Purpose:
Fast, accurate, and objective detection of disease imaging biomarkers is crucial for the study and diagnosis of ophthalmic diseases. Chiu et al. previously developed a generalized framework based on graph theory and dynamic programming (GTDP) to segment layered structures in ocular SDOCT images. However, aside from layered structures, there is a need for automatically segmenting closed-contour features in ophthalmic images such as cysts. In this work, we extend the application of our GTDP technique to segment cysts seen on retinal SDOCT images.
Methods:
Our method is an extension of the GTDP technique previously introduced for segmenting layered (line) structures. Our algorithm is based on the observation that with the appropriate transform, closed-contour features in the Cartesian domain can be represented as lines in a pseudo-polar domain. This novel transform allows us to use our previous GTDP method to segment the object as if it were a layered structure. We have utilized this method to segment retinal cysts on SDOCT images of pediatric and diabetic macular edema patients.
Results:
The fully automated segmentation efficiently isolated large and small cystoid structures with irregular contours in complex pathologies. In Fig.1, we demonstrate the effectiveness of this method in retinal SDOCT images of a) a premature infant eye with macular edema and b) an eye with diabetic macular edema.
Conclusions:
The preliminary results in Fig. 1 using our automatic GTDP technique for cyst segmentation demonstrate the extensibility of our framework to segment not only layered, but also closed-contour structures, in ocular images. This is highly encouraging for reducing the time and manpower required to segment cysts and other closed features in ophthalmic studies.
Keywords: image processing • retina