The Improvement of Extraction Retinal Pigment Epithelium (RPE) Layer in Optical Coherence Tomography (OCT) Images by Using Piecewise Dynamic Programming Method in Patients with Persistent Epithelial Defect (PED)

Age-related macular degeneration (AMD) is one of the disorders in the retina that causes central vision disorders. Optical coherence tomography (OCT) images are used to diagnose this disease, and given the changes in the disease caused by elevations in the retinal pigment epithelium (RPE) layer of the retina, the complication is diagnosed.

In the proposed method, the starting point of the RPE layer was labeled by the user on a limited number of OCT slides to eliminate the possibility of mistaking other layers such as the retinal nerve fiber layer (RNFL). The graph-based algorithm was then applied to the image in low width parts; an algorithm was used to maintain the continuity of the parts to eventually estimate the location of the RPE layer. With a similar algorithm, the bruch layer was also located at each scan, and by estimating the distance of the two layers, PED elevations were identified.

The proposed method was evaluated on three datasets with 35, 15, and 10 patients. Compared to the graph-based method in the first, second, and third datasets, respectively, the unsigned error in the RPE layer improved from 4.3392 to 2.7827, 3.3340 to 2.1623, and 6.4842 to 2.3924 pixels, and the bruch layer improved from 5.7576 to 4.8473, 4.3353 to 2.6023, and 6.67 to 2.5446 pixels, respectively.

The proposed method is valid in PED images, and can be effective in diagnosing AMD.