Potenzial von Methoden der künstlichen Intelligenz für die Qualitätssicherung

Opportunistic deep learning of retinal photographs: the window to the body revisited

“…Owing to the general lack of data sharing in ophthalmic research, most work in this field cannot offer such extensive external validation, and open access to validation databases remains one of the major unmet needs in artificial intelligence in retinal imaging. 5 Chronic kidney disease is a condition with a low prevalence, 6 which is reflected in some of the datasets used by the authors. In a low prevalence setting, the often-used outcome measure of the area under the receiver operating characteristics curve can overestimate the performance of the model, and the area under the precision-recall curve could be a more accurate outcome measure, 7 as the authors acknowledge.…”