Figure 1.

There are five steps in the workflow. (A) The data collection step with HFA data from the Harvard database, external validation datasets from XJTU, and Octopus data from ZJU and PKU. (B) Process of converting VF data to Voronoi images. (C) Process of labeling. (D) Transfer learning process, including deep-learning models, FGG-H, and FGG-O. (E) Glaucoma grading step consisting of the classification of VF defect and saliency map. VF, visual field; HFA, humphrey field analyzer; VF, visual field; XJTU, xi'an Jiaotong university; ZJU, zhejiang university; PKU, peking university.