Detection of diabetic retinopathy lesions from spectral retinal images
Turbabin, Fedor (2017)
Diplomityö
Turbabin, Fedor
2017
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe201705236781
https://urn.fi/URN:NBN:fi-fe201705236781
Tiivistelmä
Long-term diabetes may lead to a diabetic retinopathy, an eye disease, which is one of the
most frequent causes of blindness. Its early symptoms can be detected from the photos
of retina, thus, making the investigation into the automatic lesion detection method to be
a valuable undertaking. Hyperspectral images provide additional information about the
characteristics of the imaging target, which may improve accuracy in comparison to sim-
ple color images, but at the same time introduce problems of higher dimensionality. The
purpose of this work is to develop an effective algorithm for detecting diabetic retinopathy
lesions and find out whether hyperspectral images are beneficial for diabetic retinopathy
diagnosis in comparison to colour images. The experiments provide evidence that proper
algorithmic use of spectral images allows to achieve significantly better results.
most frequent causes of blindness. Its early symptoms can be detected from the photos
of retina, thus, making the investigation into the automatic lesion detection method to be
a valuable undertaking. Hyperspectral images provide additional information about the
characteristics of the imaging target, which may improve accuracy in comparison to sim-
ple color images, but at the same time introduce problems of higher dimensionality. The
purpose of this work is to develop an effective algorithm for detecting diabetic retinopathy
lesions and find out whether hyperspectral images are beneficial for diabetic retinopathy
diagnosis in comparison to colour images. The experiments provide evidence that proper
algorithmic use of spectral images allows to achieve significantly better results.