Development of a retinal pathology classification model using transfer learning and open datasets

The retina is a multilayer structure that perceives and transmits visual stimuli to the brain via the optic nerve; however, its complex structure makes it vulnerable to damage. Optical coherence tomography (OCT) is the gold standard for diagnosing retinal pathologies. This study aimed to develop a deep learning model for the classification of retinal pathologies based on OCT data. The study design included the use of transfer learning and training a model with a ResNet-50 architecture on three open datasets containing over 110.000 OCT images and 10 retinal condition classes, including normal and 10 pathological conditions. As a result, the model demonstrated a high classification accuracy of 95 %. Thus, transfer learning and significant dataset expansion provide high classification model accuracy. The study also highlights that open access to data significantly impacts the development of artificial intelligence technologies in healthcare.