ESB(k)=ReFig. 1. To isolate the potential candidate eigenvalues for the ballistic component, we propose an adaptive algorithm based on the generalized likelihood ratio test method.3232. E. Conte, A. De Maio, and G. Ricci, “GLRT-based adaptive detection algorithms for range-spread targets,” IEEE Trans. Signal Process. 49, 1336–1348 (2001). https://doi.org/10.1109/78.928688 In practice, the sphericity test3333. D. N. Lawley, “Tests of significance for the latent roots of covariance and correlation matrices,” Biometrika 43, 128–136 (1956). https://doi.org/10.2307/2333586 is more robust in separating the low-rank eigenvalues of the sample covariance matrix. The proposed algorithm uses the sphericity test metric C(q) [Eq. (8)] to isolate the eigenvalues,C(q)=sq−11P∑n=qPsn.(8)Since the largest eigenvalue (i.e., s1) is always accepted, the algorithm starts from q = 2 to calculate all C(q), q = 2, …, P. Then, the maximum value of C(q) is identified at q0. All eigenvalues with an index less than q0, i.e., Q = q0 − 1, are accepted as low-rank eigenvalues. We used MC simulation to confirm that as the ballistic component increases, the eigenvalues of the ballistic become larger, especially the first few eigenvalues. At the same time, the multiple scatterings always remain very low. The different Q eigenvalues of the hybrid signal accepted by the adaptive algorithm have a low-rank property to recover the ballistic signal whose numbers of chosen eigenvalues are 4, 4, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, and 2 with the ratio of ballistic to multiple scattering increasing from 1 to 14. Finally, the ballistic component matrix ZB can be recovered by ,where Λ̂ is a N × Q diagonal matrix with as its elements.

III. METHOD

Section:

ChooseTop of pageABSTRACTI. INTRODUCTIONII. THEORYIII. METHOD <<IV. RESULTSV. DISCUSSIONSVI. CONCLUSIONSUPPLEMENTARY MATERIALREFERENCES

A. Monte Carlo simulation of vis-OCT

We developed an SD-OCT Monte Carlo (MC) simulation platform for vis-OCT by referring to Malektaji et al.’s work.3434. S. Malektaji, I. T. Lima Jr., M. R. Escobar I., and S. S. Sherif, “Massively parallel simulator of optical coherence tomography of inhomogeneous turbid media,” Comput. Methods Programs Biomed. 150, 97–105 (2017). https://doi.org/10.1016/j.cmpb.2017.08.001 With acceleration of CUDA-GPU (Compute Unified Device Architecture - Graphics Processing Unit), one B-scan of vis-OCT can be obtained in 1 min. A four-layer 3D phantom model of the fundus tissue was used in the MC simulation, as shown in Fig. 3(a). The model layers correspond to the retina, retina pigment epithelium (RPE), choroid, and sclera, respectively.3535. W. Liu, S. Jiao, and H. F. Zhang, “Accuracy of retinal oximetry: A Monte Carlo investigation,” J. Biomed. Opt. 18, 066003 (2013). https://doi.org/10.1117/1.jbo.18.6.066003 A cylindrical blood vessel segment was intentionally placed in the retina layer with a diameter of 110 µm and a wall thickness of 10% of the lumen diameter.3636. M. Ritt and R. E. Schmieder, “Wall-to-lumen ratio of retinal arterioles as a tool to assess vascular changes,” Hypertension 54, 384–387 (2009). https://doi.org/10.1161/hypertensionaha.109.133025 The blood within the vessel is assumed to be optically homogeneous. The optical properties of the phantom, including absorption coefficient μa (cm−1), scattering coefficient μs (cm−1), and anisotropy factor g (a.u.), were referring to Refs. 3737. M. Keijzer, R. R. Richards-Kortum, S. L. Jacques, and M. S. Feld, “Fluorescence spectroscopy of turbid media: Autofluorescence of the human aorta,” Appl. Opt. 28, 4286–4292 (1989). https://doi.org/10.1364/ao.28.004286 and 3838. M. Hammer, A. Roggan, D. Schweitzer, and G. Muller, “Optical properties of ocular fundus tissues-an in vitro study using the double-integrating-sphere technique and inverse Monte Carlo simulation,” Phys. Med. Biol. 40, 963 (1995). https://doi.org/10.1088/0031-9155/40/6/001 (see Table I).

TABLE I. Optical parameters of the phantom.

Mediumμa (cm−1) @560 nmμs (cm−1) @560 nmgnRetina53190.971.37Vessel wall92840.841.37Blood16010280.9721.37RPE93810680.841.38Choroid2249500.941.39Sclera49660.91.39We then launch a 109 photon package through a single-mode fiber at the initial location, simulating a thin beam illumination perpendicular to the surface of the tissue sample. The scattering procedure is traced according to the schematic in Refs. 3939. L. Wang, S. L. Jacques, and L. Zheng, “MCML—Monte Carlo modeling of light transport in multi-layered tissues,” Comput. Methods Programs Biomed. 47, 131–146 (1995). https://doi.org/10.1016/0169-2607(95)01640-f and 4040. G. Yao and L. V. Wang, “Monte Carlo simulation of an optical coherence tomography signal in homogeneous turbid media,” Phys. Med. Biol. 44, 2307 (1999). https://doi.org/10.1088/0031-9155/44/9/316. The same fiber collects the back-scattered photons with a maximal collecting radius of 5 μm and a maximal accepted angle of 10°. In MC simulation, the photons experiencing ballistic and multiple scattering paths are separated according to the path length of the collected photons at the depth z can be sorted into two types: (1) ballistic ones satisfied Eq. (10) and (2) multiply scattered photons violate the criteria of Eq. (10),where lp is the path length of a photon with respect to the OCT’s zero delay.

B. Ex vivo phantom preparation for vis-OCT imaging

We had an ex vivo tissue phantom experiment to validate the ballistic component estimation using RM based method. The ex vivo bovine blood (Quadfive, Ryegate, MT) was used to prepare an oxygenated (sO2 ≈ 100%) blood sample with a hematocrit of 45%. We exposed blood to a constant stream of pure oxygen with continuous stirring to oxygenate it and verified the blood oxygen level with a blood-gas analyzer (Rapidlab 248, Siemens Healthcare Diagnostics, Malvern, PA).

A capillary glass tube with an inner diameter of 200 µm was used to mimic the blood vessel and flushed with phosphate-buffered saline (PBS) and heparin solution to prevent clotting or sedimentation. The oxygenated blood sample was loaded into the prepared glass capillary tube through a syringe pump [0.03 mm (s)−1]. The tube was embedded in the middle of a homemade plastic well. Immersion oil (refractive index = 1.52) was added to the well at ∼500 µm depth.

We used a prototype vis-OCT system to acquire the B-scan data crossing the tube center, which consisted of 512 A-lines. The technical configurations of the system are shown in Fig. 2. Briefly, it was built with a supercontinuum light source (SuperK EXTREME; NKT Photonics, Birkerød, Denmark). The light was first filtered by a short-pass filter (DMSP650T, Thorlabs) to eliminate the light with a wavelength longer than 650 nm. Then, we applied the band-pass filter (FF01-560, Semrock) to separate the light for vis-OCT imaging. Filtered light was sent to a 30:70 fiber coupler (Gould Fiber Optics, Millersville, MD). The system used telescopic optics in the sample arm, forming a 7 µm focused spot size. A pair of galvanometer mirrors (Nutfield Technology, Londonderry) scanned the focused spot on the retina at a 25 kHz A-line rate. The exposure time of scanning was 37 µs. The spectrometer (Blizzard SR; Optical Health, Evanston, IL) covering a spectrum range of 506–621 nm detects the interferogram signals with 37 µs exposure time, allowing 3 µs for data readout by LabVIEW 2021 (National Instrument).

C. In vivo OCT imaging of normal human fundus

We applied the proposed RM-based method to clinical imaging of normal human fundus using a commercial vis-OCT system (Opticent Health, Evanston, IL). The experiment was approved by Northwestern University Institutional Review Board (IRB) and adhered to the Declaration of Helsinki. Two healthy volunteers (Subject A and B) were recruited during their routine clinical visits to the Ophthalmology Department at the Northwestern Memorial Hospital. The volunteers’ eyes were imaged by arc and raster scanning modes in one acquisition trial. Each arc scan is a 120° segment in the full circular scan with 8192 A-lines per scan. Sixteen B-scans were acquired in each arc scan for sO2 measurement. Each raster scan contains 64 B-scans (512 A-lines) in an area of 4.8 × 4.8 mm2. Arc scanning takes 5.25 s, and raster scan takes 1.31 s in one acquisition trial. The axial and lateral resolutions of the system were about 1 and 10 µm in the retina, respectively.

D. vis-OCT oximetry

In this study, we window the broad spectral interferogram in k-space and then take the short-time Fourier transform (STFT) to reconstruct a low axial-resolution A-line signal.99. J. Yi, Q. Wei, W. Liu, V. Backman, and H. F. Zhang, “Visible-light optical coherence tomography for retinal oximetry,” Opt. Lett. 38, 1796–1798 (2013). https://doi.org/10.1364/ol.38.001796 Twenty subband spectra ranging from 525 to 582 nm are extracted using a Gaussian window. The full width at half maximum (FWHM) bandwidth of the Gaussian window is 13 nm at 553 nm (0.35 µm−1 in k-space), resulting in the axial resolution of ∼6.5 μm.4141. B. T. Soetikno, L. Beckmann, X. Zhang, A. A. Fawzi, and H. F. Zhang, “Visible-light optical coherence tomography oximetry based on circumpapillary scan and graph-search segmentation,” Biomed. Opt. Express 9, 3640–3652 (2018). https://doi.org/10.1364/boe.9.003640 The Gaussian window also avoids spectral leakage in the STFT. In visible light, a red blood cell has strong absorption contrast than other tissue, so we can use the Beer–Lambert law (BLL) model with different subband spectra to estimate sO2 bylogI(λ,z)I0(λ,z)=−zsO2×(μaHbO2(λ)+μsHbO2(λ))+(1−sO2)×(μaHb(λ)+μsHb(λ)),(11)where μaHbO2 and μsHbO2 are absorption and scattering coefficients of oxyhemoglobin, and μaHb and μsHb are absorption and scattering coefficients of deoxyhemoglobin.

E. Metrics for evaluation

To quantitatively evaluate the improvement of the RM-based method in clinical imaging, we applied two evaluation scores commonly used in OCT imaging, i.e., the contrast-to-noise ratio [CNR, Eq. (12)] and peak signal-to-noise ratio [SNR, Eq. (13)],CNR=μs−μbσs2+σb2,(12)where σs and σb are the standard deviations of the region of interest (ROI) and background regions, while μs and μb are the mean of the ROI and background regions,where Means is the mean value of the ROI region.

IV. RESULTS

Section:

ChooseTop of pageABSTRACTI. INTRODUCTIONII. THEORYIII. METHODIV. RESULTS <<V. DISCUSSIONSVI. CONCLUSIONSUPPLEMENTARY MATERIALREFERENCES

A. Validation of RM-based method using MC simulation

Figure 3(a) shows the B-scan images of the isolated ballistic and multiple-scattering components inside the boxed area of the fundus phantom by MC simulation. The ballistic component provides precise structure layers and locations, including the vessel wall and RPE layer. The influences of multiple-scattering components [Fig. 3(c)] widely exist, and there is evident strong scattering in the blood vessel region and RPE layer. Irregular noise from multiple scatterings (red arrows) was accumulated due to the inhomogeneity of tissue, which blurs spatial localization and degrades the quality of total vis-OCT signals. Figure 3(c) shows the ballistic component (blue), multiple-scattering (red), and hybrid component (purple) along the center A-line in Fig. 3(a), respectively. The ballistic signal shows the structural details of the vessel wall and layer boundaries, and there is a more remarkable enhanced edge in the vessel wall and RPE. It was noted that the ballistic component decreased more remarkably with the depth in the vessel than the multiple-scattering component and the intensity degraded in the B-scan. The total hybrid signal was ruined by multiple scattering with depth increasing, which makes the hybrid OCT signal lose its SNR.We further use eigenvalues () to separate ballistic signals. It is noted that the eigenvalues of multiple scattering [Fig. 3(b)] follow the MP law that allows us to use the RM-based method to separate ballistic and multiple scattering. The proposed adaptive algorithm isolated the corresponding ballistic eigenvalues from the hybrid RM eigenvalues following low-rank features shown in Fig. 1, which led to reconstructing the RM-based signal. The green curve in Fig. 3(c) shows the RM-based reconstruction using the isolated eigenvalues from hybrid RM to recover the ballistic signal. In the retina, the vessel region always produces shadow effects due to the strong absorption of blood; thus RPE boundary can hardly be seen in hybrid and multiple scattering component signals (purple and red curves) A-lines. The ballistic signal can provide clear but weak signals around the RPE boundary (orange arrow pointing to the blue curve). The RM-based reconstructed signal successfully recovers the RPE boundary signal with sufficient SNR (orange arrow pointing to the green curve). Meanwhile, under the RPE layer, the reconstructed signal by the RM-based method resists such high intensity and attenuates slowly, implying its potential application for recovering the tissue features below the RPE.4242. M. G. Giacomelli and A. Wax, “Imaging beyond the ballistic limit in coherence imaging using multiply scattered light,” Opt. Express 19, 4268–4279 (2011). https://doi.org/10.1364/oe.19.004268

B. Ex vivo phantom experiment for validation of RM-based method in measuring sO2

Figure 4(a) shows the ex vivo experiment setup and the cross section of the glass capillary with sO2 = 1.0 blood sample. Figure 4(b) shows blood sO2 of the original (hybrid) signal and RM-based reconstruction signal obtained in the ex vivo phantom experiment [Fig. 4(a)]. Equation (11) is applied to fit the blood sO2 demonstrated in Sec. III. Although both ways demonstrate an exponential decay along the depth, the RM-based method enhanced the 10 μm measurement depth and it revealed a greater intensity loss with depth for sO2 measurement [orange dots and diamonds in Fig. 4(b)], implying restraining the signal lifting due to multiple scattering. The RM-based method provides better accuracy in sO2 measurement than the traditional way, especially for larger or deeper vessels.

C. RM-based method improves the imaging SNR and CNR and reveals more layer structures

Cross-sectional scan, as a standard scan mode in vis-OCT, also named raster scan, provides excellent SNR in data acquisition. Figure 5(a) shows the raster scan image details acquired from human Subject A. Compared with the traditional OCT signal in Fig. 5(a), the RM-based method [Fig. 5(b)] eliminates the effects of multiple scattering, resulting in higher imaging contrast, and shows more distinguishable structure and edges, etc., inner plexiform layer (IPL), inner nuclear layer (INL), outer plexiform layer (OPL), outer nuclear layer (ONL), inner segment (IS) of photoreceptor, outer segment (OS) of photoreceptor, and RPE layers. Figure 5(c) shows the A-line OCT signals along the dotted purple and green lines in Figs. 5(a) and 5(b). The RM-based method also enhances the detailed layers, e.g., ONL (cyan arrow). This improvement is also helpful for refined structure visualization.Figures 5(d) and 5(e) show other raster scan images containing blood vessels (red circle) using traditional and RM-based methods, respectively. The shadow effects of large vessels significantly suppress the delicate structure visualization underneath the vessel in the traditional way [the pink arrows in Fig. 5(d)]. The RM-based method overcomes the shadow effect. In Fig. 5(e), RPE contains small fiber bundles that are visible in the result of the RM-based method (the pink arrow). Figure 5(f) confirms the improvement of the RM-based method by showing the A-line OCT signals along the dotted purple and green lines in Figs. 5(d) and 5(e). Furthermore, multiple scattering makes it hard to identify the small fiber bundles in the traditional method. The structures under RPE are also preserved with high quality using the RM-based method [yellow arrow in Fig. 5(e)] compared with the traditional method. Although imaging noises still exist in Figs. 5(b) and 5(e), the improvement in layer structure benefits ophthalmologists’ recognition and diagnostics.In addition, we quantify the improvements of the RM-based method in terms of the SNR and CNR using the raster scan data. For layered structure region (from IPL to RPE) in Figs. 5(a) and 5(b), SNR [Eq. (13)] is calculated using the OCT signals [e.g., in Fig. 5(e)]. The RM-based method provides an average of 1.782 dB (95% CI, [1.587, 1.978] dB) improvement in SNR compared to the traditional method. CNR [Eq. (12)] is calculated in 14 randomly selected areas in raster scan images (seven for either method). The maximum improvement in the CNR index is 2.15 dB using the RM-based method. The RM-based method provides significantly higher CNR than the traditional method, i.e., 0.95 ± 0.40 dB vs 2.48 ± 0.47 dB (data in mean ± SEM, paired t-test, p∼161% improvement) and enables fine structure visualization for clinical diagnostics when using raster scan mode.In clinical applications, arc scan mode facilitates fast sO2 calculation in major arteries and veins in the retina. In practice, multiple scattering component degrades the SNR of the reconstructed A-line signal, making the weak-scattering layers hardly visible. Figure 6(a) shows the vis-OCT image of one arc scan using the traditional method (acquired from Subject B). There are always weak signal areas [green circles in Fig. 6(a)] in arc scan mode due to the limited SNR of the traditional method. RM-based method [Fig. 6(b)] successfully recovers the structural details.

D. RM-based method enables more accurate sO2 measurement

The vasculature supports blood circulation and metabolism in the retina. The unique ability of vis-OCT is that it can simultaneously measure the blood oxygen concentration and image the layer structure in the retina. Figures 5(d) and 5(e) contain large vessels (red circles) that have apparent shadow effects below the vessel due to the strong absorption of red blood cells in visible light. To estimate the sO2 level, we need to identify the vessel boundary and extract the spectrum data along the caliber in the cross section. Misidentifying the top and bottom boundaries may impact the signal normalization and fail sO2 fit then. Vessel top boundaries are easier to be identified in traditional vis-OCT [in Fig. 5(d)] because they locate at the top surface. Nevertheless, detecting the bottom boundaries in the deep layer is usually challenging. High absorption kills the ballistic photons, and multiple scattering is dominant in the blood because of blood flow [red circle in Fig. 5(d)] and then degrades the estimation precision of sO2. The RM-based method further enhances the boundary by improving both the SNR and CNR [red circle in Figs. 5(e) and 5(f)].The measuring principle of blood oxygen information is the BLL, preferring ballistic signals. The RM-based method can separate ballistic signals ruined by multiple scattering to apply the BLL law to fit sO2, which improves the accuracy of sO2 as validated by the phantom experiment. Since arc-scanned image scans most vessels simultaneously, it is used for sO2 measurement in human clinical experiments. Figures 6(c) and 6(f) show A-lines and sO2 of the highlighted artery and vein by traditional and RM-based methods. The intensity of the traditional A-lines across the vessel has a large deviation, while the RM-based A-lines attenuate more when depth increases, mainly because it reduces the impact of multiple scattering whose random photon trajectories lift the intensity of vis-OCT signal and reduce SNR, making sO2 measurement inaccuracy. In Figs. 6(d) and 6(e), both traditional and RM-based methods output consistent sO2 measurements in the artery [red arrow in Figs. 6(a) and 6(b)]. We further compare the accuracy in sO2 measurement using the fitting R-squared (R2) and root-mean-square error (RMSE) parameters. Compared with the traditional method, the RM-based method improves the R2 from 0.84 to 0.93. The corresponding RMSE decreases from 0.0156 (traditional method) to 0.0134 (RM-based method). For the vein [the blue arrows in Figs. 6(a) and 6(b)], the RM-based processed A-lines keep smaller deviations with depth increasing. The typical sO2 of the vein was less than 0.6. In Fig. 6(g), the traditional method presents a biased sO2 = 0.74 estimation due to uncertainty in identifying the bottom wall and multiple scattering contributions. Meanwhile, the RM-based method again provides the correct estimation sO2 = 0.56 in Fig. 6(h).

V. DISCUSSIONS

Section:

ChooseTop of pageABSTRACTI. INTRODUCTIONII. THEORYIII. METHODIV. RESULTSV. DISCUSSIONS <<VI. CONCLUSIONSUPPLEMENTARY MATERIALREFERENCESOur clinical data showed that the RM-based method effectively isolated the ballistic component from the spectral domain, revealing more layered retina structures. One of the assumptions for RM-based reconstruction is that the vis-OCT is approximately independent and identically distributed sampling in k-space. The high lateral resolution, or small scanning step, would introduce a correlation between the neighboring A-lines, which violates the assumption of independent sampling and leads to errors in RM-based estimations. The isolation of the ballistic component is also affected by the number of A-lines constructing the RM. Larger size RM always results in fewer estimation errors but can also be biased by the tissue inhomogeneity. In this study, we selected seven A-lines in the RM construction by checking the inhomogeneities in the acquired A-line signals. The scanning step of the commercial vis-OCT system is specifically designed for human retina imaging (A-line acquisition rate of 25 kHz with a step size of 10 µm). When applied to other tissues, the scanning step needs to be further optimized, or we can use an alternative strategy of repeated scanning along the fixed location, i.e., M-mode OCT.4343. Y. Yasuno, S. Makita, T. Endo, G. Aoki, M. Itoh, and T. Yatagai, “Simultaneous B-M-mode scanning method for real-time full-range Fourier domain optical coherence tomography,” Appl. Opt. 45, 1861–1865 (2006). https://doi.org/10.1364/ao.45.001861We used the Wishart random matrix model that is well-established in wave scattering phenomenons, including the electromagnetic wave scattering in multiple-input multiple-output (MIMO) systems4444. A. Zanella, M. Chiani, and M. Z. Win, “On the marginal distribution of the eigenvalues of Wishart matrices,” IEEE Trans. Commun. 57, 1050–1060 (2009). https://doi.org/10.1109/tcomm.2009.04.070143 and coherent light scattering in a random medium.4545. N. Byrnes and M. R. Foreman, “Random matrix theory of polarized light scattering in disordered media,” Waves Random Complex Media 0, 1–29 (2022). https://doi.org/10.1080/17455030.2022.2153305 Usually, Gaussian statistics are applied in the representation of wave fields under multiple scattering. In statistics, the Wishart distribution is a generalization to multiple dimensions of the gamma distribution.4646. M. J. Wishart, “The generalised product moment distribution in samples from a normal multivariate population,” Biometrika 20A, 32–52 (1928). https://doi.org/10.2307/2331939 In Bayesian statistics, the Wishart distribution is a conjugate prior to the precision parameter of the multivariate normal distribution.4747. C. Bishop, Pattern Recognition and Machine Learning (Statistical Science, 2006). The matrix gamma distribution (multivariate gamma distribution) is a more general version of the Wishart distribution.4848. A. Iranmanesh, M. Arashi, and S. M. M. Tabatabaey, “On conditional applications of matrix variate normal distribution,” Iranian Journal of Mathematical Sciences and Informatics 5(2), 33–43 (2010). https://doi.org/10.7508/ijmsi.2010.02.004 These generalizations may better represent the scattering phenomenon in the mesoscopic scale, which is out of the scope of this study. For real-valued positive-definite matrices, the inverse Wishart distribution provides tools for estimating the distributions of multivariate autoregressive processes,4949. K. Triantafyllopoulos, “Real-time covariance estimation for the local level model,” J. Time Ser. Anal. 32, 93–107 (2011). https://doi.org/10.1111/j.1467-9892.2010.00686.x thus suitable for describing the dynamic scattering process.The isolated ballistic components by the RM-based method also benefit the diagnostic in 3D. Unfortunately, in this study, there are unavoidable motions during the human retinal imaging experiments, leading to poor quality in 3D visualization. To demonstrate the improvement of the RM-based method, Fig. 7 (see visualization 1) shows the retina en-face (top view) of an anesthetized mouse. It is more obvious by the RM-based method [Fig. 7(b)] than the traditional method [Fig. 7(b)], especially for vessel outlines. For B-scan reconstructed side (y − z), the RM-based method has more clear layer structure than the traditional method shown in the zoom-ins of Fig. 7 (see visualization 2). Further development of the image registration method is required for clinical 3D diagnostics using vis-OCT.The current RM-based method considers the scattering effects only. Absorption, however, also plays a vital role in imaging biological tissues, especially within the visible light spectrum. BLL has been directly applied in vis-OCT to estimate the sO2 in large vessels, which is valid in the ballistic regime. This conventional method is based on the assumption of ballistic photon propagation, which thus can be biased by the multiple scatterings in practice. There is currently no reliable method for tissue areas with deep vessels for sO2 estimation due to insufficient sensitivity. Nevertheless, the deep tissues in the retina are important, and the multiple scattering components dominate in deep tissues because absorption eliminates some of the single scattering trajectories in a higher chance, which is more prominent in retinal tissue. The choroid supplies oxygen and nutrients to the retina. Defects in the choroideremia (CHM) gene can cause CHM.5050. R. K. Menon, W. S. Ball, and M. A. Sperling, “Choroideremia and hypopituitarism: An association,” Am. J. Med. Genet. 34, 511–513 (1989). https://doi.org/10.1002/ajmg.1320340411 The area of leaky blood vessels can further expand in the retina, causing AMD5151. W. L. Wong, X. Su, X. Li, C. M. G. Cheung, R. Klein, C.-Y. Cheng, and T. Y. Wong, “Global prevalence of age-related macular degeneration and disease burden projection for 2020 and 2040: A systematic review and meta-analysis,” Lancet Global Health 2, e106–e116 (2014). https://doi.org/10.1016/s2214-109x(13)70145-1 and diabetic retinopathy.5252. A. J. Barber, T. W. Gardner, and S. F. Abcouwer, “The significance of vascular and neural apoptosis to the pathology of diabetic retinopathy,” Invest. Ophthalmol. Visual Sci. 52, 1156–1163 (2011). https://doi.org/10.1167/iovs.10-6293 Therefore, sO2 evaluation in choroid tissue is essential for clinical diagnostics. In this study, the RM-based method reveals signals in the choroid tissue’s top layer, which supports further sO2 estimation in choroid tissue.Although the signals in this study are from vis-OCT, a special SD-OCT, the RM-based method can also be applied to TD-OCT (Time Domain OCT) or SS-OCT (Swept Source OCT). Applying the RM-based method to TD-OCT thus may not improve the SNR or CNR as well as that to SD-OCT. SS-OCT, however, sweeps the wavelength in time and provides more independent sampling, which thus reduces the cross-talk between neighbored wavelengths and makes it more suitable for the RM-based method. The RM-based method provides a tool for separating the ballistic and multiple scattering components, facilitating more accurate structural and functional imaging of the tissues. In addition, Doppler and polarization information can also be extracted more effectively from the RM-based signal reconstruction. It should be emphasized that the RM-based method is proposed as a statistical estimation other than a stand-alone denoising method. After the RM-based reconstruction, SNR can be further improved by other post-processing methods, such as wavelet-based denoising,5353. S. Chitchian, M. A. Fiddy, and N. M. Fried, “Denoising during optical coherence tomography of the prostate nerves via wavelet shrinkage using dual-tree complex wavelet transform,” J. Biomed. Opt. 14, 014031 (2009). https://doi.org/10.1117/1.3081543 anisotropy curvelet transform,5454. Z. Jian, L. Yu, B. Rao, B. J. Tromberg, and Z. Chen, “Three-dimensional speckle suppression in optical coherence tomography based on the curvelet transform,” Opt. Express 18, 1024–1032 (2010). https://doi.org/10.1364/oe.18.001024 sparsity-based denoising,5555. L. Fang, S. Li, Q. Nie, J. A. Izatt, C. A. Toth, and S. Farsiu, “Sparsity based denoising of spectral domain optical coherence tomography images,” Biomed. Opt. Express 3, 927–942 (2012). https://doi.org/10.1364/boe.3.000927 and enhanced low-rank plus sparsity decomposition.2626. I. Kopriva, F. Shi, and X. Chen, “Enhanced low-rank + sparsity decomposition for speckle reduction in optical coherence tomography,” J. Biomed. Opt. 21, 076008 (2016). https://doi.org/10.1117/1.jbo.21.7.076008 The post-processing methods improve the visualization of arc scan vis-OCT image from the traditional method. Figure 8(b) shows the enhanced result from the traditional vis-OCT image [Fig. 8(a)] after enhancement of brightness (100%) and contrast (10%). However, this improvement cannot recover the layer details (the yellow arrows) revealed in the vis-OCT image from the RM-based method [Fig. 8(c)]. Furthermore, we can also apply these post-processing methods to improve the vis-OCT image from the RM-based method. Figure 8(d) shows the result after the enhancement of brightness (100%) and contrast (10%). The enhancement of brightness and contrast also magnified the noise level. More sophisticated post-processing strategies are needed to improve ophthalmologists’ diagnostics in the future.