Visible light optical coherence tomography (vis-OCT) provides a unique tool for imaging both structure and oxygen metabolism in ophthalmology. Working in visible light bandwidth, it suffers from noises due to strong scattering, especially in the blood. This work established the random matrix (RM) description of vis-OCT’s k-space data as ballistic and multiple scattering components. The eigenvalue density of the hybrid RM follows a low-rank biased Marčenko–Pastur law. The ballistic component can thus be separated out using a generalized likelihood ratio test algorithm. The RM-based method was validated by both the Monte Carlo simulation and ex vivo pure blood phantom study. We further demonstrated that the RM-based method could significantly improve the imaging quality in the human fundus, showing more details of the layered structure than current vis-OCT with ∼23.6% increase of signal-to-noise ratio, measuring the blood oxygen value more accurately, and enabling better structure visualization than the traditional method, a 1.6-fold higher contrast-to-noise ratio in raster scan mode. The isolated ballistic component also fits the Beer–Lambert law better, giving more accurate oxygen saturation in arc scan mode. The RM-based method significantly improves the reconstruction quality in 3D and facilitates clinical diagnostics. As a general framework, random matrix description also provides a new separation strategy to estimate the ballistic component in other spectral domain OCT techniques.
Read full abstract