Skip to main content
. 2021 Mar 11;4:48. doi: 10.1038/s41746-021-00417-4

Fig. 1. Schematic diagram of the hierarchical deep learning system (HDLS) simulating doctors’ diagnostic thinking.

Fig. 1

a The process of doctors’ diagnosis of glaucoma. b The process of diagnosing glaucoma of the HDLS. This system includes three modules: The pre-diagnosis module is used for preliminary diagnosis based on the overall information of the fundus photographs. The image segmentation module is used to segment OD, OC, and RNFLD based on the pre-diagnostic results. For fundus photographs pre-diagnosed as glaucoma, a glaucoma-specific network is used to segment the OC. The final diagnosis module is used to calculate MCDR and ISNT-score based on the segmentation. Finally, these two indicators are used to make a final diagnosis in combination with the status of RNFLD. OD optic disc, OC optic cup, RNFLD retinal nerve fiber layer defects, MCDR mean cup-to-disk ratio, ISNT inferior, superior, nasal, temporal.