Purchase this article with an account.
N. Cassoux, D. Brahmi, A. Giron, P. LeHoang, B. Fertil; Automatic Detection of Main Fundus Structures and Retinal Lesions on Fluorescein Angiography by Image Analysis Methods Using a Learning by Sample Approach . Invest. Ophthalmol. Vis. Sci. 2003;44(13):3641.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Purpose: To recognise automatically the main components of the fundus and CMV retinitis on fluorescein angiography images Methods: The region of interests (RoI) were defined as the optic disc, blood vessels and CMV retinitis areas. The RoIs in each image were identified by experienced ophthalmologits for comparison with computerized methods. 150 images were collected and preprocessed. Images were sampled in little sub images. The RoIs were located using a multilayer perceptron neural network(NN). The classification method using logistic regression(LR) was used as control. The inputs were derived from a principal component analysis of the image and texture features derived from the grey level histogram. Results: The optic nerve, vessels, and CMV retinitis areas were localized in respectively 77% with LR, 92% with NN; 80.50% RL, 80.38% NN; 75% RL, 85% NN.The mean errors for the CMV retinitis segmentation was 9.6% with NN.> Conclusions: In RoIs were accurately detected. This study shows that learning by sample classification methods can be efficient to texture recognition on fluorescein angiograms.
This PDF is available to Subscribers Only