![]() Analyses of the optic nerve head (ONH) and RNFL alone, therefore, potentially overlook glaucomatous macular damage. Although these studies are enlightening, mounting evidence from the investigation of glaucomatous damage implicates early macular involvement 10, 16, 17. They showed the potential of deep learning models to provide quantitative information on the extent of neural damage from qualitative data (optic disc photographs). Recently, several studies have shown that RNFL thickness and minimum rim width relative to Bruch’s membrane opening (BMO-MRW) under SD-OCT can be successfully quantified from monoscopic optic disc photographs by deep learning models 14, 15. It is useful not only for diagnosing glaucoma but also for monitoring glaucoma progression even before apparent visual field (VF) change 12, 13. Spectral domain-optical coherence tomography (SD-OCT) is widely utilized for detection and quantitative assessment of glaucomatous structural loss of retinal nerve fiber layer (RNFL) and macular ganglion cell-inner plexiform layer (mGCIPL) 8, 9, 10, 11. This labeling process is labor-intensive as well as subjective. However, the deep learning models of most of the previous studies require ground truth labeling by human graders. reported a deep learning model that had been trained on a very large-scale dataset with over 48,000 fundus photographs and that achieved an AUC of 0.986 for referable glaucomatous optic neuropathy 6. reported on a deep learning system for multiethnic diabetic cohorts that achieved an AUC of 0.942 7. A number of studies have reported the diagnostic performance of the deep learning model as excellent in terms of area under receiver operating characteristic curve (AUC). There has been remarkable progress in glaucoma screening thanks to the development of deep learning algorithms such as convolutional neural networks (CNNs) for visual recognition 4, 5, 6. Effective screening strategies are important, as most patients do not present any symptoms before the disease has reached the advanced stage 2, 3. Glaucoma is the leading cause of visual impairment worldwide, affecting more than 70 million people 1. The trained HDLM algorithm showed a great capability for mGCIPL thickness prediction from RNFLPs. Even when the peripapillary area (diameter: 1.5 disc diameters) was masked, the correlation ( r = 0.713 P < 0.001) and agreement (MAE = 4.87 µm) were not changed significantly ( P = 0.378 and 0.724, respectively). The predicted mGCIPL thickness showed a strong correlation and good agreement with the measured mGCIPL thickness (Correlation coefficient r = 0.739 P < 0.001 MAE = 4.76 µm). The correlation coefficient and mean absolute error (MAE) between the predicted and measured mGCIPL thicknesses were calculated. An HDLM was built by combining a pre-trained deep learning network and support vector machine. A total of 789 pairs of RNFLPs and spectral domain-optical coherence tomography (SD-OCT) scans for 431 eyes of 259 participants (183 eyes of 114 healthy controls, 68 eyes of 46 glaucoma suspects, and 180 eyes of 99 glaucoma patients) were enrolled. We developed a hybrid deep learning model (HDLM) algorithm that quantitatively predicts macular ganglion cell-inner plexiform layer (mGCIPL) thickness from red-free retinal nerve fiber layer photographs (RNFLPs). ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |