Various machine-learning schemes have been proposed to diagnose glaucoma. They can classify subjects into 'normal' or 'glaucoma'-positive but cannot determine the severity of the latter. To complement this, researchers have proposed statistical indices for glaucoma risk. However, they are based on a single examination indicator and do not reflect the total severity of glaucoma progression. In this study, we propose an integrated glaucoma risk index (I-GRI) based on the visual field (VF) test, optical coherence tomography (OCT), and intraocular pressure (IOP) test. We extracted important features from the examination data using a machine learning scheme and integrated them into a single measure using a mathematical equation. The proposed index produces a value between 0 and 1; the higher the risk index value, the greater the risk/severity of glaucoma. In the sanity test using test cases, the I-GRI showed a balanced distribution in both glaucoma and normal cases. When we classified glaucoma and normal cases using the I-GRI, we obtained a misclassification rate of 0.07 (7%). The proposed index is useful for diagnosing glaucoma and for detecting its progression.
Keywords: glaucoma; machine learning; prediction; risk index.