Exudate detection in eye digital fundus images using neural networks

Albert Clos, Gerald Schaefer, Lars Nolle

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

In this paper we show that a properly trained neural network can be used to detect exudate lesions associated with diabetic retinopathy in digital fundus images of the eye. Image pre-processing is crucial for good performance and includes colour normalization to reduce the variability of colours found in fundus images as well as principal component analysis for dimensionality reduction. Furthermore, blood vessels and the optic disc are segmented out of the retina images in order to avoid confusion with exudate regions. The remaining image regions are used to train a backpropagation neural network to distinguish exudate from non-exudate areas. Experimental results based on a image set with known exudate locations confirm the efficacy of our approach.

Original languageEnglish
Title of host publication13th International Conference on Soft Computing: Evolutionary Computation, Genetic Programming, Fuzzy Logic, Rough Sets, Neural Networks, Fractals, Bayesian Methods, MENDEL 2007
Pages122-127
Number of pages6
Volume2007-January
Publication statusPublished - 2007

Fingerprint Dive into the research topics of 'Exudate detection in eye digital fundus images using neural networks'. Together they form a unique fingerprint.

  • Cite this

    Clos, A., Schaefer, G., & Nolle, L. (2007). Exudate detection in eye digital fundus images using neural networks. In 13th International Conference on Soft Computing: Evolutionary Computation, Genetic Programming, Fuzzy Logic, Rough Sets, Neural Networks, Fractals, Bayesian Methods, MENDEL 2007 (Vol. 2007-January, pp. 122-127)