Some eye diseases exhibit abnormalities or atrophy of the retina, which can impair vision and cause blindness. Certain other diseases traditionally thought to be exclusively associated with the brain are now known to be associated with changes or differences in the retina, which can also impair vision. Optical coherence tomography (OCT), an imaging device that is enabling modern in-vivo high-resolution studies of the retina, is rapidly becoming a key imaging technology both in ophthalmology, where clinical diagnoses and disease monitoring are experiencing a dramatic technical revolution, and also in neurology, where noninvasive retinal markers of brain disease are beginning to emerge. A key limitation in the current technology is the lack of detailed, accurate, and fully-automated analysis of the retinal layers that are observed throughout the retina in a typical three-dimensional spectral domain (SD)-OCT exam. Although tools are beginning to appear in association with commercial devices, they are currently lacking in sophistication and are not validated or standardized across manufacturers. This deficiency is problematic for both clinical and scientific application of SD-OCT since the measurements are not of known precision and are therefore not easily compared both in the same subject longitudinally and in population cross sections. The proposed research takes advantage of state-of-the-art algorithms that are emerging in both computer vision and medical imaging and will apply, adapt, and extend them to the specific application of layer segmentation in three-dimensional retinal SD-OCT. Advanced tools for population averaging in a normalized space will also be developed specifically for retinal SD-OCT data. The methods will be validated against manual segmentation and normalization on both normal subjects and patients experiencing retinal degeneration largely associated with thinning of selective layers of the retina. Scans of patients with diseases causing significant retinopathy such as cysts, detachment, or scarring will be evaluated, but further development beyond the present research is expected to be needed to properly segment and quantify such cases. The developed software will be developed within the open-source Java Image Science Toolkit (JIST) framework and released as open source software.
The project will develop software for the image analysis of optical coherence tomography (OCT) scans of the retina. Segmentation and quantification of the characteristics of nerve layers within the retina is expected to be of great use on both ophthalmology and neurology, where the retina can be a sensitive indicator of both vision and neurological problems or diseases. Open source software written in a highly portable language will be made available to the research community at the conclusion of the research grant.
|Lang, Andrew; Carass, Aaron; Sotirchos, Elias et al. (2013) Segmentation of retinal OCT images using a random forest classifier. Proc SPIE Int Soc Opt Eng 8669:|