Segmentation of human retinal layers from optical coherence tomography scans

Nathan M Hammes, Purdue University


An algorithm was developed in to efficiently segment the inner-limiting membrane (ILM) and retinal pigmented epithelium (RPE) from spectral domain-optical coherence tomography image volumes. A deformable model framework is described and implemented in which free-form deformations (FFD) are used to direct two deformable sheets to the two high-contrast layers of interest. Improved accuracy in determining retinal thickness (the distance between the ILM and the RPE) is demonstrated against the commercial state-of-the-art Spectralis OCT native segmentation software. A novel adaptation of the highest confidence first (HCF) algorithm is utilized to improve upon the initial results. The proposed adaptation of HCF provides distance-based clique potentials and an efficient solution to layer-based segmentation, reducing a 3D problem to 2D inference.




Tsechpenakis, Purdue University.

Subject Area

Ophthalmology|Computer science

Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server