Vascular permeability (VP) is the ability of blood vessel walls to permit small molecules like drugs and ions, and cells such as lymphocytes^[1,2]. Normally, endothelial cells act as a selective barrier, allowing limited passage of small molecules while restricting macromolecules and cells^[3]. However, in pathological conditions, vascular barrier damage increases macromolecule and cell leakage, causing edema, inflammation, and disease progression^[4,5]. In liver fibrosis, increased capillary permeability reduces plasma albumin levels^[6,7]. Similarly, kidney diseases alter capillary permeability, leading to albuminuria and edema, notably in end-stage renal disease with microvascular leakage and fibrosis^[8,9]. Thus, VP changes are closely associated with hepatic and renal fibrosis, making VP evaluation vital for early detection and monitoring fibrosis progression.

A common method for assessing VP involves intravenously injecting Evans Blue (EB), an azo dye with peak absorption at 610 nm, and using colorimetric spectrophotometry^[10,11]. EB binds to serum albumin, and under normal conditions, their complex remains within the vasculature. Increased VP, however, causes this complex to leak into extravascular tissues, visibly intensifying blue staining in areas with heightened permeability^[12,13]. Magnetic resonance imaging (MRI) is distinguished by its exceptional soft tissue contrast and proficiency in visualizing anatomical structures. However, its utility is limited by suboptimal temporal resolution and the necessity for contrast agents to accurately assess VP. Furthermore, MRI is associated with elevated costs and may not be readily available in certain settings. Positron emission tomography (PET) is highly effective for functional imaging, offering quantitative insights into metabolic activity and blood flow. Nonetheless, its use is often constrained by the need for radiolabeled substances, which pose safety concerns due to ionizing radiation exposure. Computed tomography (CT) provides rapid imaging with superior spatial resolution, rendering it particularly advantageous for the evaluation of vascular structures and pathologies. However, CT typically depends on iodinated contrast agents, which can provoke allergic reactions in some patients and may be unsuitable for those with impaired renal function^[14–17]. In contrast, photoacoustic microscopy (PAM) offers high-contrast, high-resolution imaging with superior spatial resolution, tissue penetration depth, label-free imaging capability, and dynamic monitoring of blood flow and oxygenation status, rendering it particularly suitable for imaging tissue microvasculature^[18–20]. Furthermore, PAM’s effectiveness in quantitative VP assessment has been demonstrated in studies investigating blood–brain barrier integrity^[21,22].

In this study, we employed dual-wavelength PAM to quantitatively evaluate VP in the context of liver and kidney fibrosis. By utilizing the distinct absorption spectra of hemoglobin and EB at wavelengths of 532 and 610 nm, respectively, we effectively assessed changes in the microvascular structure and permeability. By spectral unmixing techniques we demonstrated the efficacy of this methodology in investigating alterations in VP in fibrotic liver and kidney tissues.

Female ICR mice, aged 6 to 8 weeks and weighing about 20 g, were obtained from the Guangdong Medical Laboratory Animal Center. They were housed in individually ventilated cages at 24°C with a 12-h light/dark cycle and had free access to food and water. Twelve mice were randomly assigned to three groups: liver fibrosis ($n=3$), unilateral ureteral obstruction (UUO, $n=3$), and control ($n=6$). Liver fibrosis was induced using carbon tetrachloride (${\mathrm{CCl}}_4$), while renal fibrosis in the UUO group was modeled by left ureteral ligation, following established protocols^[23–25]. All experiments complied with the Ethics Committee guidelines of Guangdong Provincial People’s Hospital.

532 nm is a commonly used wavelength for imaging hemoglobin, as hemoglobin exhibits strong absorption at this wavelength, enabling efficient capture of hemoglobin photoacoustic signals in blood vessels. We measured two absorption peaks for blood at 540 and 577 nm [Fig. 1(b)]. To avoid potential overlap between the hemoglobin signal and EB detection at 610 nm, and considering the practicality of using 532 nm as a standard wavelength for commercial lasers, we chose 532 nm for hemoglobin imaging. 610 nm corresponds to the absorption peak of EB, which is commonly used to assess VP. At this wavelength, hemoglobin absorption is significantly reduced, minimizing interference from hemoglobin on EB signals. Therefore, 532 and 610 nm were selected for spectral unmixing of hemoglobin and EB in this study.

As shown in Fig. 2(a), the PAM imaging system utilizes two lasers to generate pulsed excitation light at 532 and 610 nm. The first laser (VPFL-G-HE-30, Spectra-Physics) emits pulsed light at 532 nm, while the second laser (AO-N-532, CNI) produces 610 nm light via a Raman effect in a 10 m single-mode fiber (HB450-SC, FiberCore). The 610 nm light is then filtered using a narrow bandpass filter. Both lasers operate at 10 kHz, with a 500 ns delay between pulses to differentiate the photoacoustic signals from the two wavelengths.

To focus the combined laser beams to a spot size of approximately 3.2 µm, a short-pass dichroic mirror (DMSP550, Thorlabs) and a 50 mm focal length achromatic doublet lens (AC127-050-A, Thorlabs) are used. A custom photoacoustic beam combiner ensures coaxial alignment of excitation light and acoustic waves, utilizing uncoated right-angle prisms, aluminum-coated prisms, and plano–concave lenses. To reduce aberrations from mirror reflections, a top correction lens is positioned before the excitation light enters the beam combiner. The excitation light reflects off the aluminum-coated sample surface, while the generated acoustic wave is focused by the plano–concave lens and detected by an ultrasound transducer (V214-BC-RM, OLYMPUS), which converts the acoustic signal into an electrical signal. The transducer operates at a center frequency of 50 MHz, with a $-6\mathrm{dB}$ bandwidth of about 70% and a sensitivity of $-13\mathrm{dB}$.

To evaluate the system resolution, a blade was scanned using the 532 and 610 nm lasers with incremental steps of 3 and 5 µm, respectively, at a scanning speed of 5 mm/s. The one-dimensional X-scan signal obtained from the blade’s photoacoustic image, as shown in Fig. 2(b), was analyzed to compute the transverse resolution. This analysis, which involved Gaussian fitting, revealed a full width at half-maximum (FWHM) of 8.06 µm at 532 nm and 14.23 µm at 610 nm, respectively, as presented in Fig. 2(c).

Mice were anesthetized with 2% isoflurane for imaging procedures. For liver imaging in the one-week fibrosis model, a 5 mm abdominal incision exposed the liver lobe, and a 532 nm laser was used for structural imaging. The parameters included a scan range of $2\mathrm{mm}\times 2\mathrm{mm}$, a step size of 5 µm, and a laser energy of 200 nJ per pulse. For renal imaging in the one-week UUO model, a 5 mm incision was made along the right side of the spine to access the kidneys, with scanning parameters set to an imaging range of $0.9\mathrm{mm}\times 0.9\mathrm{mm}$, a step size of 3 µm, and a laser energy of 250 nJ per pulse.

Spectral unmixing was performed using excitation wavelengths of 532 and 610 nm to differentiate between hemoglobin and EB concentrations in microvessels. Photoacoustic signals were acquired and processed with MATLAB (R2022b, MathWorks) for spectral unmixing analysis^[26–29]. To effectively reduce high-frequency noise and outliers, a combined strategy of Gaussian and median filtering was applied. The algorithm employs a two-wavelength linear mixed model based on non-negative constrained least squares (NNLS) [refer to Eqs. (1) and (2)] to effectively separate hemoglobin and EB signals. The threshold for hemoglobin signal detection is specifically set to exclude background plasma interference, while the EB signal threshold is adjusted to minimize false positives from low-concentration signals. These thresholds are calibrated based on pre-experimental data. A binarized mask based on vessel diameter ($>10\mathrm{\mu m}$) is applied to filter out signals from non-relevant regions, thereby reducing unmixing errors. The concentrations and distributions of EB were calculated using the following equations: $$P_{532}=2.303\eta \mathrm{\Gamma }·[\frac{e_{532-\mathrm{Hb}}}{M_{\mathrm{Hb}}}·{\rho }_{\mathrm{Hb}}+\frac{e_{532-\mathrm{EB}}}{M_{\mathrm{EB}}}·{\rho }_{\mathrm{EB}}]·F_{532},$$(1)$$P_{610}=2.303\eta \mathrm{\Gamma }·[\frac{e_{610-\mathrm{Hb}}}{M_{\mathrm{Hb}}}·{\rho }_{\mathrm{Hb}}+\frac{e_{610-\mathrm{EB}}}{M_{\mathrm{EB}}}·{\rho }_{\mathrm{EB}}]·F_{610},$$(2)where $P_{532}$ and $P_{610}$ are the photoacoustic signals at 532 and 610 nm, $M_{\mathrm{Hb}}$ and $M_{\mathrm{EB}}$ are the molar masses of hemoglobin and EB, $F_{532}$ and $F_{610}$ represent the laser intensities at 532 and 610 nm, $e$ is the molar absorption coefficient at a specific wavelength, and ${\rho }_{\mathrm{Hb}}$ and ${\rho }_{\mathrm{EB}}$ are the concentrations of hemoglobin and EB.

To evaluate the spectral unmixing algorithm, simulations were conducted using a phantom model that mimicked conditions post-EB injection. Blood samples were prepared by mixing 950 µL of newborn bovine blood with 50 µL of saline for the control group. In the experimental groups, 950 µL of blood was combined with 50 µL of EB solutions at concentrations of 8, 12, 16, 20, and 24 mg/mL, resulting in final EB concentrations of 0.4, 0.6, 0.8, 1.0, and 1.2 mg/mL. These mixtures were injected into a 0.3 mm diameter transparent tube submerged in water to simulate blood vessels.

Microvascular permeability in the livers and kidneys was assessed with EB dye^[30]. An indwelling needle was placed in the tail vein and connected to an EB solution, with a syringe pump injected EB steadily. After baseline imaging, a 2% EB solution at 5 mL/kg was used for liver imaging, and a 1% EB solution at 1 mg/kg was used for kidney imaging. Images were taken every minute for 15 min post-injection, with EB distribution and concentration quantified via spectral unmixing.

Statistical significance was evaluated using a two-tailed Student’s t-test, with a P-value of less than 0.05 considered significant. Prior to conducting the t-test, we verified the assumptions of normality and homoscedasticity. Normality was assessed using the Shapiro–Wilk test for each experimental group, with all p-values exceeding 0.05, indicating that the data conformed to a normal distribution. Homoscedasticity was tested using Levene’s test, which showed no significant differences in variances between the groups. To ensure the accuracy of our results, we applied the Bonferroni correction when conducting multiple comparisons. Specifically, we adjusted the significance level ($\alpha =0.05$) by dividing it by the number of comparisons made. All analysis was performed using SPSS software (version 27, IBM).

In the phantom study, red blood samples and a blue mixture (blood with EB) were prepared to simulate hemoglobin and EB-albumin complexes, respectively [Fig. 1(a)]. The absorption spectra, shown in Fig. 1(b), revealed strong absorption peaks at 540 and 577 nm for blood, and a distinct peak at 610 nm for EB. Using a dual-wavelength PAM system, we acquired photoacoustic images at both wavelengths [Fig. 1(c)]. A spectral unmixing algorithm was then applied to extract the concentration distributions of blood and EB, as displayed in Fig. 1(d).

The results showed that, at EB concentrations below 0.6 mg/mL, the spectral unmixing technique could not accurately distinguish the EB signal from the blood background. However, above 0.6 mg/mL, the extracted EB signal intensity increased proportionally with concentration, aligning well with theoretical predictions. Linear regression revealed a high correlation coefficient for EB concentration, with an $R^2$ value of 0.99, as illustrated in Fig. 1(e).

This study employed ${\mathrm{CCl}}_4$-induced liver fibrosis and UUO models in mice to investigate fibrosis-related changes in microvascular structure using PAM imaging. PAM images of healthy liver tissue revealed uniformly distributed microvessels with a well-defined lobular architecture [Fig. 3(a)]. The portal vein and hepatic artery branches converge into the central vein, forming a hexagonal lobular zone that supports efficient blood circulation. In the one-week fibrosis model, hepatic lobular structures appeared disorganized, featuring sparse and irregularly arranged vasculature with malformed small vessels [Fig. 3(b)].

PAM images of healthy kidneys showed evenly distributed peritubular capillaries [Fig. 3(c)], while UUO kidneys displayed sparse, disrupted capillaries and prominent deep macrovessels [Fig. 3(d)]. We conducted a multiparametric analysis of vascular changes, processing hemoglobin distribution images to extract vascular networks. Results showed increased mean vessel diameter (control: $24.0\mathrm{\mu m}\pm 1.5\mathrm{\mu m}$; fibrosis: $35.5\mathrm{\mu m}\pm 3.6\mathrm{\mu m}$, $P<0.05$) and decreased vessel density (control: $71.1\%\pm 2.1\%$; fibrosis: $48.0\%\pm 1.5\%$, $P<0.01$) in fibrotic livers [Fig. 3(e)]. In UUO kidneys, the mean vessel diameter also increased (control: $33.3\mathrm{\mu m}\pm 1.2\mathrm{\mu m}$; UUO: $36.8\mathrm{\mu m}\pm 1.2\mathrm{\mu m}$, $P<0.05$), with a reduction in vessel density (control: $63.6\%\pm 0.7\%$; UUO: $61.5\%\pm 0.9\%$, $P<0.05$) [Fig. 3(f)]. These results confirm fibrosis disrupts microvascular morphology in both livers and kidneys.

Under physiological conditions, EB is confined to the vascular system but can diffuse through the endothelial barrier at high concentrations^[31]. The spectral unmixing algorithm indicated increased microvascular permeability in fibrotic livers, allowing for accurate differentiation and visualization of EB extravasation. The control group exhibited minimal EB leakage, indicating preserved vascular integrity [Fig. 4(a)]. In the one-week fibrosis model, significant EB accumulation occurred in the interstitial space, with slight alignment along microvessels [Fig. 4(b)]. Quantitative analysis revealed a notable increase in EB extravasation in the fibrosis model compared to a healthy liver [Fig. 4(c)]. Fibrosis reduced the hemoglobin percentage area (control: $61.1\%\pm 3.9\%$; liver fibrosis: $36.7\%\pm 2.6\%$, $P<0.001$) [Fig. 4(d)], while the EB extravasation percentage area at 15 min post-injection was significantly higher in a fibrotic liver (control: $20.0\%\pm 2.6\%$; liver fibrosis: $30.7\%\pm 1.5\%$, $P<0.01$) [Fig. 4(e)].

After the administration of EB via the tail vein, we assessed renal microvascular permeability using time-lapse imaging. Control kidneys showed minimal EB extravasation [Fig. 5(a)], while UUO-treated kidneys exhibited increased peritubular capillary diameter and significant EB leakage along microvessels, with minor aggregation in branching capillaries [Fig. 5(b)].

Quantitative analysis indicated significantly higher EB extravasation in the fibrotic model compared to healthy kidneys [Fig. 5(c)], although the visual contrast of EB permeability was less distinct in the kidneys than in the livers. This discrepancy may be attributed to increased intrarenal pressure in the UUO model. Furthermore, fibrosis led to a reduction in the percentage area of hemoglobin (control: $74.8\%\pm 4.6\%$, UUO: $63.8\%\pm 3.4\%$, $P<0.01$) [Fig. 5(d)], while the percentage area of EB extravasation at 15 min post-injection was significantly greater in the fibrotic kidney compared to the control group (control: $15.3\%\pm 1.7\%$, UUO: $26.2\%\pm 2.1\%$, $P<0.01$) [Fig. 5(e)], displaying a pattern similar to that observed in the liver.

Histological analysis confirmed the successful establishment of liver and kidney fibrosis. Hematoxylin and eosin (HE) staining revealed disorganized hepatocytes with degeneration, necrosis, and inflammatory cell infiltration in fibrotic livers [Fig. 6(a)]. Masson staining showed increased collagen deposition within the fibrotic liver tissue. In the UUO model, HE staining indicated renal tubule atrophy with expanded interstitial spaces and inflammatory infiltration [Fig. 6(b)]. Masson staining reflected an enlarged area of positive staining, indicating increased collagen content in the renal tissue.

Microvascular permeability is essential for maintaining in vivo homeostasis^[32]. Endothelial cells create a dynamic barrier under physiological conditions that allows small molecules to pass. Pathological conditions can compromise this barrier, resulting in the leakage of larger molecules, such as plasma albumin^[33,34]. Increased VP is associated with various diseases, including hypoalbuminemia, as the liver is a primary site for albumin synthesis^[35,36]. This study utilized dual-wavelength PAM to monitor and quantify structural and permeability changes in the renal microvasculature with liver fibrosis and UUO. Previous research has employed PAM with EB dye to assess blood–brain barrier integrity^[21,37]. In those studies, EB was administered based on body weight, and time-lapse imaging quantified dye extravasation by counting pixels outside the vasculature. While this approach minimizes background interference, it may also reduce the useful signal, compromising sensitivity, signal-to-noise ratio, and accuracy^[38]. In our study, we implemented a dual-wavelength PAM system with lasers at 532 and 610 nm, leveraging their distinct absorption characteristics to accurately calculate the content and distribution of EB. Our findings align with prior studies, showing microvascular changes in fibrotic liver and kidney tissues^[23,39–42]. After one week of fibrosis induction, the liver microvasculature exhibited distortion, characterized by increased vessel diameter and decreased density. In UUO-induced renal fibrosis, peritubular capillaries displayed structural disorganization.

PAM offers high-resolution microvascular imaging; however, its limited penetration depth poses challenges for visualizing deeper tissues, restricting our ability to image deeper vessels. Future studies could address these limitations by reducing optical scattering or by combining PAM with photoacoustic computed tomography for multi-scale imaging^[43–46]. Additionally, although PAM effectively utilized spectral unmixing to quantitatively assess microvascular permeability in the liver and kidney, in vitro experiments indicated that EB signals were undetectable below 0.6 mg/mL. This highlights the need for further optimization of the spectral unmixing algorithm to enhance sensitivity and accuracy.

In conclusion, this study examined microvascular structures and permeability alterations in liver fibrosis and UUO models. By implementing a dual-wavelength PAM system at 532 and 610 nm and applying spectral unmixing techniques, we quantitatively evaluated changes in VP during fibrotic progression. Our results highlight the methodological viability and potential of this approach in elucidating abnormal permeability mechanisms, informing future therapeutic strategies aimed at preserving the liver and kidney microvasculatures.