Early glaucoma screening Decision Tree using in-depth clinical analysis of the optic nerve head

To assess the performance of an end-to-end computer vision algorithm for the early screening and diagnosis of glaucoma from fundus photographs. Development and evaluation of a computer vision approach for the extraction of physiologic measurements from the optic nerve head (ONH) and the detection of glaucoma. 101 fundus photographs were included for the development and validation of the screening algorithm. We automatically detect and extract the ONH within the retina, and compute two methods for segmenting of the optic cup (OC) and the optic disc (OD). Referable ophthalmic features including the cup-to-disc ratio (CDR), the Inferior- Superior-Nasal-Temporal (ISNT) rule and the neuroretinal rim (NRR) area are exploited, to quantify the structural changes within the ONH. Then, we elaborate a clinical protocol to provide trustworthy, easily interpretable glaucoma screening process, and indications about the stages of disease spreading.Proposed approach allows to accurately estimate the clinical measurements with 0.11 ± 0.08 to 0.15 ± 0.12 mean-std error rates on CDRs, ISNT sectors and NRR area. Also, outstanding glaucoma screening performance was achieved with 93.07% (95% confidence interval (CI), 86.24% to 97.17%) accuracy, 98.57% (95% CI, 92.30% to 99.96%) sensitivity and 80.65% (95% CI, 62.53% to 92.55%) specificity rates, competing with the state-of-the-art. We formulated a clinical protocol aiming at evaluating the structural changes occurring within the ONH, for assessing the different spreading stages of the neuropathy. Excellent performance has been obtained on glaucoma screening process, allowing to integrate such algorithm into computer-aided diagnosis systems and support ophthalmologists.