Fig. 1: Overview of the Ci-SSGAN framework. | npj Digital Medicine

Fig. 1: Overview of the Ci-SSGAN framework.

From: Clinically informed semi-supervised learning improves disease annotation and equity from electronic health records: a glaucoma case study

Fig. 1

a Abundant unlabeled ophthalmology notes compared with limited expert annotations. b Ci-SSGAN generator combines unlabeled text embeddings (blue arrow), demographics (orange arrow), and noise (dashed black arrow) to produce synthetic data, with the discriminator classifying glaucoma subtypes and output evaluated to ensure fairness by measuring each subgroup disparities. Non-GL=non-glaucoma, OAG/S= open angle glaucoma/suspect, ACG/S= angle closure glaucoma/suspect, XFG/S= exfoliation glaucoma/syndrome, PDG/S= pigmentary dispersion glaucoma/syndrome, and SGL= secondary glaucoma. R= real, F= fake. Parts of this figure were created with BioRender.com.

Back to article page