Purpose:
To implement and evaluate an algorithm that achieves accurate segmentation of the Optic Nerve Head (ONH) in Optical Coherence Tomography (OCT) images of the retina. The geometrical features of the ONH differ between healthy subjects and glaucoma patients and can be useful indicators for early glaucoma detection. This makes the importance of providing robust automated techniques for the identification of such geometrical structures within the retina obvious.
Methods:
We acquired the volume OCT scans with a Spectralis HRA+OCT (Heidelberg Engineering, Heidelberg, Germany), that provides a spatial resolution of approximately 10 μm in the transversal direction. The data consists of 10 volume scans (7 glaucoma patients and 3 normal subjects). As a preprocessing step we resampled the data to an equidistant pixel spacing. Then, a recent implementation of the level-sets method (Li, 2010) was modified and applied on the enface images (a summation along the A-Scans) of the available datasets. Afterwards, morphological erosion was performed in order to avoid the undesired detection of the blood vessels that are located adjacently to the ONH. The automatically extracted results were compared with those determined manually after detailed examination of every B-Scan and the corresponding Scanning Laser Ophthalmoscope (SLO) image. The manual segmentation served as our gold standard in order to have a quantitative evaluation of the results.
Results:
The automated segmentation algorithm provides a sensitivity of 0.8533±0.0972 and a specificity of 0.9853±0.0195 in average on our dataset.
Conclusions:
The proposed algorithm provides accurate segmentation of the ONH in OCT images and can find important applications in ophthalmology (e.g. automated glaucoma detection from geometrical features of the ONH).
Keywords: image processing • optic disc • shape and contour