Table 3 Overall performance summary of Ci-SSGAN vs. ICD code

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

Metric

ICD Code

Ci-SSGAN (25%)

Ci-SSGAN (100%)

Gain vs. ICD (25%)

Gain vs. ICD (100%)

Accuracy

0.743

0.773

0.853

\(+\)2.96

\(+\)10.91

F1 score

0.746

0.765

0.856

\(+\)1.93

\(+\)10.97

AUROC

0.847

0.942

0.949

\(+\)9.46

\(+\)10.19

AUCPR

0.625

0.821

0.847

\(+\)19.62

\(+\)22.22

  1. All values are calculated on the test dataset. Bold values show the best scores. AUROC and AUCPR calculated using one-vs-rest strategy with macro-averaging across six glaucoma classes. Bold values indicate the best-performing result for each metric across the compared methods (ICD code, Ci-SSGAN (25%), and Ci-SSGAN (100%)).
Back to article page