Abstract
Purpose :
Optical Coherence Tomography is becoming an indispensable tool in imaging the eye. The retinal tomography is able to provide pathology localization that is more accurate anatomically enabling the diagnostician to ascertain the provenance of the pathology and a more deterministic prognosis of the progress of the disease compared to Colored Fundus Cameras. Quick anatomical segmentation of layers of the retina enables the ophthalmologist to explain the issues in an interactive environment to the patient
Methods :
Current state of the art systems used in various industry standard medical equipment use classical graph cuts based methods and deep learning methods. We have designed an efficient system that has optimised the speed of generating the segmentations, analysis and visualisation on a commodity nVidia GPU that enables the computational pipeline to visualise these segmentations. All computations have been parallelized and moved to the GPU using CUDA.
Results :
The resulting optimised algorithm is able to generate the thickness of the profile of layers of the retina under a second after the acquisition of the OCT volume. This enables the circular tomography and its segmentation to be generated nearly instantaneously after the 3D scan of the optic disc. The interactive radial and circular tomograms helps the patients understand the scan.
Conclusions :
Providing the delineation of retinal layers quickly would enable an ophthalmologist to explain with more clarity about the nature of the diseases even as simple as thinning of the retinal layers, disruption in layers and comparing normative databases immediately to the patient and building accurate normative databases offline quickly.
This abstract was presented at the 2024 ARVO Imaging in the Eye Conference, held in Seattle, WA, May 4, 2024.