TY - JOUR T1 - Challenges in estimating the accuracy of imaging-based detection methods for glaucomatous progression JF - British Journal of Ophthalmology JO - Br J Ophthalmol SP - 385 LP - 386 DO - 10.1136/bjophthalmol-2012-302841 VL - 97 IS - 4 AU - Koenraad A Vermeer AU - Hans G Lemij Y1 - 2013/04/01 UR - http://bjo.bmj.com/content/97/4/385.abstract N2 - The accuracy of glaucoma diagnosis by imaging devices that are commonly used in clinical practice, such as the GDx, the Heidelberg retinal tomograph and optical coherence tomography (OCT) is at a very high level, often surpassing the accuracy of trained ophthalmologists.1 Because glaucoma specialists are considered to be the gold standard that imaging devices are tested against, the interobserver and intraobserver agreement of these specialists impose a limit on the possibilities to further improve the diagnostic accuracy of these imaging devices. Even if these devices perform better than clinicians, our estimates of their accuracy will not reflect this because any deviation from these clinicians’ judgment will be flagged as erroneous. These problems are not unique to glaucoma diagnosis: similar limits also hinder further improvement of automated detection of diabetic retinopathy.2 Despite these problems regarding diagnostic accuracy, imaging devices may have various other advantages, such as better reproducibility, ease-of-use, speed of diagnosis and documentation, price and versatility, and their performance in these aspects is enhanced continuously, benefiting both clinicians and patients.The next challenge, which has high clinical relevance for a chronic, progressive disease, is monitoring glaucoma over time. Compared with glaucoma diagnosis, validation of glaucomatous progression detection by imaging devices is much more complicated: a reference standard for progression is, unfortunately, not available. Current clinical guidelines only define progression in general terms, as a progressive process of functional and/or structural damage,3 ,4 demonstrating that … ER -