Optical coherence tomography (OCT) has emerged as an important noninvasive imaging modality in the field of ophthalmology, especially with the introduction of spectral domain OCT (SD-OCT) [1
]. Due to its greater imaging speed and resolution, the layers of the outer retina can be identified on SDOCT scans and the effects of outer retinal diseases, such as retinitis pigmentosa (RP), can be quantified.
RP is a progressive retinal disease that affects the receptors resulting in a severe loss of vision. The structural damage of the outer retina seen on OCT scans has been quantified and used for guidance in therapeutic interventions [5
]. With SD-OCT, the thickness of the outer segments (OS) of the receptors can be measured and regions of thinning can be identified as well [9
]. Thinning of the OS layer precedes changes in other receptor layers such as the outer nuclear layer [11
]. Recently, Hood et al. argued that the IS/OS contour, the locus of points at which one of the outer retinal borders, the IS/OS line disappears and the OS thickness goes to zero, shows promise as a clinical measure of disease progression [12
]. Consequently, it is important to develop and validate automated procedures for segmenting the layers of the outer retina and quantifying the thickness of the OS layer.
Manual segmentation has been employed in previous OCT studies with RP patients [9
]. However, manual segmentation is a time consuming process and may exhibit subjective variations among different segmentation experts. To provide objective measures and to extend the research study to a larger scale, it is highly desirable to develop a segmentation and quantification methodology that is both reliable and automated. Recently, we have developed a fully automated segmentation algorithm based on gradient information in dual scales, and it showed high accuracy and repeatability in the segmentation of three dimensional (3D) OCT volume scans of normal subjects [13
]. Several other algorithms in the literature also have demonstrated automated retinal boundary segmentation based on intensity variation, gradient information, or textural features [14
]. Among these algorithms, some have shown high accuracy of detecting multiple inner and outer retinal boundaries in normal subjects [18
]. Yet the morphological changes in RP diseased eyes that leads to the loss of the IS/OS boundary and thinning of the ONL have remained a technical challenge for the automated segmentation of the OCT images.
The purpose here is to describe modifications to our automated segmentation algorithm [13
] to allow delineation of the inner limiting membrane (ILM) and three outer retinal boundaries in RP OCT images and to provide a validation of this algorithm by comparing the results to those obtained with manual segmentation. In addition, an automated method is proposed to quantify the OS thickness loss in a 3D volume scan and to identify the IS/OS contour in SDOCT images from RP patients.