Abstract
Purpose :
Non-invasive imaging of lamina cribrosa’s structure and optical property is important for myopia and glaucoma investigation. This paper describes an automatic segmentation method of lamina beam using a multi-functional Jones matrix optical coherence tomography (JM-OCT).
Methods :
12 eyes of 6 subjects (normal and myopic) were scanned by a custom-made JM-OCT with 1.06-mm probe. The lateral scan range was 3 mm × 3 mm. A single measurement of JM-OCT provides multi-contrast tomographies including OCT, attenuation coefficient, OCTA,, birefringence, and polarization uniformity.
The multi-contrast tomography is processed by a newly developed tissue classification framework consisting of 2 steps: (1) Generation of training dataset for a tissue classifier. The pixels in training images are clustered based on multiple features derived from the multiple contrasts. Tissue labels are manually assigned to each cluster. This process uses only a minimum knowledge of human experts, but generates a training dataset to train machine learning based tissue classifier. (2) Supervised tissue classification. A random forest classifier is trained by using the generated training dataset. Finally, each pixel of test images are classified into tissue types. Six JM-OCT volumes were used in this study
En-face birefringence maps of the lamina beam is created by using the segmented lamina beam.
Results :
Fig. 1(a) shows the OCT cross-section of a 32 years old female (IOP = 14 mm Hg; axial length = 23.7 mm). Fig. 1(b) shows segmented tissue labels obtained from the classifier. Unlike, conventional lamina segmentation, lamina beam is distinguished from other lamina tissue.
Figure 2 shows the en face intensity map (a) and the segmented lamina beam (b). By overlaying the lamina beam (red) on the intensity (c), the rationality of the segmentation is demonstrated. Fig. 2(d) shows the birefringence map of the lamina beam and sector-wise mean birefringence map (f) with sectors as shown in Fig. 2(e). The nasal sector shows higher birefringence than other sectors.
Conclusions :
Automated segmentation of lamina beam based on multi-functional JM-OCT is demonstrated. It provides birefringence map which is sensitive only for lamina beam.
This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.