Visible light optical coherence tomography for assessing retinal structure and function in diseased mouse models

Abstract

The measurement of retinal structure and function is critical for the early diagnosis of blinding eye diseases. Here, we develop a visible light OCT system that integrates high-resolution imaging, GPU-accelerated computing, and automated retinal layer segmentation algorithms that allow for three-dimensional high-definition retinal imaging and blood flow and oxygen saturation measurement for mouse models. We conducted a series of imaging experiments on diseased mouse models, including those with retinitis pigmentosa and glaucoma, and compared them with wild-type mice, and evaluated the OCT imaging results with histological analysis. We observed a reduction in retinal thickness in the retinitis pigmentosa model, along with damage to the inner/outer segment junctions and the outer nuclear layer. In the glaucoma model, the intensity of the ganglion cell layer relative to the internal limiting membrane was reduced. Both disease models exhibited decreased retinal arteriovenous and total blood flow, increased venous oxygen saturation, and a reduced arteriovenous oxygen saturation difference. We also compared the visible light OCT system with an 840 nm spectral domain OCT and a 1060 nm swept source OCT for mouse retinal imaging. The visualization and monitoring of retinal function and metabolic capacity using mouse models through visible light OCT may provide new opportunities for understanding the pathology of ocular diseases.

1. Introduction

The retina, situated in the innermost layer of the eye, uniquely allows non-invasive observation of blood vessels and serves as the sole accessible site for in vivo investigation of the central nervous system (CNS) [1]. Individuals with retinitis pigmentosa (RP) may exhibit disruption of the inner/outer segment junctions [2], decreased thickness of the outer nuclear layer, and signs of retinal pigment epithelial degeneration [3–5]. Additionally, the retina's venous oxygen saturation often increases due to reduced oxygen consumption following photoreceptor loss [6]. Glaucoma, another prevalent retinal disease leading to eventual blindness [7], is characterized by retinal ganglion cell (RGC) death. Research indicates that individuals with glaucoma frequently experience diminished blood flow in various ocular tissues [8], with advanced stages often associated with higher venous oxygen saturation and lower arteriovenous difference [9]. Similarly, in patients with Alzheimer's disease (AD), notable retinal abnormalities such as extensive loss of RGCs [10], reduced retinal nerve fiber layer (RNFL) thickness [11], and diminished retinal blood flow have been observed [12]. Assessing the retina and measuring its functional indicators for early disease screening is crucial to prevent the disease entering irreversible blindness-leading stage.

Animal models are used as surrogates of humans for biomedical research due to the ethical restrictions of working with cell and tissue samples from human. Small animals such as mice are widely used mammalian model systems due to their small size, ease of maintenance and handling, short reproductive cycle, and most importantly, sharing of similar genomic and physiological properties with humans, and the ability to be readily manipulated genetically [13]. Mouse models have already played important roles in many studies such as AD [14], diabetic retinopathy [15], glaucoma [16], and atherosclerosis [17]. In previous studies, electron microscopy and biochemical techniques were often used to visualize the retina of mice [18], which can cause irreversible damage to the model. The ability to noninvasively monitor the entire course of disease progression and treatment response without sacrificing the mouse model can greatly improve our understanding of the disease.

Optical coherence tomography (OCT) is based on the principle of low-coherence interferometry that allows for non-invasive or minimally invasive, real-time, high-speed, high-resolution imaging of living biological tissues, which provides the gold standard for ophthalmic disease diagnosis. To date, most OCT devices have used near-infrared (NIR) light because of its deep penetration and the relative ease of obtaining a commercial light source [19]. With the advent of supercontinuum (SC) light sources [20], visible light OCT (vis-OCT) is beginning to attract significant attention for imaging weakly scattering tissues such as the eye [21–23], where higher resolution images can be obtained compared to the NIR along with an extra range of functional index measurements [24–27]. The earliest report of vis-OCT dates back to 2002 as demonstrated by Považay et al. [28]. Since then, vis-OCT has been employed for retinal structural imaging [21,26,29] and functional imaging [22,30–38] in both small animals such as rat [24,34–38] and mice [39,40], and humans [22,30,31]. Song et al. reported the first vis-OCT angiography (vis-OCTA) for human retinal imaging [41], and they later employed dual-channel OCT to simultaneously acquire vis-OCT and near-infrared OCT images, and validated that glaucoma patients exhibited reduced reflectivity of the peripapillary retinal nerve fiber layer [42]. Meanwhile, Shreesha Rao D. S. et al. concluded that supercontinuum-based spectral-domain optical coherence tomography (SD-OCT) can achieve shot-noise-limited detection and successfully imaged ex vivo mouse retinas [43].

Due to the high axial resolution of OCT and the fact that the extinction coefficients of blood oxyhemoglobin $\left(Hb{O}_2\right)$ and deoxyhemoglobin $\left(Hb\right)$ for visible light are two orders of magnitude higher than those of NIR, vis-OCT is advantageous of performing accurate measurements of blood oxygenation. The integration of blood flow and oxygen saturation enables the quantification of the oxygen metabolic rate within the retinal circulation [35]. In 2013, Yi et al. first demonstrated the application of vis-OCT oximetry in measuring blood oxygenation levels within major retinal arteries and veins in vivo using pigmented rat models [38], followed by a serious of studies using vis-OCT to quantify circulating oxygen saturation $s{O}_2$ in the retina [38,44,45]. In comparison, competitive non-invasive methods for quantifying retinal blood oxygen saturation, such as dual-wavelength and multi-wavelength fundus cameras estimate $s{O}_2$ [46] by measuring the optical density of retinal vessels, which cannot distinguish the spectral signals between the blood vessels and other tissues due to the lack of the depth-resolved imaging capability, leading to the introduction of artifacts such as pigmentation [46]. Other recent techniques, such as photoacoustic ophthalmoscopy for measuring $s{O}_2$ , require eye contact and the use of contrast agents, making them still far from clinical translation [47].

However, current vis-OCT system technology is not sufficiently mature for widespread use for translational applications. The visible light can cause discomfort for human eyes, leading to difficulties of routine retinal imaging in clinic. For small animals, especially for mice, their eye has a much smaller diameter and is limited by lower sensitivity, as well as susceptibility to visible light-induced cataracts and other discomfort. The measurement of retinal blood flow and blood oxygen levels in diseased mice has not been examined thoroughly either.

In this study, we developed a vis-OCT system specifically for small animal imaging that allows for structural and functional measurement of mouse retina with improved designs of imaging interface, real-time volumetric visualization capability, and automated retinal layer assessment. We imaged the mouse retinas of RP and glaucoma, and evaluated the retinal layer thickness, arteriovenous blood flow, and blood oxygen saturation between the diseased models and controls. Histological evaluation was also conducted and compared with the quantitative findings from OCT. Additionally, we compared vis-OCT and NIR OCT systems for mouse retinal imaging and discussed further prospects for the application of vis-OCT.

2. Method

2.1. Vis-OCT imaging system

As shown in Fig. 1(A), we utilized a supercontinuum light source with a central wavelength of 560 nm and a bandwidth of 100 nm (SuperK EXTREME, NKT photonics, Denmark), operating at a repetition rate of 78.1 MHz. Ten percent of the light was directed towards the sample with an incident power of 0.8 mW on pupil, which is within the ANSI laser safety standard [48], while 90% of the light was directed towards the reference arm. The sample arm consisted of a fiber collimator (F240APC-A, Thorlabs, USA), an adjustable focus lens (A-39N0-P16, Edmund, USA) to compensate for individual mouse ocular aberrations, a pair of galvanometer mirrors (Galvo Set, Cambridge Technology, USA) for beam scanning, and a pair of focusing lenses (f1 = 75 mm, f2 = 19 mm, Thorlabs, USA) to reduce the beam diameter entering the mouse eye. Additionally, we added a miniature contact lens after the focusing lens to keep the mouse eye moist and suppress cataracts. The reference arm used the same type of fiber collimator and focusing lens as the sample arm for dispersion compensation. Moreover, a diaphragm was added to the reference arm to strictly control its power, preventing excessive intensity and help reduce noise. Light back-scattered/reflected from the sample arm and reference arm interfered after passing through the coupler, and the interferometric spectrum was collected by a homebuilt spectrometer consisting of a collimating lens, a grating (1200 lines/mm, Thorlabs, USA), a camera lens (85 mm, f/1.8 G, Nikon, Japan), and a high-speed line scan camera (2048 pixels, e2 V, UK) with a line scan rate set to 50 kHz. The axial resolution of the system was measured to be 1.96 μm in retina with an average refractive index of 1.38 [49](Fig. 1(B)). The measured system sensitivity was 89 dB with a -6 dB roll-off range of 0.6 mm (Fig. 1(C)).

Fig. 1.

vis-OCT system and its characterization. (A) Schematic diagram of the system. SC, supercontinuum light source; FC, fiber coupler; CL, collimating lens; DM, diaphragm; FL, focusing lens; M, mirror; TL, tunable lens; GM, galvanometer scanning mirror; L1, mini-contact lens; DG, diffraction grating; LSC, line-scan camera. (B) The impulse response of the system showing an axial resolution of 1.96 μm in tissue. (C) Measurement of system sensitivity roll-off at six depth positions (in air). Imaging depth at -6 dB is approximately 0.6 mm. The point spread functions (PSFs) with different colors represent signals collected at various depths.

A custom-built software was used to perform data acquisition and real-time image display. The main functions included system control, data acquisition, image processing and display which were achieved using C++ with multi-threaded implementation. One thread controlled the synchronization of the system and configured hardware components, including the galvanometer, acquisition card, and camera. With the control software, the line scan camera was triggered by the control electronics for the galvanometer at the start of scanning to achieve synchronization between beam scanning and data acquisition. Since mice develop cataracts relatively quickly after anesthesia, it is necessary to quickly locate the optic nerve head within minutes [50]. Therefore, processing of OCT signals was implemented on a NVIDIA GeForce RTX 2080 graphics card using CUDA such that the B-scan and en face images could be visualized in quasi-real-time to locate the optic nerve head. Specifically, another thread was used for real-time data post-processing, where upon completing the acquisition of a B-scan, it transferred the data to the GPU, where different kernel functions performed window shaping, linear-in-k interpolation, and FFT, and generated the B-scan image. After obtaining all B-scans in the volume, the en face image was constructed on the GPU via maximum or mean intensity projection along the depth direction. At the imaging speed of 50kHz, it took 1.5 seconds to generate a 256 × 256 en face image with GPU acceleration (compared with 7.5 seconds without GPU), which was limited by the A-scan rate of the camera.

2.2. Blood flow measurement

When light interacts with moving blood flow, the frequency of the reflected light changes, and Doppler OCT measures this frequency change to obtain blood flow information [51]. Under the assumption of single scattering particles, when the axial position of the sample arm is $z_S$ and the fixed length of the reference arm is $z_R$ , the interference signal detected by the detector can be represented as $I(k)\propto cos(2(z_R-z_S)k)$ . When the scattering particle undergoes an axial motion, the change in $z_S$ between repeated measurements leads to a phase change, which can be extracted from the acquired interference spectrum via Fourier transform:

$$\mathrm{\Delta }\phi =\frac{4\pi nTvcos\theta }{{\lambda }_0}$$ (1)

where n is the refractive index of the medium; T is the sampling period of the charge-coupled device (CCD) camera; v is the speed of the scattered particle; ${\lambda }_0$ is the central wavelength of the incident light; $\theta$ is the Doppler angle between the particle motion and the incident beam. Blood flow rate F is obtained by taking the product of the vascular blood flow velocity v and the cross-sectional area S perpendicular to the direction of blood flow:

$$F=v\cdot S$$ (2)

2.3. Oxygen saturation measurement

Changes in retinal oxygen consumption affect $s{O}_2$ , which is the ratio of oxyhemoglobin to the total hemoglobin concentration. If quantified accurately, it can potentially serve as a biomarker for monitoring retinal metabolism and provide a valuable early indicator of retinal diseases. Following the work done by Ji Yi et al [25,38], we used a vis-OCT-based method to quantify $s{O}_2$ in the retina, using the contrast in light absorption between $Hb{O}_2$ and $Hb$ . Assuming that the concentrations of $Hb{O}_2$ and $Hb$ , i.e. $C_{Hb{O}_2}$ and $C_{Hb}$ respectively, are homogeneous, the incident light traverses the vessel and is then reflected from the bottom vessel wall and passes twice through the vessel lumen. The reflected intensity $I(z,\lambda )$ can be represented:

$$I(z,\lambda )=I_0(\lambda )R_{0}r(\lambda )e^{-2(z-z_0)[C_{Hb{O}_2}{\epsilon }_{Hb{O}_2}(\lambda )+C_{Hb}{\epsilon }_{Hb}(\lambda )]}$$ (3)

where $z_0$ is the depth of the top vessel boundary with respect to the incident light, $I_0(\lambda )$ is the incident light intensity at wavelength $\lambda$ , $R_0$ is the wavelength-independent reference arm reflectance, $r(\lambda )=A{\lambda }^{-\alpha }$ is the reflectance in the vessel, which is wavelength-dependent, and the extinction coefficient ${\epsilon }_{Hb{O}_2}(\lambda )$ and ${\epsilon }_{Hb}(\lambda )$ are taken from the literature [52].

Taking the natural logarithm of the normalized spectrum yields Eq. (4):

$$\begin{array}{ll}OD(z,\lambda )& =In\left(\frac{I(z,\lambda )}{I_0(\lambda )}\right)\\ & =-2(z-z_0)[C_{Hb{O}_2}{\epsilon }_{Hb{O}_2}(\lambda )+C_{Hb}{\epsilon }_{Hb}(\lambda )]-\alpha ln(\lambda )+ln(A{R}_0)\end{array}$$ (4)

The wavelength-dependent $OD(z,\lambda )$ can be obtained by processing the spectral signal with short-time Fourier transform (STFT). Specifically, we applied a set of Gaussian windows with a width of 15 nm and a spacing of 3 nm to the interference spectrum, resulting in 19 split spectra. Based on the STFT-processed signal, the source spectrum from the anterior and the posterior border (( $I_0(\lambda )$ and $I(z,\lambda )$ in Eq. (3), respectively) of the vessel wall were obtained [53]. The blood oxygen level was determined using a nonnegative least-square fitting model [25]. Subsequently, $s{O}_2$ was calculated according to $s{O}_2=C_{Hb{O}_2}/(C_{Hb{O}_2}+C_{Hb})$ .

2.4. NIR OCT systems

We also compared the vis-OCT with two other home-built NIR systems at a center wavelength of 840 nm and 1060 nm, respectively, specifically for mouse retinal imaging. The 840 nm system is a spectral domain OCT (SD-OCT) system equipped with a broadband light source (cBLMD-T-850-HP, Superlum, Ireland) with a center wavelength of 840 nm and a bandwidth of 160 nm. The system is essentially similar in construction to the vis-OCT system and is capable of achieving an axial resolution of ∼3.95 µm in air and a system sensitivity of 99 dB.

The 1060 nm system is a swept source OCT system (SS-OCT) equipped with a wavelength swept laser (SL102081, Thorlabs, USA) with a center wavelength of 1060 nm and a sweep range of 100 nm. The output beam from the light source was split such that 20% of the light was directed to the sample arm, which employed optical components analogous to those in the visible-light OCT system. Specifically, two achromatic doublet lenses (AC254-075-B-ML and AC080-016-B-ML, Thorlabs, USA) were utilized to accommodate the spot size entering the mouse pupil. In the reference arm, a pair of fiber collimators (F230APC-1064, Thorlabs, USA) formed an adjustable optical delay line. Interference signal was detected by a built-in balanced photodetector inside the light source, and was subsequently acquired with optical clocking to a computer by a high-speed data acquisition card (ATS9373, AlazarTech, Canada). The axial resolution of the SS-OCT system was 13.5 µm in air. The sensitivity of the system was 95.7 dB, and the -6 dB sensitivity roll-off range was 4.44 mm.

2.5. Retinal layer segmentation

We developed a deep learning method for mouse retinal layer segmentation, to automatically delineate ILM (internal limiting membrane), IPL (inner plexiform layer), INL (inner nuclear layer), OPL (outer plexiform layer), ONL (outer nuclear layer), ELM (external limiting membrane), and BM (Bruch's membrane). The AI model is based on ResUNet architecture [54]. Given the scarcity of mouse retinal data, we employed a transfer learning approach. Specifically, we first trained a segmentation model on a dataset of 2,169 human retinal OCT B-scan images annotated with retinal layer segmentation (acquired using Zeiss HD-OCT 5000 and OptoVue Avanti RTVue XR). Subsequently, we fine-tuned this model on a small annotated dataset of vis-OCT mouse retinal images. For quantitative analysis, segmentation was reviewed and errors were corrected by a trained analyst. The automated algorithm was able to reduce the retinal layer thickness analysis time by 95% from 180s to 10s per image.

3. Experimental procedures

3.1. Animal preparation

We imaged three groups of mouse models, wild-type (controls), RP and glaucoma mice, with each group consisting of 5 mice. For controls, male C57BL/6J mice (6-12 weeks old, weighing 20 g or more) were used (Chengdu Lilai Biotechnology Corp. Chengdu, China). To induce RP mouse model, N-methyl-N-nitrosourea (N837326, Mackin, China), a chemically-induced photoreceptor degeneration drug [55] freshly prepared with sterile physiological saline at a dose of 50 mg/kg, was injected into the fundus. Glaucoma mouse model was inducted through intraocular injection of 2μL of 5μM N-Methyl-D-Aspartate (NMDA, M3262, Sigma, USA), which is known to result in the damage of RGCs [56]. The mice were then performed respiratory anesthesia by 2.5% isoflurange (RWD life science, Shenzhen, China) in oxygen and the eyes was in topical anesthesia by Proparacaine Hydrochloride Eye Drops (Alcon Laboratories, Beijing, China). After injection, ointment (Tobramycin Ointment, ALCON CUSI, S.A., ESP) was placed over the eyes. Mice were free fed after recovery, followed by subsequent experiments after 5 days.

Before OCT image acquisition, mice were anaesthetized by intraperitoneal injection of 1% pentobarbital sodium solution (50ug/g). The pupils were well dilated with compound tropicamide eye drops. A 2.5-mm-diameter mini-contact lens with lubricant eye gel (GenTeal Tears, Alcon, USA) was placed on the cornea of the mice to delay drying and suppress cataract. A heating pad was used to maintain body temperature during image acquisition. All experiments were performed in accordance with the protocol approved by the Ethics Committee of the University of Electronic Science and Technology of China (protocol No.: 106142024030728797).

3.2. Imaging experiments

3.2.1. OCT imaging

We employed a series of scan patterns to image a square field of view of 2 × 2 mm² of the mouse retina. For general OCT imaging, 1200 B-scan locations were scanned over a 2-mm range, with each B-scan consisting of 1200 A-lines. For high-definition structural imaging, 20 B-scan locations were scanned over the same 2-mm range, and 50 repetitive B-scans were performed at each location, with each B-scan consisting of 1500 A-lines. For retinal blood flow analysis, 50 B-scan locations were scanned with 10 repetitive B-scans performed at each location, with each B-scan comprising 3000 A-lines to oversample the fast axis for Doppler measurement. For retinal $s{O}_2$ analysis, 500 B-scan locations were scanned, with each B-scan consisting of 3000 A-lines.

3.2.2. Histology and immunohistochemistry

After the mice were sacrificed, eyes were removed and put into 4% paraformaldehyde for 2 hours [57]. Whole-mount retinas were dissected, and embedded in PBS containing Triton X-100 (0.1%, v/v) and donkey serum (10%, v/v). Retinas were incubated overnight with a rabbit-anti Brn3a primary antibody (Abcam, ab245230, UK) at 4℃, and nuclei were stained with DAPI. The secondary antibody was incubated with an Alexa Fluor 594 donkey anti-rabbit (Invitrogen, Life Technologies, USA) for 2 hours in room temperature. Confocal fluorescence microscopy images were taken by a confocal laser microscope LSM 900 (Zeiss). Quantitative analysis was performed using ImageJ software (version 3.8, National Institutes of Health, Bethesda, Washington).

3.3. Statistical analysis methods

Statistical analysis in this study was conducted using GraphPad Prism 8 (GraphPad Software Inc., USA). Mean and standard deviation of each group were calculated. Diseased and control groups were compared using paired t-test if data were normally distributed, or using non-parametric test otherwise. A p-value < 0.05 was considered statistically significant.

4. Results

4.1. Retinal imaging results

The retinal imaging results for controls, RP, and glaucoma mice are shown in Fig. 2. The B-scan images (b-c) clearly show the optic nerve head region, the retinal layering and blood vessel cross-sections with high definition, with a zoomed-in view displayed in (e), where the ultra-high resolution of the vis-OCT imaging system allows for clear resolution of the ten retinal layers. (d) shows the retinal layer segmentation using the proposed algorithm. Examination of the B-scan images of RP mice reveals disruption in the IS/OS and RPE layers, with obvious degenerative cavities, representing lesions (Fig. 2(e2)). This indicates damage to the photoreceptor and pigment epithelial cells. The brightness of the B-scan images of the glaucomatous mice is reduced, but there are still distinct layers visible. In comparison, the glaucoma group exhibits relatively intact retinal layers (Fig. 2(e3)).

Fig. 2.

En face and B-scan images of mouse retina. Representative OCT images of (A) control, (B) RP, and (C) glaucoma mouse retina. Scale bar: 200 µm. Abbreviations: RP: retinitis pigmentosa; ILM, internal limiting membrane; RNFL, Retinal Nerve Fiber Layer; GCL, ganglion cell layer; IPL, inner plexiform layer; INL, inner nuclear layer; OPL, outer plexiform layer; ONL, outer nuclear layer; ELM, external limiting membrane; IS, inner segment junction; OS, outer segment; RPE, retinal pigment epithelium; BM, Bruch's membrane. (a1-a3): Red and green dashed lines indicate two distinct B-scan frames, corresponding to the colored borders of their corresponding B-scans below. (e1-e3): Magnified views of yellow solid-line regions in (c1-c3). (e2): Red arrow: lesions. (e3): Blue arrow: GCL.

4.2. Retinal layer analysis

Figure 3 (A) compares the total retinal thickness (TRT) among the three types of mice, defined as the distance between ILM and BM. The TRT in controls was 216.83 ± 3.25 µm, close to the reported value of 212.3 ± 6.2 µm in the literature [58], confirming the accuracy of our measurements. The ONL thickness was 54.40 ± 1.82 µm in controls, 6.34 ± 3.48 µm in RP, and 53.62 ± 4.36 µm in glaucoma mice (Fig. 3(B)). The average ELM-BM thickness was 62.44 ± 1.94 µm in controls, 52.52 ± 6.47 µm in RP, and 56.38 ± 1.92 µm in glaucoma mice (Fig. 3(C)). Compared to controls, the TRT, ONL and ELM-BM thickness in RP mice was significantly thinner. There was no significant difference between glaucoma and control mice.

Fig. 3.

Retinal thickness statistics in controls, RP, and glaucoma mice. (A) Total retinal thickness; (B) ONL thickness; (C) ELM-BM thickness. ***: p < 0.001; *: p < 0.05. Abbreviations: RP: retinitis pigmentosa; TRT: total retinal thickness; ONL: outer nuclear layer; ELM-BM: between external limiting membrane (ELM) and Bruch's membrane (BM).

The GCL of the retina, adjacent to the ILM, provides valuable optical characteristics reflecting ganglion cell density distribution. Research has shown that the thickness change of the GCL is a potential biomarker for neurodegenerative diseases [59,60]. To better investigate the changes in the retina of mice with ganglion cell injury, we analyzed the intensity of the GCL relative to the ILM by subtracting the ILM intensity from the GCL intensity in logarithmically compressed signals. The thin GCL's proximity to the ILM makes its boundaries in OCT images hard to identify for thickness measurement. We identified the GCL layer by shifting a fixed number of pixels (9 pixels, 9.78 µm) below the ILM [58]. We calculated the average intensity by averaging the intensities of a small window (3 pixels) around this position across all A-lines. Both the control and glaucoma mice followed this procedure for determining the intensity of the GCL and ILM layers. As shown in Fig. 4, the average relative GCL intensity in glaucoma mice was nearly 1.5 dB lower than that of control mice, suggesting a decrease of number or an impairment of function of ganglion cells.

Fig. 4.

GCL intensity relative to the ILM layer in control and glaucoma mice. *: p < 0.05.

4.3. Retinal blood flow measurement

We firstly constructed an in vitro environment to verify the accuracy of blood flow measurements, using a medical syringe pump set to flow rates of 0.1-1.5 mL/h with 0.1 mL/h increments and a capillary glass tube to simulate blood vessels. As shown in Fig. 5(F), the measured flow rate obtained using the vis-OCT system closely matched the expected results (r = 0.99, p < 0.001). After verifying the reliability of Doppler OCT measurement, the same methodology was used to measurement retinal blood flow. By extracting the central coordinates of vascular cross-sections in different B-scans intersecting the vessel, the flow direction and its angle with the z-axis (i.e., the incident light vector) were obtained, yielding the Doppler angle θ in Eq. (1). Then, Δφ can be determined by extracting the phase difference between adjacent A-scans at the segmented vascular cross-section. Finally, blood flow velocity v and flow rate were determined according to Eq. (1) and Eq. (2), respectively.

Fig. 5.

Retinal blood flow measurement in control, RP and glaucoma mice, showing en face, B-scan, Doppler OCT images, and arteriovenous flow statistics. (A) Control mice; (B) RP mice; (C) Glaucoma mice; (D) Comparison of retinal arterial and venous blood flow in three groups; (E) Comparison of total retinal arterial and venous blood flow in three groups. (F) Validation of measured flow rates with set flow rates in an in vitro experiment. Bar: 200 µm. Abbreviations: BF: blood flow; TRBF: total retinal blood flow.

For in vivo measurements, we analyzed the major retinal vessels around the optic nerve head. Figure 5 shows retinal OCT images and blood flow measurements of the three representative mouse models, where the red and green markers indicate arteries and veins respectively and the results are further presented in Table 1. As illustrated in Fig. 5(D)(E), the venous, arterial, and total blood flow in the diseased mice is reduced compared to the controls, with a more pronounced decrease observed in the venous blood flow.

Table 1. Quantitative analysis of retinal blood flow in diseased and control mice. ***: p < 0.001; *: p < 0.05. Abbreviation: BF: blood flow, TRBF: total retinal blood flow.

Type	BF(µL/min)			TRBF(µL/min)
	vein	artery		vein	artery
Control	0.69 ± 0.47	0.71 ± 0.47		3.7 ± 1.84	4.42 ± 1.25
RP	0.30 ± 0.22***	0.51 ± 0.32		1.62 ± 0.26*	2.27 ± 1.06*
glaucoma	0.45 ± 0.34*	0.71 ± 0.30		2.70 ± 1.53	3.98 ± 2.05

4.4. Retinal oxygenation measurement

As shown in Fig. 6, Venous oxygen $\left(V\mathrm{\_}s{O}_2\right)$ was 59.1%±6.9% in controls, 74.8%±2.2% in RP, and 73.7%±2.7% in glaucoma mice. Arterial oxygen $(A\mathrm{\_}s{O}_2)$ was found to be 93.1%±3.6% in controls, 84.4%±2.7% in RP and 80.5%±4.2% in glaucoma mice. Additionally, the arteriovenous difference is notably higher in controls (34.0%) than in RP (9.6%) and glaucoma mice (6.8%). In RP and glaucoma mice, $V\mathrm{\_}s{O}_2$ is increased compared to controls, while $A\mathrm{\_}s{O}_2$ is decreased, resulting in a reduced arteriovenous difference.

Fig. 6.

Oxygen saturation measurements in the retina of control, RP, and glaucoma mice. Abbreviations: A-V: arterial and venous difference.

4.5. Comparison with NIR OCT systems

Figure 7(A)(B) shows the retinal images of the same wild-type C57BL/6J mouse imaged by the vis-OCT system and the 840 nm SD-OCT system. (C)(D) are a side-by-side comparison of the same wild-type C57BL/6J mouse by the vis-OCT and 1060 nm SS-OCT system. Although the other two imaging systems provide better depth penetration, the visible light SD-OCT system clears exhibited significantly higher image resolution, better contrast and clearer retinal layer stratification, which are especially favorable for mouse retinal imaging.

Fig. 7.

Comparison of retinal images of a wild-type C57BL/6J mouse with the (A, C) vis-OCT system, (B) 840 nm SD-OCT system, and (D) 1060 nm SS-OCT system. Bar: 200 µm. (a1-a4): Red and green dashed lines indicate two distinct B-scan frames, corresponding to their matching B-scans with colored borders.

4.6. Histological evaluation of diseased mouse models

After the imaging experiment, we also analyzed the retinas at the pathological level 5 days after drug injection (Fig. 8(A)). By measuring multiple regions at the same plane of pathological sections, we found that the ONL (Fig. 8(B)(D)) in RP mice (13.71 ± 3.4 µm) is significantly thinner compared to the controls (60.6 ± 2.85 µm). Besides, we observed a decrease in the nuclear number of ONL and they became looser, and more irregular, indicating a loss of ONL. RP is typically characterized by progressive visual field loss and vision decline, which may eventually lead to blindness. This is significantly correlated with the loss of retinal photoreceptor cells. The outer nuclear layer, which we assess in our histopathological examinations, is the layer where the cell bodies of the retinal photoreceptor cells are located. The pathological findings are consistent with the observed significant reduction of ONL in RP mice by vis-OCT systems (Fig. 3(B)).

Fig. 8.

Significant pathological changes were observed in both groups of disease models. (A) MNU induced RP and NMDA induced glaucoma mouse models were induced in male C57BL/6J mice (6-12 weeks old). (B) Retina sections from control and RP mice were analyzed by immunohistochemistry (5 retinas collected in each group, 4′, 6-diamidino-2-phenylindole (DAPI) (blue) used as nuclear marker). Bar: 55 µm. (C) Schematic drawing of different retinal layers and their correspondent cell types, and a corresponding OCT image of different retinal layers. (D) The thickness of the outer nuclear layer (ONL) was evaluated in both control and glaucoma mice and correlated with OCT findings. (E) Diagram of the 24 areas (each area size: 0.1 mm²) that were examined in whole mount retina. Purple boxes represent the central area (central), orange boxes represent the middle-peripheral area (middle), and green boxes represent the peripheral area (peripheral). (F-H) 24 Brn3α-labeled RGCs areas (each area size: 0.1 mm², Brn3α (red): an RGC marker) in whole mount retina were quantified, and the total RGC loss was also calculated in both control and glaucoma mice (5 retinas collected in each group). ***: p < 0.001; **: p < 0.01.

RGCs are the principal cells that constitute the GCL [61]. Glaucoma is characterized by the loss of RGCs and optic nerve damage [7]. To examine RGC damage in the glaucoma model, we quantified the RGC loss in retinal whole mount (24 areas, 0.1mm² per area) using immunofluorescence staining for the bodies of RGC marker Brn3a (Fig. 8(E)(F)). We found significant loss of RGC in NMDA induced glaucoma mice compared with controls (Fig. 8(G)(H)). The pathological findings were once again consistent with the vis-OCT findings which showed GCL in glaucomatous mice was significantly thinner (Fig. 4), both demonstrating damage to the inner-layer neurons. However, with vis-OCT, we can potentially inspect the retinal damage in mice at different time points without sacrificing the animal.

5. Discussion

In this study, we have developed a vis-OCT system which offers higher resolution, better contrast, clearer stratification of mouse retinal layers compared to near-infrared OCT systems, and additionally enables the observation of interlayer damage and the measurement of blood flow and arteriovenous oxygenation in diseased mice. Despite the use of contact lenses to delay cataract formation, cataracts can still develop over time that could affect the quality of mouse retinal imaging using vis-OCT, particularly with prolonged anesthesia and exposure of visible light, Thus, we also utilized GPU parallel programming to achieve rapid data preview and acquisition. With the aid of the proposed automatic segmentation algorithm, rapid and extensive quantitative analysis is feasible.

We have found that MNU induced RP mice exhibited a decrease in total retinal thickness, along with degenerative cavities in the IS/OS and RPE layers, which is likely associated with photoreceptor damage, consistent with the findings demonstrated in the literature [2–6]. For NMDA induced glaucoma mice, we found that the intensity of the GCL layer relative to the ILM was decreased, indicating potential damage or reduction in ganglion cells [56]. In terms of blood flow measurement, both RP and glaucoma mice showed a reduction in individual venous and arterial blood flow as well as total blood flow, reflecting changes in vascular function [62]. Regarding oxygen saturation, RP mice exhibited an increase in venous $s{O}_2$ , possibly linked to the reduction and damage of photoreceptor cells with high oxygen consumption, resulting in a decreased arteriovenous difference [6]. Glaucoma mice showed similar patterns, likely due to reduced oxygen consumption associated with neural damage [9].

Due to individual variability among mice and environmental influences, the accuracy of blood oxygen measurements merits further investigation. In addition to the retinal diseases discussed in this study, the capability of vis-OCT to measure retinal thickness and functional indicators can be extended to other diseases that manifest in the retina, such as Alzheimer's disease [11,12]. This potential is currently being explored in ongoing research. Continued advancements in imaging technology, combined with a deeper understanding of the biological significance of the morphological and functional features, will pave the way for improved early diagnosis and treatment strategies for retina-related diseases.

Finally, we compared retinal imaging between vis-OCT and NIR OCT. Overall, vis-OCT offers higher lateral and axial resolution, making it suitable for imaging retinal microstructures in small animals, and it performs better in measuring retinal blood flow and oxygen saturation due to the higher optical absorption in the visible light range. In contrast, NIR OCT is better suited for imaging deeper tissues and is widely used for imaging human eyes.

6. Conclusion

We developed a vis-OCT system capable of retinal imaging in mice, demonstrating high-resolution retinal imaging and quantitative analysis of blood flow and oxygen saturation in various C57BL/6J disease models. Our findings indicate that this system provides clear retinal layer stratification and can effectively detect interlayer damage in diseased mice. Additionally, the blood flow and oxygen saturation measurements show significant differences between diseased mice and wild-type controls. With these capabilities, our proposed method holds great potential for the early evaluation of blinding eye diseases.

Funding

National Natural Science Foundation of China 10.13039/501100001809 ( 62075033, 62135002); Sichuan Provincial Science and Technology Support Program 10.13039/100012542 ( 2023YFS0022, 2024YFHZ0174); Medico-Engineering Cooperation Funds from University of Electronic Science and Technology of China ( ZYGX2021YGCX004); Fundamental Research Funds for the Central Universities 10.13039/501100012226 ( ZYGX2021J009).

Disclosures

The authors declare no conflicts of interest.

Data availability

Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.