RNFL alone (0.914).

Molecular Genetics of Glaucoma: Subtype and Ethnicity Considerations

“…For example, Guidoboni et al, have created mathematical models for ocular blood flow in order to understand underlying vascular RFs [ 239 ]. Similarly, Seo and Cho utilized deep learning techniques, specifically a deep neural network, to evaluate the association between specific optical coherence tomography-based parameters and NTG [ 240 ]. As these techniques develop further and more genetic information is made available for use, these research techniques hold promise in elucidating a genetically driven model for glaucoma pathogenesis.…”

Glaucoma Detection Using Support Vector Machine Method Based on Spectralis OCT

“…To our knowledge, our study was the few ones which evaluated the application of machine learning technique in complicated Spectralis OCT parameters for glaucoma detection, including ppRNFL, ONH, and macular parameters. Several published literatures have explored the use of Spectralis OCT parameters to construct machine learning classifiers for glaucoma diagnosis [ 16 , 37 , 38 , 39 ]. Kim et al developed several machine learning models, including SVM for glaucoma diagnosis, using ppRNFL parameters and clinical features (age, IOP, and corneal thickness) and visual field information, and they found the random forest model had the best performance, with an AUC value of 0.979 and AUC value of the SVM model at 0.967 [ 37 ].…”

“…Park et al used a multilayer neural network to combine BMO-MRW and ppRNFL parameters for glaucoma diagnosis, which showed better performance than using either BMO-MRW or ppRNFL data alone [ 38 ]. A deep learning classification model was adopted by Seo et al for discriminating early normal tension glaucoma from glaucoma, which suspected and showed the best performance, considering three OCT-based parameters together (BMO-MRW, ppRNFL, and the color classification of ppRNFL), with an AUC value of 0.966 [ 39 ]. Though it is difficult to directly compare our results with previous research, due to the differences in the subjects included, as well as the OCT parameters and machine learning methods used.…”

Comparison of Different Machine Learning Classifiers for Glaucoma Diagnosis Based on Spectralis OCT

“…Seo et al investigated the diagnostic accuracy of six MLCs using BMO-MRW, cRNFL, and cRNFL color codes from SD-OCT to discriminate between early normal tension glaucoma patients from glaucoma suspects. The RF classifier was the second-best performing MLC with an AUC of 0.947, and the deep neural network model was the best one with an AUC of 0.966 [ 23 ]. These studies suggest that RF is a powerful and reliable machine learning method.…”

“…Machine learning classifiers (MLCs) are well-established analytical methods especially good at detecting the relationship between a huge amount of input parameters, eventually facilitating the diagnosis of a condition [ 14 ]. In fact, some reports suggest that MLCs are as good as [ 15 , 16 ], or even better [ 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 ] than, currently available techniques for glaucoma diagnosis. However, what most people criticize MLCs for is the analysis process being a “black box” [ 25 , 26 , 27 ], for it produces results based solely on the input data using an algorithm, which prevents clinicians from understanding how variables are being combined to make such a prediction.…”