Local optic axis estimation under noise in polarization sensitive optical coherence tomography

Polarization-Sensitive Optical Coherence Tomography (PS-OCT) extends standard OCT by measuring the polarization state of the returned signal, providing depth-resolved maps of optical retardance and optic-axis orientation. This sensitivity to microstructural anisotropy makes PS-OCT powerful for brain imaging. However, the measured optic-axis at any depth is affected by all overlying layers. To recover local orientation, previous methods model each top layer as a pure retarder and iteratively cancel its effect by inverting rotation matrices or tracing Stokes-vector trajectories on the Poincaré sphere. Local optic axis recovery is challenging due to noise, speckle, and phase wrapping. To improve robustness, we introduce super-pixel clustering that groups pixels by polarization and spatial proximity. We compute cluster-level mean orientations and sum retardance as centroids to guide reconstruction, reducing noise while preserving continuity. Using the super-pixel approach, we improved upon the two standard methods and validated our method on simulated single- and double-layer models and applied it to ex vivo brain tissue, achieving robust noise suppression and structural resolution.