Automatic Anisotropic Diffusion Filtering and Graph-search Segmentation of Macular Spectral-domain Optical Coherence Tomographic (SD-OCT) Images.

BACKGROUND

Optical Coherence Tomography (OCT) is a non-invasive medical imaging technique that provides high-resolution cross-sectional images of the retina. There is a need to develop algorithms for obtaining quantitative and qualitative information about the retina which are essential for assessing and managing eye conditions.

METHODS

This work emphasizes on an automated image processing algorithm for segmenting retinal layers. It involves preprocessing of the acquired retinal SD-OCT image (B-scan) using the proposed automatic Anisotropic diffusion filter, followed with contrast stretching to suppress intrinsic speckle noise without blurring structural edges. Graph search segmentation using Dijkstra algorithm with a combination of threshold and axial gradient as the cost function is used to segment the retinal layer boundaries.

RESULTS

The algorithm was performed and the average thickness of the segmented retina was computed for the 3D retinal scan (128 B-scans) of 8 subjects (4 normal and 4 abnormal) using Early Treatment Diabetic Retinopathy Screening (ETDRS) chart.

CONCLUSION

Segmentation was evaluated using manually segmented B-scan by an Ophthalmologist as ground truth and accuracy was found to be 99.14 ± 0.27%.