Characterization of Leukocyte Crawling in the Living Human Retina Using Adaptive Optics Scanning Light Ophthalmoscopy Authors: Lily A. Greenberg, Luis Muncharaz Duran, Aaron Brown, Alexander Barash, Deep Parikh, Meenakashi Gupta, Jordan Bellis, Jeffrey A. Glassberg, Richard B. Rosen, Toco Y. P. Chui Published: 2026 ABSTRACT PurposeDirect visualization of leukocytes within the retinal microvasculature is essential for investigating local immune responses. In this study, we characterize leukocyte crawling dynamics in the retinal vasculature of healthy volunteers and three patients with systemic diseases using adaptive optics scanning light ophthalmoscopy (AOSLO).MethodsTwenty-three healthy volunteers (age range, 20–75 years) and three patients with systemic diseases (diabetes, sickle cell disease, and syphilis) were imaged using a commercial AOSLO device. Consecutive phase-contrast AOSLO images of the temporal retina were acquired over 5 minutes at two to four preidentified regions of interest. Leukocyte crawling events were tracked manually in the 5-minute time-lapse video; vessel type, lumen diameter, speed, and pattern were assessed.ResultsA total of 232 intraluminal leukocyte crawling events were identified in healthy volunteers. Crawling events were more frequently observed in venules (89.66%) than arterioles (9.92%). Lumen diameter with leukocyte crawling ranged from 10.2 to 67.8 µm; mean ± SD crawling speed was 0.3 ± 0.1 µm/s. Four distinct crawling patterns, with respect to direction of blood flow, were identified: with, against, backtrack, and across. No significant difference in crawling speed was found between venules and arterioles (P = 0.67) or among crawling patterns (P = 0.197). In addition to crawling, various leukocyte dynamics, including adhesions, reverse transmigration, and extravascular cell motilities, were observed in patients with systemic diseases.ConclusionsThis study demonstrates the feasibility of utilizing AOSLO to noninvasively visualize leukocyte crawling in the living human eye. The ability to characterize leukocyte dynamics in their natural environment offers invaluable insight into the localized retinal immune status. Leukocytes play a central role in immune defense by detecting pathogens, coordinating immune response, and repairing tissue damage.1 However, in disease states, these processes can become hypoactive (impairing host defense) or hyperactive (driving pathologic inflammation).2–4 A deeper understanding of leukocyte dynamics may clarify their role in physiological and pathological immunity. Leukocytes utilize the circulatory system as a sophisticated transit system to patrol their environment. Leukocyte rolling is the initial phase of immune surveillance and typically takes place on the luminal surface. Leukocytes roll at velocities slower than free-flowing blood, which allows for exposure to chemokines on the endothelial surface.5 Once leukocytes are activated, they enter the adhesion phase that leads to firm leukocyte arrest on the endothelial surface. Subsequently, leukocytes enter the crawling phase, an intraluminal process that enables relocation to optimal sites for transendothelial migration, also known as extravasation.6 Extravasation is the process by which leukocytes exit blood vessels and migrate into the surrounding tissue.7 Reverse transmigration occurs when previously extravasated leukocytes travel back across the endothelium into the bloodstream.8 As the only transparent organ in the human body, as well as one of the few immune-privileged sites, the eye offers a unique window into studying the immune system. Adaptive optics scanning light ophthalmoscopy (AOSLO) enables high-resolution imaging of the retina at the cellular level.9–17 It employs a wavefront sensor to detect optical aberrations due to refractive error, tear film irregularities, and imperfections in the cornea and lens. A compensating deformable mirror corrects these aberrations.10 AOSLO imaging is typically separated into two channels: confocal and non-confocal. The confocal channel captures singly backscattered light, providing visualization of high-contrast retinal features such as photoreceptors9,10,13 and the retinal nerve fiber layer.18–20 The non-confocal channel captures multiply scattered light, providing visualization of low-contrast or transparent retinal structures such as blood vessels,14,21,22 retinal ganglion cells,16 and immune cells.17,23–25 Several non-confocal detection schemes have been developed, including offset aperture,14,21 split detection,15 quadrant detection,26 and multidetection.27 These advances in non-confocal imaging have facilitated an increasingly detailed investigation of real-time vascular and cellular dynamics in the living human retina. Previous studies have explored the application of AOSLO for studying immune cell dynamics in the retina. In humans, motility and morphology of hyalocytes, macrophages, and microglia have been characterized under healthy23 and inflammatory conditions.25 Additionally, leukocyte dynamics have been explored in mice17 and nonhuman primates.28 In mouse models, stages of the leukocyte cascade, including rolling, crawling, and transmigration, have been studied longitudinally and manipulated to stimulate inflammatory conditions.17 In nonhuman primates, leukocyte crawling has been characterized by morphology, vessel location, speed, and direction.28 These studies established a vital foundation for exploring analogous dynamics in humans. In this study, we utilized a commercially available AOSLO to visualize leukocyte dynamics in the eyes of healthy volunteers and patients with systemic diseases. We have achieved high-contrast, label-free imaging of leukocyte crawling events, enabling manual identification, tracking, and characterization. Direct visualization of leukocyte dynamics in their natural environment offers invaluable insights into immunopathogenesis and leverages leukocytes as quantitative biomarkers for disease progression and treatment response. Methods Participants This study was performed at the New York Eye and Ear Infirmary of Mount Sinai. All procedures adhered to the tenets of the Declaration of Helsinki and were approved by the Institutional Review Board of Mount Sinai Hospital (identifier 17–00041). Written informed consent was obtained from all participants after thorough explanation of possible risks. A total of 23 healthy volunteers with no known retinal pathology and systemic disease were recruited based on self-report. Fundus photographs and optical coherence tomography (OCT) angiograms (OCT-A) were reviewed by an on-site ophthalmologist to confirm healthy status. Additionally, three patients with retinopathies due to different systemic diseases (diabetes, sickle cell disease, and syphilis) were recruited to demonstrate clinical feasibility of leukocyte dynamics imaging. Eligibility required one eye with pupil dilation to at least 5 mm. Exclusion criteria were severe nuclear sclerosis, vitreous opacities, and macular edema. One eye from each participant was imaged. Prior to imaging, one drop of 2.5% phenylephrine and one drop of 0.5% to 1% tropicamide were used for pupil dilation and cycloplegia. Angles and intraocular pressure were examined before dilation for participants with no history of dilation. Fundus, OCT-A, AOSLO, and Axial Length Instruments Fundus photographs were acquired using a color fundus camera (DRI OCT Triton Plus; Topcon, Tokyo, Japan). OCT-A scans were acquired using a spectral-domain OCT device, which utilizes an illumination source centered at 840 nm, a bandwidth of 45 nm, an axial resolution of 5 µm, and an acquisition rate of 70,000 A-scans per second (Avanti RTVue-XR; Optovue, Fremont, CA, USA). AOSLO videos were acquired using an AOSLO device, which utilizes an 840-nm beam for both imaging and wavefront sensing with 640 µW peak power measured at the pupil (MONA IV; Robotrak Technologies, Nanjing, China). Confocal lateral resolution was measured by imaging a USAF1951 test target using a 19-mm focal length achromatic lens and determined to be ∼2 µm horizontally and vertically. Established microscopy formulas were used to derive the corresponding confocal axial resolution (see Supplementary Material provided by Robotrak Technologies).29,30 To estimate the non-confocal resolution, the non-confocal channel was modeled as a confocal channel with a diffraction-limited illumination beam and a larger, finite detection pinhole, allowing comparable resolution calculations. Optical section thickness was estimated using the confocal resolution measurement and theoretical optical modeling of the system and determined to be ∼36 µm.31 The imaging device simultaneously collected one confocal and six non-confocal videos. Confocal videos were collected using a circular aperture of ∼1.1 Airy disk diameter. Non-confocal videos were captured using six evenly spaced circular offset apertures located ∼8 Airy disk diameter from the confocal aperture. Each offset aperture has a diameter of ∼7 Airy disks. In this study, all videos were captured with a 2.4° × 2.4° field of view at a frame rate of ∼14 Hz for 6 seconds (86 frames per video, 1024 × 1024 pixels). The built-in image-processing software (MonaPro V1.00.04.251030, Robotrak Technologies, Nanjing, China) was used to generate the demotioned videos for the one confocal channel and six non-confocal channels. The software utilized a strip-based registration approach to calculate the relative motion of each strip.32 Initial registration utilized the confocal channel, and the resulting transformations were subsequently applied to the six non-confocal channels. Frames with poor image quality were excluded and appear as dark frames in the demotioned videos. The six demotioned non-confocal videos were then used to create three split-detection videos at 100°, 160°, and −140°. Each of the three split-detection videos was used in the convolution calculation, with the corresponding 5 × 5–pixel emboss filter mask as described previously.23,33,34 The three emboss-filtered split-detection videos were subsequently merged to create a single phase-contrast video using mean intensity projection. The software generates an average confocal and phase-contrast image from the demotioned confocal and phase-contrast videos, respectively (Figs. 1A, 1B), in addition to a perfusion image from the standard deviation of the demotioned phase-contrast video (Fig. 1C). Simultaneously acquired AOSLO images in a representative healthy volunteer (60-year-old male) using a commercial device. (A) Averaged confocal, (B) averaged phase-contrast, and (C) perfusion images. All images have been contrast enhanced for display. The axial length of the studied eye was measured using an ocular biometer (IOL Master; Carl Zeiss Meditech, Dublin, CA, USA). Fundus, OCT-A, AOSLO, and Axial Length Measurement The entire imaging session, including fundus photograph, OCT-A, AOSLO, axial length, and breaks for participant comfort, lasted approximately 1.5 hours. One fundus photograph centered at the macula and one 3 × 3–mm OCT-A centered at ∼9° temporal to the fovea were captured at the beginning of the session. The fundus photograph and superficial layer OCT-A were utilized to differentiate between arterioles and venules based on vessel color and the presence of the periarteriolar capillary-free zone, respectively. Next, two to four regions of interest (ROIs) with the presence of arterioles and venules were identified from OCT-A images for subsequent AOSLO imaging (Fig. 2). Methodological overview of the imaging session in a representative healthy volunteer (29-year-old male). (A) Color fundus photograph showing the 3 × 3–mm temporal retina imaged with OCT-A (black box). (B) The 3 × 3–mm superficial OCT-A acquired approximately 9° temporal to the fovea, with arterioles and venules delineated by red and blue overlays, respectively; four 2.4° × 2.4° regions of interest imaged with AOSLO are indicated. (C) AOSLO phase-contrast image corresponding to the yellow box region highlighted in B. All images have been contrast enhanced for display. For each ROI, consecutive 6-second AOSLO videos were acquired over 5 minutes (average 26 acquisitions per ROI). During AOSLO imaging, participants were encouraged to blink between video acquisitions, and some participants received artificial tear drops between ROI imaging sessions for better patient comfort or image quality. Lastly, the axial length of the study eye was measured to correct individual retinal magnification factor. Registering Confocal, Phase-Contrast, and Perfusion Images Across 5 Minutes Image registration was performed to optimize visualization of the leukocytes and track their movement over time. Averaged confocal, averaged phase-contrast, and perfusion images from each ROI were registered using the Register Virtual Stack Slices plugin with advanced setup on ImageJ (US National Institutes of Health, Bethesda, MD, USA).35 Primary image registration was performed mainly on confocal images; if registration was inadequate, perfusion images were used. To facilitate registration, images underwent pseudo flat-field correction with a blurring radius of 50 pixels using the BioVoxxel toolbox ImageJ plugin36 and contrast enhancement by saturating 0.1% of the pixels using Enhance Contrast in ImageJ, to correct uneven illumination and achieve sufficient contrast, respectively. The computed transformation matrices were subsequently applied to the phase-contrast images using the Transform Virtual Stack Slices ImageJ plugin. A 5-minute time-lapse video of the registered phase-contrast images was then generated for visualization of leukocyte crawling. Identifying and Tracking of Leukocyte Crawling Over 5 Minutes Leukocyte crawling events were identified by two graders following repeated review of the 5-minute time-lapse phase-contrast video from each ROI. Each leukocyte crawling event was tracked manually with the TrackMate ImageJ plugin.37 The leukocyte travel displacement between consecutive time points was converted to distance in microns (µm) using the retinal scaling factor derived from the Bennett–Littmann ocular scaling formulas.38 Timestamps from each AOSLO-averaged image were used to determine the time elapsed between frames and ultimately to calculate mean speed. Speed of leukocyte crawling was calculated for each cell using distance traveled divided by the time elapsed between timestamps. Each leukocyte track was also characterized by vessel type (arteriole and venule), crawling pattern in relation to blood flow, and lumen diameter (µm). Lumen Diameter Measurement Lumen diameter at the site of the leukocyte crawling event was measured using various ImageJ plugins (Fig. 3). First, an average of the phase-contrast 5-minute time-lapse was generated (Fig. 3A). Next, the margin of the blood column corresponding to the leukocyte crawling event was manually traced and linearized using the Straighten plugin.39 The resulting vertically straightened segment (Fig. 3B) was processed using the Differentials plugin to compute the gradient magnitude (Fig. 3C), representing the blood column and vessel wall boundaries.40,41 Intensity profiles of the straightened segment were then extracted using Plot Profile, and peaks corresponding to the vessel wall edges were identified using the BAR peak analysis tool (Fig. 3D).42 In general, four peaks corresponding to the vessel walls and blood column were identified. The two outermost peaks represent the distance between the vessel walls and were used to calculate the lumen diameter (Fig. 3D, blue data points). The two innermost peaks represent the width of the blood column (Fig. 3D, red data points). The distance between each outermost peak and its corresponding innermost peak represents the width of the apparent cell-free layer on either side of the blood column (Fig. 3D; Supplementary Video S1).43 For leukocyte crawling events spanning multiple vessels, only the lumen diameter along the path where the leukocyte traveled the greatest distance was measured. Lumen diameter measurement in a representative healthy volunteer (39-year-old female). (A) Average of AOSLO phase-contrast 5-minute time-lapse. White box indicates location of leukocyte crawling event. (B) Corresponding vessel segment after vertical straightening. (C) Gradient magnitude of straightened vessel segment. (D) Graph of intensity versus distance (µm). Four peaks corresponding to the vessel walls and blood column were identified manually. The two outermost peaks (blue) represent the distance between the vessel walls (lumen diameter), while the two innermost peaks (red) represent the width of the blood column. The distance between each outermost peak and its corresponding innermost peak represents the width of the apparent cell-free layer. See Supplementary Video S1 for the demotioned phase-contrast video, demonstrating the blood flow column and apparent cell-free layer. All images have been contrast enhanced for display. Statistical Analysis All statistical analyses were performed using SPSS version 24.0 (IBM Corporation, Chicago, IL, USA). Descriptive statistics were computed for all variables. The normal distribution of all datasets was assessed using the Shapiro–Wilk test. Parametric independent samples t-tests were used to compare the number of leukocyte crawling events between sex (male vs. female), as well as mean leukocyte crawling speed between sex (male vs. female) and between vessel type (arteriole vs. venule). Pearson's correlation was used to assess the relationship between speed and lumen diameter, as well as between speed and age. One-way analysis of variance was performed to compare mean speed among four crawling patterns. A P value <0.05 was considered statistically significant. Results Leukocyte Crawlings in Healthy Volunteers Eighty-one ROIs in 23 healthy volunteers were imaged using AOSLO (age range, 20–75 years; mean ± SD: 38.0 ± 18.2 years; 10 male and 13 female) (Table). Leukocyte crawling was observed in 22 of 23 healthy volunteers and 61 of 81 ROIs. Leukocytes appeared as distinctive structures within the vascular lumen on phase-contrast images. Their morphology ranged from flat to ovoid. A 5-minute time-lapse revealed leukocyte crawling along distinct and continuous paths within the apparent cell-free layer (Fig. 4; Supplementary Video S2). Although some events appeared to occur within the blood column on the longitudinal cross section of the blood vessel, these likely represent leukocytes crawling within the cell-free layer at either the ceiling or floor of the vessel lumen. Table.Summary of ROIs and Number of Leukocyte Crawling Events in Each Healthy VolunteerParticipantSexAge, YEye ImagedAxial Length, mmTotal No. of ROIs Imaged (No. of ROIs With Crawlings)Total No. of Arterioles Imaged* (No. of Arterioles With Crawlings)Total No. of Venules Imaged* (No. of Venules With Crawlings)Total No. of Crawlings1F26OD24.684 (2)5 (1)6 (1)22M24OS23.214 (1)6 (0)4 (1)13M30OS25.333 (2)4 (0)4 (2)54F24OS23.034 (2)3 (0)5 (2)45M29OD24.124 (3)3 (0)5 (3)266M26OD25.674 (3)5 (2)3 (2)67M20OD25.033 (3)3 (1)5 (4)128M26OS24.914 (4)4 (1)4 (4)159F66OS23.334 (3)4 (1)4 (2)510F27OS26.243 (3)4 (1)2 (2)511M25OD25.334 (3)4 (1)5 (4)3212F28OS23.244 (4)5 (0)5 (4)1513M25OD24.744 (1)5 (0)3 (1)214F27OS24.764 (4)3 (1)4 (4)2615M28OD23.564 (4)4 (3)4 (2)916F23OS23.433 (3)5 (0)3 (3)717F39OD24.743 (2)3 (1)4 (1)918M60OD22.564 (4)4 (2)5 (3)1219F65OS24.404 (3)6 (0)5 (3)1720F75OD22.082 (0)2 (0)2 (0)021F67OD22.662 (2)1 (0)2 (2)422F53OD23.514 (3)4 (1)3 (3)1223F62OD23.962 (2)3 (0)2 (2)5Total————81 (61)90 (16)89 (55)231*Number of vessels was counted based on branch-point connectivity. All vessel segments connected through visible branch points were counted as a single vessel. Segments without a visible branch-point connection were counted as separate vessels. Summary of ROIs and Number of Leukocyte Crawling Events in Each Healthy Volunteer Number of vessels was counted based on branch-point connectivity. All vessel segments connected through visible branch points were counted as a single vessel. Segments without a visible branch-point connection were counted as separate vessels. Manual tracking of leukocyte crawling events on AOSLO phase-contrast 5-minute time-lapse in a representative healthy volunteer (28-year-old female). (A) Nine leukocyte crawling tracks are coded by different colors. A, arteriole; V, venule. Black arrows indicate blood flow direction. Yellow arrow indicates an individual leukocyte crawling event represented in B. (B) Sequential phase-contrast images showing representative leukocyte crawling (yellow arrows) in the apparent cell-free layer (top: white arrows) of the venule, with corresponding timestamps (in seconds) indicated for each image. See Supplementary Video S2 for the complete sequence. All images have been contrast enhanced for display. A total of 232 leukocyte crawling events were identified. Leukocyte crawling was observed in the lumen diameter ranging from 10.21 to 67.81 µm (Fig. 5; Supplementary Videos S3–S6). Across all ROIs, 90 arterioles and 89 venules were imaged. Capillaries were abundant within each ROI and were not individually counted. In total, 208 (89.66%) leukocyte crawling events were found in venules, 23 (9.92%) were found in arterioles, and 1 (0.43%) was found in a capillary. The mean ± SD speed of leukocyte crawling was 0.3 ± 0.1 µm/s (range, 0.1–0.7 µm/s). Four distinct crawling patterns with respect to the direction of blood flow were identified: with (traveling in the same direction as blood flow), against (traveling in the opposite direction as blood flow), backtrack (traveling in one direction and subsequently reversing direction along the same track), and across (traveling from one side of the vessel lumen to the contralateral side) (Supplementary Video S7). The most frequently observed leukocyte crawling pattern was with (44.6%), followed by against (34.6%), backtrack (8.2%), and across (5.2%). Some leukocytes demonstrated multiple crawling patterns over the course of imaging (7.4%). Phase-contrast AOSLO images of a range of lumen diameters with leukocyte crawling events in four healthy volunteers. (A) A 26-year-old male, (B) a 60-year-old male, and (C) a 65-year-old female, demonstrating two leukocytes (cyan and yellow arrows) crawling along the same trajectory. The leukocyte indicated by the yellow arrow bulges into the blood flow column, creating a localized perturbation of the adjacent blood flow. (D) A 27-year-old female. Each row shows three representative time points across the 5-minute imaging session (arranged in temporal order from left to right, with time points in seconds shown in the top right corner of each panel); yellow arrows indicate the leukocyte crawling within the apparent cell-free layer at each time point. See Supplementary Videos S3 to S6 for the complete sequences. All images have been contrast enhanced for display. Although numerous capillaries were visualized, only one capillary leukocyte crawling event was identified, precluding meaningful analysis of capillary crawling behavior. Therefore, the single capillary event (Fig. 5A; Supplementary Video S3) was excluded from statistical analysis, leaving 231 leukocyte crawling events. There was no significant difference in the number of crawling events between sexes (male versus female) (P = 0.349) or across age (P = 0.361). There was no significant difference in crawling speed between vessel type (arterioles versus venules) (P = 0.67) (Fig. 6A), between sexes (male versus female) (P = 0.686), or across age (P = 0.837). There was no correlation between speed and lumen diameter (P = 0.436) (Fig. 6B; Supplementary Videos S4–S6). For crawling patterns in venules, excluding the leukocyte crawling events with more than one pattern, there was no significant difference in crawling speed (P = 0.197) (Fig. 7). Comparison of leukocyte crawling speed across vessel type and lumen diameter. (A) Boxplot comparing leukocyte crawling speed in arterioles versus venules; no significant difference was observed (independent samples t-test, P = 0.67). (B) Scatterplot of leukocyte crawling speed versus lumen diameter in arterioles and venules; no significant correlation between speed and lumen diameter was observed (Pearson's correlation, P = 0.436). (C) Histogram showing the frequency distribution of leukocyte crawling events across lumen diameters. Boxplot of leukocyte crawling speed across different crawling patterns in venules. Leukocyte crawling events exhibiting more than one crawling pattern were excluded from analysis. No significant differences were demonstrated (one-way analysis of variance, P = 0.197). Leukocyte Dynamics in Systemic Diseases Leukocyte crawling along the vascular wall was observed in all three patients. These three cases demonstrate the feasibility of applying these methods in real-world clinical populations to explore leukocyte crawling behavior. In addition, several pathology-specific leukocyte behaviors were identified (Fig. 8; Supplementary Videos S8–S10). In eyes with diabetic retinopathy and sickle cell retinopathy, leukocyte adhesion to the venular vessel wall was observed (Figs. 8A, 8B; Supplementary Videos S8–S9). In the eye with sickle cell retinopathy, the adhered leukocyte was later observed detaching from the vessel wall and reentering the bloodstream. Reverse transendothelial migration of leukocytes from tissue into bloodstream was also visualized in the same eye with sickle cell retinopathy. In the eye with posterior uveitis due to syphilis, numerous extravascular cell motilities along the vessels were clearly evidenced in the time-lapse video (Fig. 8C; Supplementary Video S10). The sizes of these extravascular cells were relatively smaller and rounder compared to the intraluminal crawling leukocytes. Leukocyte dynamics of three patients with retinopathies due to various systemic diseases over the course of 5 minutes. (A) A 48-year-old female with type 2 diabetes mellitus complicated by nonproliferative diabetic retinopathy. (B) A 23-year-old female with sickle cell disease (HbSS genotype) complicated by sickle cell retinopathy. (C) A 39-year-old male with posterior uveitis (syphilitic chorioretinitis). Only leukocytes with distinct paths were tracked in C. Leukocyte crawling, leukocyte adhesion (Ad), reverse transendothelial migration (rTEM), and extravascular cell motility (EXV) were identified in patients. A, arteriole; MA, microaneurysm; V, venule. Black arrows indicate blood flow direction. See Supplementary Videos S8 to S10 for the complete sequences. All images have been contrast enhanced for display. Discussion Leukocyte crawling is a critical phase of the leukocyte cascade, enabling adherent leukocytes to actively migrate along the endothelium in search of optimal sites for transmigration through blood vessels into adjacent inflamed tissue. Characterization of leukocyte crawling behaviors offers a window into some of the intricacies of immune surveillance and response. Our study demonstrates the feasibility of utilizing AOSLO to visualize leukocyte crawling in the eyes of healthy volunteers and patients with systemic diseases. The ability to characterize leukocyte dynamics, noninvasively and label free, in their natural environment offers invaluable insight into the localized retinal immune status. In total, we analyzed 231 leukocyte crawling events in healthy volunteers. The highest frequency of crawling events was found in vessels with a lumen diameter between 25 and 45 µm. Despite nearly equal numbers of arterioles (90) and venules (89) imaged, most leukocyte crawling events were found in venules (89.66%), as compared to arterioles (9.92%), consistent with prior literature.28 Several factors may contribute to the predominance of leukocyte crawling in venules, including differences in wall shear stress and vessel wall composition between arteries and veins. Studies have shown that the pattern of leukocyte migration in arteries, where shear stress is higher, is preferentially in the direction of blood flow, whereas no clear directional preference is observed in veins.44 Our findings are consistent with this observation, which suggests increased shear stress may influence leukocyte migration by limiting effective crawling directions. However, even when arterial shear stress was experimentally manipulated to match the vein, the predominance of leukocyte crawling in veins remained, suggesting factors other than hemodynamics contribute to this phenomenon.45 Studies have shown that the venule endothelium has higher baseline expression of integrins responsible for upregulation of adhesion during inflammation.46 This increases the vessels’ capacity for leukocyte adhesion, likely contributing to the greater prevalence of leukocyte crawling events observed. The mean ± SD speed of leukocyte crawling in our study was 0.3 ± 0.1 µm/s, as compared to prior studies carried out in the retina of nonhuman primates (0.2 ± 0.1 µm/s).28 These crawling speeds are consistently slower than leukocyte rolling, which has been measured in human oral mucosa (range, ∼4–60 µm/s) and mice cremaster muscle (range, ∼5–30 µm/s) under varying conditions.47,48 This may reflect the different molecular mechanisms and biological activities of these two phases of the leukocyte cascade.5 The prior study carried out in nonhuman primates identified two crawling patterns with respect to the direction of blood flow (with and against)28; our study expanded on this, identifying four distinct patterns (with, against, backtrack, and across; Supplementary Video S7). Both observations are consistent with prior literature, which suggests leukocyte crawling can occur in multiple directions in relation to blood flow, while leukocyte rolling exclusively occurs in the direction of blood flow.5 Our analysis revealed no significant differences in leukocyte crawling speed between arterioles and venules, across lumen diameters, or among crawling directions. This finding aligns with the understanding that leukocyte crawling is an active process mediated by interactions with integrins and selectins.5,45,48 Therefore, the composition of the vessel wall likely has the greatest influence on leukocyte crawling ability and speed, rather than factors such as hemodynamics or vessel size. While the primary aim of our study was to quantitatively characterize leukocyte crawling events in healthy volunteers, the ability to visualize these crawling events, as well as other leukocyte dynamics, in patients with systemic disease demonstrates the feasibility of applying this approach in nonhealthy human eyes. Leukocyte adhesions were visualized in our diabetic and sickle cell retinopathy cases. In diabetes, previous studies have not only demonstrated a positive association between leukocyte adhesion and hyperglycemia but also highlighted the relationship between leukocyte adhesion and diabetic micro- and macrovascular alterations.49,50 In sickle cell disease, inflammation plays a central role in vaso-occlusive events. Activated leukocytes become adherent to the endothelium, upregulating expression of additional adhesion molecules. This promotes the recruitment of circulating platelets and sickled red blood cells, which contributes to vascular congestion, occlusion, and ischemia.51 Extravascular cellular movement was visualized in our syphilitic choroiditis case. We believe these cells and their behavior closely resemble those previously documented in the mouse retina in response to endotoxin-induced inflammation.17 Demonstrating this phenomenon in a clinical case underscores the translational relevance of these findings to human uveitis and highlights potential implications for assessing active inflammation. The limitations and potential sources of error in this study are important to note. While leukocyte crawling events were identified manually by two graders, we did not assess intergrader variability in this study. Future assessment of intergrader agreement could provide insight into the reliability and consistency of manual leukocyte crawling event identification. Additionally, the identification of the cell-free layer remains observational in nature. Complementary validation approaches, such as fluorescent labeling, would likely be required to further characterize and confirm this observation. During AOSLO imaging, the plane of focus was optimized at the two sides of the vessel's long axis. As a result, leukocyte crawling events occurring on the ceiling or floor of the vessel may have been out of focus and therefore not identified, particularly in vessels with a larger lumen diameter. Moreover, given the two-dimensional nature of our measurements, we likely underestimated the total distance traveled and, therefore, the true leukocyte speed. Leukocytes traveling along the z-axis likely cover a greater distance than we could measure. Further, each ROI was imaged for 5 minutes to prioritize participant comfort; however, not all leukocyte crawling events began and ended within this time frame, potentially leading to incomplete paths of leukocyte crawling. Lastly, at times, it was difficult to differentiate between movements of hyalocyte processes overlying the vessel wall and leukocyte crawling events within the lumen. In cases of uncertainty, events were excluded, which may have led to an underestimation of the total number of leukocyte crawling events observed. This study is a proof of concept. Future studies should include more patients to enable a quantitative assessment of leukocyte dynamics in the pathologic microvasculature. Comparing leukocyte motility across different stages and severities of disease could further elucidate how these cellular behaviors relate to disease progression or resolution and response to treatment. Further work is needed to identify the specific leukocyte subtypes involved in crawling behavior, potentially through immunohistochemical staining or fluorescent tagging, to differentiate cell types and clarify their respective roles in vascular inflammation. Additionally, our averaged time-lapse method did not permit visualization of leukocyte rolling; however, future protocols should be adapted to study this migration phase of the cascade as well. Conclusions We have demonstrated the ability to visualize leukocyte crawling events noninvasively in the living human eye, without exogenous labeling. Clinical imaging of leukocyte crawling dynamics shows promise of revealing a more detailed picture of immune surveillance for managing retinal inflammatory disorders, including many systemic conditions that affect retinal vasculature. Future studies should focus on the diagnostic value of visualizing leukocyte crawling events, not only to elucidate underlying pathology but also to establish high-resolution biomarkers for distinguishing disease regression from progression and shaping future developments of targeted therapies. Supplementary Material Supplement 1 Supplement 2 Supplement 3 Supplement 4 Supplement 5 Supplement 6 Supplement 7 Supplement 8 Supplement 9 Supplement 10 Supplement 11