Analysis of retinal nerve fiber layer via Markov random fields in color fundus images

Since images of ocular fundus are commonly available in ophthalmic practice, automatic assessment of the retinal nerve fiber layer (RNFL) can be useful for early diagnosis of glaucoma. This contribution presents a texture analysis method for the description of RNFL status and proposes appropriate textural features for medical diagnosis. The method uses Gaussian Markov random fields to model visual appearance of the texture. The results show that the proposed textural features can be used for reliable classification of healthy RNFL areas and areas with focal wedge-shaped RNFL loss. Experiments show a close correlation of the proposed features with the real measurement of the RNFL thickness by optical coherence tomography (OCT). Quantitative comparison of results from the two modalities - fundus camera and OCT - successfully validates relevancy of the proposed features.