RESUMEN
SIGNIFICANCE: This study evaluates the reliability and validity of an automatic method of the external and internal limbal points identification from anterior segment optical coherence tomography (OCT) images in comparison with manual delineation. PURPOSE: The purpose of this work was to evaluate the repeatability and precision of a previously proposed automatic method of external and internal limbal points identification and to compare them with the manual delineation by experienced clinicians in terms of limbus diameter. METHODS: Optical coherence tomography tomograms obtained for 12 healthy volunteers without a history of eye diseases were analyzed. Fifteen OCT tomograms were captured for each patient. For all the images, the external and internal limbal points were determined using both the automatic and manual methods. The external and internal limbus diameters were used as the comparative parameter between the methods under consideration. The statistical analysis included mean, standard deviation, the Passing-Bablok regression, and the Pearson correlation coefficient. RESULTS: A strong linear dependence between the automatic and manual methods was identified. While compared with the subjective estimates from clinicians, the automatic technique overestimated the external limbus diameter (bias equals 0.21 mm for optometrist and 0.23 mm for ophthalmologist) and slightly underestimated the internal limbus diameter (bias equals 0.13 mm for optometrist and 0.04 mm for ophthalmologist). The automatic method showed significantly better repeatability than the manual method in the case of external limbal points identification and comparably high repeatability for internal limbal points recognition. CONCLUSIONS: Because of high precision and excellent repeatability, the automatic method of limbal points identification may be successfully used for estimation of the dynamic changes in the geometry of the anterior segment of the eye, where the large number of captured OCT images needs to be processed automatically with high precision.