In cooperation with Faculty hospital of Ostrava – Ophthalmology we aret working on development of methods that have a potential to recognize automatically areas of retinal lesions from retinal pictures made by system RetCam 3. These automatic methods based on artificial intelligence have potential to improve quality and objectivity of diagnostics of retinopathy of prematurity (ROP) in context of analysis of retinal lesions.

In comparison with highly used method of machine learning that requires a training based on huge amount of data, our method uses so called online semisupervision classification. It means that the method learns to recognize retinal lesions directly during the running on the base of specific signs of retinal lesions without any special training that is time-consuming a lot.

The outcome of this model is detected retinal lesions with options of their quantification in time it enables monitoring of progress of this illness in time. At the same time the system can detect an optical disk that is used for referential object when we observe the progress of retinal lesions in time.