Rijul Saurabh Soans¹ (), Susana T.L. Chung¹; ¹University of California, Berkeley

Fixation stability assessment is important for understanding oculomotor function in many eye disorders. It can be quantified using eyetrackers, microperimeters and scanning laser ophthalmoscopes. These devices are not always available in typical clinical settings, whereas optical coherence tomography (OCT) is more accessible. Can the OCT be used to quantify fixation stability given that it does not collect a lot of “samples” in a single scan? We developed a computer vision framework to test the feasibility of quantifying fixation stability from OCT volume scans. The framework reconstructs scan paths by cross-correlating en-face projections from individual B-scans, with the infrared confocal scanning laser ophthalmoscopy reference image. Planned scan positions were extracted from OCT metadata to calculate deviations from intended acquisition paths. En-face profiles were generated by projecting intensity values from the outer retinal layers of each B-scan. The framework employs multi-orientation Gabor filtering for vessel enhancement, followed by template matching using normalized cross-correlation. It uses adaptive warping to account for compression artifacts caused by eye movements during OCT acquisition. Detected positions were refined using Hampel outlier detection and spline interpolation. To validate the framework, we compared fixation stability using this framework with that derived from a 10-second fixation task using a microperimeter from the right eye of 30 observers with normal vision. Fixation stability was quantified as the 68% isoline area (ISOA) of gaze distributions in both modalities. In general, ISOAoct was higher than ISOAfixation, by an average of 0.21 deg² (statistically different from 0, p<0.001), but there was a significant correlation between the two values (Pearson ρ=0.55, p=0.002). The limits of agreement based on Bland-Altman analysis were: upper=0.57 deg²; lower=-0.14 deg². These results demonstrate that it is feasible to quantify fixation stability from routine OCT scans without additional hardware or exam time, but OCT may overestimate the absolute fixation stability values.