Purpose To evaluate the performance of various tests with automated and subjective evaluation for primary diagnostics of glaucoma in a normal clinical setting. Methods Subjects referred because of suspicion of glaucoma were recruited. All subjects had full ophthalmologic evaluation with stereophotography of the optic nerve head (ONH), red-free retinal nerve fiber layer (RNFL) photography, scanning laser polarimetry (SLP), optical coherence tomography (OCT), and standard automated perimetry (SAP). Analysis of all results was made jointly by 5 glaucoma specialists, with a consensus for each eye as healthy, glaucomatous, or glaucoma suspect. Results from each method were evaluated against this consensus, followed by a subjective evaluation of each method by a glaucoma specialist, comparing this with automated classification by the devices. Results Of the 101 subjects and 202 eyes examined, 23 eyes were by consensus glaucomatous, and 23 were glaucoma suspect. Sensitivity was best with ONH photos and worst with SAP, while SLP had a better sensitivity but poorer specificity than did RNFL photographs and OCT. Subjective evaluation of SLP, OCT, and SAP data gave better sensitivity and specificity than did classification by numeric values from the devices only, with OCT performing better in automatic classification than did SLP or SAP. Conclusions None of the current methods is superior to others in diagnosing glaucoma, and the accuracy of automated tests was better when subjectively evaluated by an experienced ophthalmologist. Diagnosis of glaucoma should be based on a combination of test results interpreted by a clinician.
TRk-CNN: Transferable Ranking-CNN for image classification of glaucoma, glaucoma suspect, and normal eyes
