Retinal nerve fiber layer analysis via Markov random fields texture modelling

The texture analysis of the retinal nerve fiber layer (RNFL) in colour fundus images is a promising tool for early glaucoma diagnosis. This paper describes model-based method for detection of changes in the RNFL. The method utilizes Gaussian Markov random fields (GMRF) and the least-square error (LSE) estimate for the local RNFL texture modelling. The model parameters are used as a texture features and non-linear classifier based on the Bayesian rule is used for classification of healthy and glaucomatous RNFL tissue. The proposed features are tested in the sense of classification errors and also they are applied for segmentation of RNFL defects in a high-resolution colour fundus-camera images. The results are also compared with the Optical Coherence Tomography images regarded as a gold standard for our application due to the possibility of quantitative RNFL thickness measurement.