Microbubble-assisted pLIFU enhances riboflavin delivery and corneal cross-linking in an in vivo rabbit keratoconus-like model

Abstract

Epithelium-off corneal collagen cross-linking (CXL) slows keratoconus progression but disrupts the epithelial barrier, whereas epithelium-on CXL is constrained by limited stromal riboflavin loading and insufficient evidence that physical delivery enhancement improves CXL outcomes in vivo keratoconus-like ectasia. Here, we developed microbubble-assisted pulsed low-intensity focused ultrasound (pLIFU) to enhance riboflavin transport through cavitation and evaluated its therapeutic effect in a collagenase-induced rabbit keratoconus-like model. Passive cavitation detection and temperature monitoring identified a treatment condition of 1 Hz pulse repetition frequency, 20 % duty cycle, 1.43 MPa focal-zone peak negative pressure, and 200 μl SonoVue solution, with maximum temperature rise limited to 1.5 °C. Hydrophone measurements showed low posterior-segment acoustic action, with a retinal-position mechanical index of approximately 0.076. Confocal fluorescence imaging showed that pLIFU with microbubbles increased peak stromal riboflavin fluorescence by 9.27 ± 2.16-fold (p < 0.0001) and effective penetration depth by 4.09-fold (p < 0.001) relative to control. In vivo, pLIFU-microbubble-enhanced CXL produced greater corneal flattening and a larger shear-wave-speed response than the US group. One month after treatment, central corneal curvature decreased by 45.85 ± 6.32 % and shear-wave speed ($c_s$) increased by 52.50 ± 12.89 % relative to pretreatment in the USMB group. Short-term histology and inflammatory cytokine assays showed no sustained inflammatory elevation or overt retinal laminar disruption within the observation window. These findings support the early feasibility of pLIFU cavitation-enhanced riboflavin delivery as a physical augmentation strategy for epithelium-on CXL in keratoconus-like ectasia, with favourable short-term ocular tissue tolerance for the assessed endpoints; they do not establish non-inferiority to epithelium-off CXL or long-term safety and durability.

1. Introduction

Keratoconus is a progressive corneal ectasia characterised by focal stromal thinning, corneal protrusion, irregular astigmatism, and progressive visual loss [1], [2]. In cases of advanced disease, corneal transplantation may ultimately be necessary; however, this procedure is constrained by factors including donor availability, the risk of rejection and infection, and the necessity for long-term postoperative management [3]. The clinical rationale for early-to-moderate keratoconus treatment is therefore to increase corneal resistance to ectatic deformation before irreversible thinning and scarring occur and to reduce the future need for transplantation [4], [5]. Corneal collagen cross-linking (CXL) represents the only established intervention with the capability to slow or halt keratoconus progression [6], [7], [8], [9]. Standard CXL involves the removal of the central epithelium to allow for riboflavin penetration prior to UVA (wavelengths 320–400 nm) photoactivation. However, it has been observed that epithelial debridement can lead to disruption of the corneal surface barrier, resulting in an increased risk of pain, infection, inflammation, haze, and delayed epithelial healing [[10], [11]]. The epithelium-on CXL procedure was introduced to reduce these epithelial complications; however, the intact epithelium and tight junctions strongly restrict hydrophilic riboflavin transport, leading to weaker cross-linking depth and less consistent stiffening than the epithelium-off procedure [12], [10], [11], [13]. A plethora of research has been conducted on the utilisation of chemical enhancers, iontophoresis, electroosmosis, and carrier-based systems with the objective of enhancing ocular drug delivery. However, the implementation of these methodologies is constrained by factors such as epithelial toxicity, incomplete delivery, and procedural complexity [[10], [11], [13], [14], [15]]. A physical method that improves stromal riboflavin delivery while preserving the epithelial barrier is a critical need for safe and effective epithelium-on CXL.

Pulsed low-intensity focused ultrasound (pLIFU) provides a spatially targetable acoustic stimulus which can modulate tissue permeability via cavitation-related mechanical effects, microstreaming, acoustic radiation force, local shear stress and transient remodelling of intercellular junctions [16], [17], [18], [19], [20]. To our knowledge, there is limited direct journal evidence on pLIFU-mediated transepithelial riboflavin delivery. The existing evidence on corneal riboflavin delivery is mainly based on non-focused or low-frequency, low-intensity ultrasound systems. Lamy et al. and Sun et al. demonstrated that ultrasound can facilitate riboflavin absorption into the corneal stroma and induce a certain degree of corneal stiffness enhancement following CXL [21], [22]. However, ultrasound alone usually requires longer treatment times, higher acoustic input or lower frequencies, and may be accompanied by superficial epithelial changes or a local temperature rise [21], [22]. Therefore, while ultrasound-only delivery demonstrates the feasibility of physical enhancement for corneal riboflavin delivery, it is challenging to meet the requirements of delivery efficiency, cavitation monitoring and thermal control for safe and effective CXL augmentation under in vivo disease conditions simultaneously.

Microbubbles can act as exogenous cavitation nuclei, amplifying the mechanical effects of ultrasound at low acoustic intensity. Under pulsed ultrasound, microbubbles oscillate stably or transiently collapse, generating microstreaming, shear stresses, transient membrane pores and junctional openings that can increase local transport [23], [24], [25], [26], [27]. However, direct studies of ultrasound–microbubble treatment in corneal tissue remain scarce, with most research focusing on plasmid or gene delivery rather than riboflavin delivery [23], [24], [25], [26], [27], [28], [29]. Although a conference abstract has reported on the use of high-intensity ultrasound with microbubbles for corneal riboflavin phonophoresis, this approach differs from that of low-intensity focused ultrasound [30]. Furthermore, parameters currently utilised for ocular ultrasound-mediated delivery vary significantly. Frequencies range from 20 kHz [31] to 2.25 MHz [32], while peak negative pressures range from 0.10 MPa [33] to 3.50 MPa [30], [34]. Pulse repetition frequencies (PRF) range from 1 Hz [33], [34] to 1000 Hz [35], pulse durations range from 10 ms [34], [36] to 500 ms [35], and treatment durations range from 5 min [36] to 60 min [31]. Additionally, there is a lack of standardised protocols regarding the type and dosage of microbubbles used [34], [35], [36], [37]. As the corneal epithelium, blood-retinal barrier and sclera differ markedly in terms of their tissue architecture, acoustic propagation pathways, drug transport routes and safety constraints, parameters from existing ocular ultrasound-microbubble studies cannot be transferred directly to pLIFU-microbubble-mediated stromal riboflavin delivery. To achieve safe and efficient stromal riboflavin delivery under epithelium-on conditions, therefore, the effects of acoustic pressure, duty cycle, pulse repetition frequency and microbubble dose on delivery efficiency, cavitation activity and temperature rise need to be systematically defined. This will provide a rational basis for pLIFU-microbubble-enhanced CXL [38], [39], [40].

Cavitation is central to microbubble-assisted delivery; however, excessive transient cavitation can generate undesirable thermal and mechanical effects, including high local pressure, shock waves and microjets [40], [23], [24], [25], [26], [27]. This is a particular concern in the eye because ocular tissues are small, spatially confined and sensitive to thermal or mechanical perturbation. Previous studies of ultrasound in the eye have reported that high acoustic pressure or intense microbubble activity can cause retinal haemorrhaging, inflammation, or structural abnormalities [33], [34], [41]. Thermal analyses of ultrasound-mediated ocular delivery further indicate that temperature rise depends on frequency, intensity and insonation time, and that corneal temperature elevation can alter epithelial stress-response proteins [42], [43]. The WFUMB thermal safety recommendations advise against increasing the temperature of ocular tissue by more than 1.5 °C above physiological levels [44]. Therefore, riboflavin delivery efficiency alone is insufficient as an evaluation endpoint. For pLIFU-microbubble delivery, efficient delivery requires moderate cavitation activity; however, excessive transient cavitation can induce undesirable thermal effects and tissue damage, thereby reducing safety. Existing ocular ultrasound-microbubble studies have inadequately monitored cavitation and temperature during delivery, and no study has simultaneously evaluated cavitation signals, temperature increase, and corneal riboflavin distribution. Passive cavitation detection can record harmonic, subharmonic, ultraharmonic and broadband noise signals to help distinguish between stable and transient cavitation activity. This provides quantitative evidence to inform the selection of safe and effective acoustic parameters and microbubble doses [40], [45], [46], [47].

Furthermore, existing studies on the use of low-intensity ultrasound to enhance corneal delivery have mainly focused on normal corneas or those of healthy animals, with limited in vivo validation of disease models. The endpoints of these studies have primarily included riboflavin fluorescence intensity, tissue penetration depth, drug permeability, and short-term epithelial changes [16], [21], [22]. While these studies demonstrate that ultrasound can enhance delivery, they do not address whether this can lead to improvements in curvature, thickness stabilisation and stiffness enhancement after CXL in a pathologically ectatic cornea. Compared with genetic, light-induced or oxidative stress models, collagenase II treatment can induce elevation of corneal curvature, central corneal thinning, collagen disorganisation and mechanical weakening within a relatively short time, approximating several key pathological features of keratoconus [48], [49], [50]. For keratoconus therapy, demonstrating riboflavin entry into the corneal stroma is insufficient to support therapeutic efficacy. A more critical question is whether enhanced delivery can lead to quantifiable improvements in corneal morphology and stiffness after CXL in an ectatic disease state [7], [8], [9]. High-frequency B-mode ultrasound, shear-wave elastography and optical coherence elastography can track changes in corneal thickness, curvature and stiffness. This provides a technical basis for in vivo therapeutic evaluation [51], [52], [53], [54], [55]. However, existing research lacks systematic in vivo validation of the therapeutic efficacy of pLIFU combined with microbubble-enhanced CXL under pathological conditions. Furthermore, there is a lack of safety assessments that simultaneously integrate changes in corneal morphology, corneal stiffness, posterior segment histology, and inflammatory markers. In other words, further clarification is needed on whether pLIFU combined with microbubbles can safely and effectively treat keratoconus.

In summary, the key question is not just whether ultrasound can promote the entry of riboflavin into the cornea via the epithelium, but also whether pLIFU-microbubble-enhanced riboflavin delivery can further improve the efficacy of epithelium-on CXL in a pathologically ectatic cornea while ensuring the safety of ocular tissue. To address this, we first identified a pLIFU-microbubble condition in ex vivo corneas with a maximum temperature rise of no more than 1.5 °C. We then applied this condition to a collagenase-induced rabbit keratoconus-like model. We used high-frequency B-mode ultrasound, shear-wave elastography, histology and inflammatory cytokine assays to evaluate post-CXL corneal curvature, SWE-derived apparent stiffness and the short-term ocular tissue response. This study aims to demonstrate that microbubble-assisted pLIFU can serve as a physical adjunct that combines efficient delivery, temperature-rise limitation and short-term tissue tolerance to achieve safe and effective CXL therapy for keratoconus-like ectasia by directly embedding enhanced stromal riboflavin delivery into therapeutic evaluation in an in vivo disease model.

2. Results

2.1. Temperature-rise-limited riboflavin delivery enhancement

First, we established a parameter-selection framework that treated cavitation activity and temperature rise as coupled outputs rather than independent optimisation targets (see Fig. 2). Fig. 2 illustrates the changes in cavitation activity and temperature under various pLIFU parameters and microbubble doses. The control group exhibited only background acoustic signals (Fig. 2a), whereas the US group produced limited cavitation responses, with transient cavitation intensity primarily maintained at around 2000 to 2500 a.u. (Fig. 2b and 2f). Following the addition of microbubbles, the USMB group exhibited significantly increased broadband noise, with an initial peak approaching 7,000 a.u., which was approximately 2.8 to 3.5 times higher than that observed in the US group (see Fig. 2c and 2g). This enhanced cavitation response was maintained in the presence of riboflavin (Fig. 2d and h), indicating that the microbubbles were the primary factor in amplifying the pLIFU-induced cavitation activity. Under the selected condition, the peak inertial cavitation metric (ICM_PV) was 6028.55 ± 1047.96 a.u. with riboflavin (USMB-Rib) and 6363.98 ± 2137.38 a.u. without riboflavin (USMB-NoRib), reported as mean ± s.d. The value with riboflavin was 335.43 a.u. (5.27%) lower; this descriptive difference did not indicate loss of microbubble-mediated acoustic responsiveness.

Fig. 2.

Temperature-rise-limited cavitation optimization defines effective pLIFU–microbubble treatment parameters. a–d, Representative passive acoustic emission spectrograms acquired during control treatment, pLIFU alone, pLIFU with microbubbles, and pLIFU with microbubbles plus riboflavin. e–h, Corresponding time-resolved cavitation intensity profiles showing the temporal evolution of acoustic activity under each condition. i–l, Quantitative effects of pulse repetition frequency, duty cycle, focal-zone peak negative pressure and microbubble volume on cavitation magnitude and temperature rise. The selected condition, 1 Hz pulse repetition frequency, 20 % duty cycle, 1.43 MPa focal-zone acoustic pressure and 200 μl SonoVue® solution, provided robust cavitation activity while maintaining the maximum temperature rise within 1.5 °C. Data are presented as mean ± s.d.; statistical significance is indicated in the panels. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Figs. 2i to 2l show a further comparison of the effects of PRF, duty cycle, focal-zone peak negative pressure and microbubble dose on transient cavitation and temperature rise. A PRF of 1 Hz, a duty of 20% and an acoustic pressure in the focal zone of 1.43 MPa produced strong transient cavitation while keeping the maximum temperature rise to within 1.5 °C. However, when the duty was increased to 30%, or the acoustic pressure was increased to 2.15 MPa, the maximum temperature rise exceeded the preset thermal control requirement of 1.5 °C. Regarding the microbubble dose, a volume of 200 μl of SonoVue solution produced the strongest cavitation response; increasing the dose to 300 μl reduced the cavitation signal (see Figs. 2i to 2l). Therefore, a PRF of 1 Hz, a duty of 20%, a focal-zone peak negative pressure of 1.43 MPa and a SonoVue solution dose of 200 μl were selected for subsequent delivery and in vivo CXL experiments. This condition was chosen to achieve effective cavitation activity while controlling temperature rise rather than simply maximising cavitation intensity.

Microbubble-assisted pLIFU increased both the intensity and depth of stromal riboflavin delivery under temperature-rise-limited conditions (Fig. 3). Confocal fluorescence imaging showed weak riboflavin signals in the control group, moderate superficial-to-mid stromal fluorescence in the US group, and the strongest and deepest riboflavin distribution after pLIFU with microbubbles (Fig. 3a to c). The depth-resolved profiles confirmed this graded response: fluorescence decayed rapidly in controls, increased modestly under low-duty-cycle or low-pressure microbubble conditions, was further enhanced by pLIFU alone, and reached the highest level under the selected 20% duty cycle, 1.43 MPa pLIFU–microbubble condition (Fig. 3d). Peak fluorescence intensity was 5.26 ± 0.68 a.u. (p < 0.0001) in controls, 17.90 ± 1.68 a.u. (p < 0.0001) under 1%–1.43 MPa–MB, 13.94 ± 2.37 a.u. (p < 0.0001) under 20%–0.07 MPa–MB, 34.38 ± 0.84 a.u. (p < 0.0001) under 20%–1.43 MPa–NoMB, and 48.79 ± 9.48 a.u. under 20%–1.43 MPa–MB (Fig. 3e). Similarly, maximum effective penetration depth increased from 55.0 ± 10.80 μm in controls to 140.0 ± 15.00 μm under 1%–1.43 MPa–MB, 140.0 ± 21.02 μm under 20%–0.07 MPa–MB, 190.0 ± 14.72 μm under 20%–1.43 MPa–NoMB, and 225.0 ± 23.63 μm under 20%–1.43 MPa–MB (Fig. 3f). Thus, although pLIFU alone and suboptimal microbubble conditions improved riboflavin penetration to varying degrees, the selected pLIFU–microbubble condition produced the greatest stromal riboflavin accumulation and depth extension.

Fig. 3.

Microbubble-assisted pLIFU markedly increases stromal riboflavin accumulation and penetration depth. a–c, Representative depth-resolved confocal fluorescence images of riboflavin distribution in control, US and USMB groups. Scale bars, 400 μm. d, Depth-dependent fluorescence intensity profiles showing stromal riboflavin distribution under different delivery conditions. e, f, Quantification of peak fluorescence intensity and maximum effective penetration depth. Microbubble-assisted pLIFU produced the strongest stromal fluorescence signal and the deepest riboflavin penetration under temperature-rise-limited conditions. Data are presented as mean ± s.d.; statistical significance is indicated in the panels. **p < 0.01, ****p < 0.0001.

Hydrophone-constrained modeling indicated that the large rabbit–porcine thickness difference primarily redistributed the local focal field rather than substantially altering pressure magnitude. In the thickness-isolated comparison between a 0.290-mm keratoconus-like rabbit cornea and a 0.833-mm ex vivo porcine cornea, loaded-field maximum PNP changed by 2.34%, the axial maximum shifted by 0.35 mm, lateral −6-dB width changed by 3.72%, and posterior-corneal and 2-mm-posterior PNP changed by 1.48% and 1.57%, respectively (Supplementary Fig. S1 and Table S1). The species-informed analysis yielded comparable posterior-corneal and 2-mm-posterior decreases of 1.10% and 1.58%, respectively.

2.2. Safe and effective CXL treatment validation in a keratoconus-like model

We established a collagenase-induced ectatic cornea model to evaluate whether enhanced riboflavin delivery could translate into improved epithelium-on CXL outcomes (Fig. 4). After collagenase type II treatment, high-frequency B-mode ultrasound showed a visibly steeper anterior corneal surface with central protrusion (Fig. 4a), and central corneal curvature increased from approximately 0.17 mm⁻¹ to approximately 0.29 mm⁻¹ (Fig. 4d). Shear-wave elastography showed weaker and less continuous wave propagation after model induction (Fig. 4b), accompanied by a decrease in shear-wave speed ($c_s$) from 6.20 ± 0.23 to 4.58 ± 0.10 m/s (Fig. 4d). Corneal thickness also decreased after modeling (Fig. 4e). Masson trichrome staining revealed stromal loosening, enlarged interlamellar spaces, and disrupted collagen organization (Fig. 4c), which was supported by reduced collagen-occupied area fraction, increased void fraction, higher orientation entropy, and increased angular median absolute deviation (Fig. 4f to 4i). These multimodal findings confirmed the formation of a keratoconus-like phenotype with corneal steepening, thinning, reduced shear-wave speed, and stromal collagen disorganization.

Fig. 4.

Collagenase induction establishes a keratoconus-like corneal phenotype with curvature elevation, reduced shear-wave speed and stromal disorganization. a, Representative high-frequency B-mode ultrasound images before (I) and after collagenase induction (II), showing increased corneal steepening. b, Representative shear-wave displacement maps indicating reduced wave propagation after model induction. c, Representative Masson trichrome staining images showing stromal loosening and collagen disorganization. d–e, Quantification of central corneal curvature, shear-wave speed and corneal thickness. f–i, Histomorphometric analysis of collagen area fraction, void fraction, orientation entropy and angular median absolute deviation. These multimodal readouts confirm the formation of key keratoconus-like features before therapeutic evaluation. Data are presented as mean ± s.d.; statistical significance is indicated in the panels. *p < 0.05, **p < 0.01, ***p < 0.001.

We next assessed therapeutic responses after epithelium-on CXL (Fig. 5). Longitudinal B-mode ultrasound showed persistent corneal steepening in controls, whereas the US and USMB groups showed progressive recovery of the anterior corneal contour during follow-up (Fig. 5a to 5c). This recovery was most evident in the USMB group, in which the central cornea became flatter by 1 month after treatment (Fig. 5c). Quantitatively, central corneal curvature remained elevated at 0.24 ± 0.02 mm⁻¹ in controls, decreased to 0.17 ± 0.01 mm⁻¹ in the US group, and further decreased to 0.16 ± 0.02 mm⁻¹ in the USMB group at 1 month (Fig. 5g). Relative to the pretreatment state, curvature decreased by 45.85 ± 6.32% in the USMB group, significantly greater than the 17.35 ± 5.45% decrease in controls (p < 0.001) and showing an additional improvement trend compared with the 40.32 ± 4.84% decrease in the US group (Fig. 5h).

Fig. 5.

Microbubble-assisted pLIFU-enhanced CXL improves corneal shape recovery and stiffness restoration in keratoconus-like ectasia. a–c, Representative longitudinal B-mode ultrasound images from control, US and USMB groups before treatment and during follow-up after CXL. d–f, Representative shear-wave displacement maps at selected time points showing treatment-dependent changes in wave propagation. g, h, Central corneal curvature and relative curvature change over time. i, j, Shear-wave speed and relative speed change over time. Compared with control and pLIFU alone, microbubble-assisted pLIFU-enhanced CXL produced more pronounced curvature reduction and stiffness restoration in the keratoconus-like model. Data are presented as mean ± s.d.; statistical significance is indicated in the panels. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Shear-wave displacement maps showed treatment-dependent changes consistent with the curvature results (Fig. 5d to 5f). Compared with the weak propagation pattern in controls, the US group showed improved wave propagation, whereas the USMB group displayed stronger and more continuous displacement patterns, indicating greater stiffness restoration (Fig. 5d to 5f). At 1 month, $c_s$ was 4.99 ± 0.29, 6.60 ± 0.54, and 7.71 ± 0.65 m/s in the control, US, and USMB groups, respectively (Fig. 5i). Relative to pretreatment, $c_s$ changed by −1.42 ± 5.76%, 30.47 ± 10.58%, and 52.50 ± 12.89%, respectively (Fig. 5j). As a secondary textual estimate only, Young’s modulus calculated from the individual $c_s$ values using $E\cong 3\mu \cong 3\rho c_s^2$ was 79.49 ± 9.21, 139.68 ± 22.99, and 190.93 ± 32.22 kPa, respectively; Young’s modulus is not plotted. Thus, microbubble-assisted pLIFU converted enhanced riboflavin delivery into measurable corneal flattening and a larger protocol-matched shear-wave-speed response after epithelium-on CXL in the keratoconus-like model.

2.3. Short-term ocular tissue safety evaluation

Finally, we evaluated short-term ocular tissue response under the selected condition (Fig. 6). The maximum corneal temperature rise during treatment did not exceed 1.5 °C. Hydrophone measurement at a retinal-position surrogate showed a peak negative pressure of 0.066 ± 0.008 MPa after propagation through the intact eye. The corresponding retinal-position mechanical index was approximately 0.076, below the FDA output limit for ophthalmic diagnostic ultrasound and used here as a conservative reference for posterior-segment acoustic action [56].

Fig. 6.

Microbubble-assisted pLIFU-enhanced CXL maintains short-term ocular tissue integrity without sustained inflammatory activation. a–c, Representative corneal H&E staining images from control, US and USMB groups after treatment. d–f, Representative retinal H&E staining images showing preserved retinal layer organization under the optimized treatment condition. g, h, Quantitative analysis of corneal thickness and retinal layer thickness. i–l, Serum and aqueous humor levels of IL-1β and MIP-1 measured before treatment and during short-term follow-up. The measured endpoints support short-term tissue tolerance within the observation window but do not constitute a clinical ocular-surface safety assessment. Data are presented as mean ± s.d.; statistical significance is indicated in the panels. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

During routine unaided follow-up observation, no obvious corneal surface damage or clinically apparent corneal edema was noted. This was not a standardized slit-lamp or fluorescein assessment and therefore does not exclude subtle epithelial defects, haze, hyperemia, or anterior-segment inflammation.

Histology showed that the corneal architecture remained recognizable after pLIFU-microbubble-enhanced CXL, with a relatively compact stromal structure that was consistent with the B-mode ultrasound evidence of corneal contour improvement (Fig. 6a to 6c and 6g). Retinal H&E staining showed that the major retinal layers remained identifiable in all groups, without overt laminar disruption (Fig. 6d to 6f and 6h). Although selected retinal layer thicknesses were lower in the USMB group than in controls, no definitive structural disruption was observed within the short-term window. Serum and aqueous humor cytokine measurements showed no sustained elevation of IL-1β or MIP-1 at 7 days after treatment, and the USMB group did not show a more persistent inflammatory response than pLIFU alone (Fig. 6i to 6l). Together, these results support favorable short-term ocular tissue tolerance under the optimized pLIFU-microbubble condition.

3. Discussion

3.1. Parameter selection and delivery-efficiency comparison

This study links microbubble-assisted pLIFU, cavitation-enhanced riboflavin delivery, and epithelium-on CXL efficacy in a single disease-relevant experimental chain. The central advance is not simply that microbubbles increased riboflavin penetration. Rather, pLIFU-microbubble-enhanced delivery was accompanied by greater early corneal curvature reduction and a larger shear-wave-speed response in the keratoconus-like model. This delivery-to-response link is essential because improved riboflavin fluorescence alone does not demonstrate that an epithelium-on CXL procedure can modify the shape or mechanical response of an ectatic cornea.

Through internal parameter comparisons, this study identified the key parameters regulating pLIFU-microbubble-mediated corneal riboflavin delivery. PRF, duty cycle, focal-zone acoustic pressure, and microbubble dose all affected cavitation activity and temperature rise, but the selected condition did not simply correspond to the highest cavitation intensity. A PRF of 1 Hz produced a strong transient cavitation response, whereas increasing the PRF to 500 or 1000 Hz markedly reduced cavitation signals. Previous studies have shown that pulse parameters can substantially alter microbubble cavitation responses and temporal stability. Under short pulse intervals, microbubble destruction by preceding pulses, depletion of effective cavitation nuclei, and insufficient replenishment of fresh microbubbles may reduce cavitation signals induced by subsequent pulses [40], [47]. A 20% duty cycle and a 1.43 MPa focal-zone acoustic pressure maintained effective cavitation while keeping the maximum temperature rise within 1.5 °C [44], whereas increasing the duty cycle to 30% or acoustic pressure to 2.15 MPa exceeded the preset thermal-control range. Regarding microbubble dose, 200 μl SonoVue solution produced a strong transient cavitation response, whereas increasing the dose to 300 μl decreased the cavitation signal. Based on previous work, this phenomenon may be related to nonlinear prefocal acoustic attenuation, enhanced scattering, and focal energy shielding caused by high microbubble concentrations [57]. Therefore, the final condition of PRF 1 Hz, 20% duty cycle, 1.43 MPa, and 200 μl SonoVue solution represented a balance between effective cavitation and temperature-rise control rather than a maximal acoustic-input combination.

Compared with previous corneal or ocular ultrasound delivery studies, the selected parameters in the present study were tailored to corneal riboflavin delivery and epithelium-on CXL therapy. In terms of frequency, Lamy et al. used 880 kHz nonfocused ultrasound to promote riboflavin entry into the corneal stroma under intact epithelium [21], and Park et al. used 690 kHz focused ultrasound for blood-retinal barrier opening [34], whereas the present study used 756 kHz pLIFU, within the sub-megahertz range commonly used in corneal riboflavin delivery and posterior-segment barrier-opening studies. Compared with 20 kHz ocular permeation enhancement by Zderic et al. [31] and 40 kHz low-frequency ultrasound by Sun et al. [22], the higher frequency used here is more readily focused, facilitating concentration of acoustic action in the central cornea. Regarding PRF, Park et al. and Bourdin et al. both used a PRF of 1 Hz in posterior-segment barrier-opening studies [33], [34], and the internal comparison in this study also showed that 1 Hz produced stronger transient cavitation than 500 and 1000 Hz, suggesting that a lower PRF facilitates microbubble replenishment and oscillation recovery between pulses [40], [47]. Regarding duty cycle, continuous or prolonged low-frequency ultrasound can enhance riboflavin delivery but requires additional cooling to control thermal effects [22]. The present study used a 20% duty cycle, maintaining effective transient cavitation while limiting the maximum temperature rise to within 1.5 °C. Regarding acoustic pressure, Bourdin et al. used 0.15 MPa as the optimized pressure for posterior-segment barrier opening [33], Park et al. compared 0.81 to 1.10 MPa and observed higher pressure-related retinal damage [34], and another work on high-intensity ultrasound with microbubbles for corneal riboflavin delivery used approximately 3.50 MPa [30]. The present study selected 1.43 MPa at the focal zone and confirmed by hydrophone measurement that the pressure at the retinal position was markedly attenuated in the presence of the eye, balancing corneal focal delivery enhancement and posterior-segment acoustic-level control. Regarding microbubble dose, previous ocular ultrasound-microbubble studies vary widely in microbubble type, administration route, and dose [32], [33], [34], [35], [36], [37], and therefore cannot be directly applied. The present study compared 20 to 300 μl SonoVue solution and identified 200 μl as a suitable dose. Regarding treatment duration, Lamy et al. treated corneas for 6 min [21], Sun et al. used 15 min of low-frequency ultrasound [22], and Zderic et al. reported ocular ultrasound delivery for up to 60 min at 20 kHz [31]. The present study achieved marked stromal riboflavin delivery enhancement with 5 min of pLIFU, shorter than the low-frequency ultrasound treatment by Sun et al. and comparable to the treatment duration used by Lamy et al.

The manufacturer-specified 6-h post-reconstitution period is a handling limit rather than an individual-bubble lifetime [58]. During the 5-min pulsed exposure, the onset broadband PCD response was followed by an immediate reduction and a lower but detectable response during most of the remaining insonation period, indicating an evolving acoustically responsive bubble population rather than survival of each initial bubble for 300 s. ICM_PV was 6028.55 ± 1047.96 a.u. in USMB-Rib and 6363.98 ± 2137.38 a.u. in USMB-NoRib (mean ± population s.d.), a descriptive reduction of 5.27% with riboflavin. The near-neutral riboflavin/PBS formulation may have modestly altered phospholipid-shell resonance or damping through ionic/interfacial conditions or surrounding-fluid viscosity [[59], [60], [61], [62], [63]], but bubble-size distribution, viscosity, interfacial properties, sound speed, and attenuation of the final formulation were not measured. We therefore infer neither an individual-bubble lifetime nor a specific physicochemical mechanism.

Under the above temperature-rise-limited condition, this study obtained higher stromal riboflavin delivery efficiency. Lamy et al. reported that nonfocused ultrasound increased stromal riboflavin fluorescence by up to approximately 3.6-fold [21]. Sun et al. used 40 kHz ultrasound at MI = 0.8 for 15 min and achieved riboflavin delivery approaching the epithelium-off CXL protocol; based on figure-derived estimation, the maximal delivery enhancement over passive epithelium-on soaking was approximately 8.5-fold [22]. In the present study, the USMB group achieved a 9.27 ± 2.16-fold (p < 0.0001) increase in peak riboflavin fluorescence and an 4.09-fold (p < 0.001) increase in maximum effective penetration depth within a 5-min treatment. Compared with previous ultrasound-only permeation enhancement, the major improvement of this study was not only increased delivery efficiency but also the use of cavitation activity, temperature rise, and stromal riboflavin distribution as parameter-selection endpoints within a short treatment time, thereby providing a delivery condition that combines delivery enhancement and thermal constraint for subsequent in vivo CXL validation.

3.2. CXL therapeutic efficacy in the disease model

Collagenase exposure was confined to the 15-min model-induction step. The enzyme solution was then removed and the cornea was immediately rinsed three times with sterile 0.9% saline; no collagenase was administered thereafter. Microbubble-assisted ultrasound delivery and riboflavin/UVA CXL were performed approximately four weeks later, after the keratoconus-like phenotype had developed. Published rabbit ectasia models similarly terminate collagenase exposure by solution removal and saline rinsing, with thinning and steepening persisting for weeks; a recent collagenase-model study subsequently applied an epithelium-preserving riboflavin-delivery/CXL intervention after the ectatic phenotype had developed [[49], [50], [64]]. The finite exposure, triple rinse, and four-week interval make a biologically meaningful pool of freely active collagenase at CXL highly unlikely. Moreover, all groups underwent the same collagenase exposure, rinsing, interval, riboflavin exposure, and UVA irradiation, so a common residual-enzyme effect cannot explain the additional USMB response. Because residual enzyme activity was not assayed directly, complete absence cannot be proven analytically. We therefore interpret the findings as treatment-associated stabilization of the existing enzyme-injured matrix—consistent with CXL-induced intermolecular cross-linking and increased resistance to subsequent enzymatic digestion [65]—rather than de novo collagen synthesis or mere neutralization of an acute collagenase bath.

The ex vivo porcine experiments were used for parameter screening rather than as evidence of species-invariant corneal acoustic dosing. Literature-informed benchmark cases differ substantially in central thickness (0.290 mm for collagenase-induced rabbit ectasia versus 0.833 mm for fresh ex vivo porcine cornea) and also differ in curvature and reported material properties [[49], [50], [66], [67]]. Hydrophone-constrained simulations showed that increasing central thickness from 0.290 to 0.833 mm produced a modest axial redistribution of the focal field. Thus, 1.43 MPa is reported as the hydrophone-calibrated nominal free-field/focal-zone pressure, not an identical intrastromal dose across species. Direct in vivo corneal pressure mapping was not performed; the model therefore quantifies expected cross-species redistribution but not subject-specific intrastromal dosimetry.

Efficient riboflavin delivery into the corneal stroma alone is insufficient to demonstrate therapeutic efficacy of epithelium-on CXL. For keratoconus, the more important question is whether delivery enhancement can be converted into improved corneal morphology and stiffness in a pathologically ectatic cornea. Previous ultrasound-enhanced corneal delivery studies have mainly focused on ex vivo normal corneas or healthy animal corneas, with endpoints such as riboflavin fluorescence intensity, penetration depth, or short-term epithelial changes [16], [21], [22], and therefore do not establish therapeutic value in a keratoconus-like disease state. The present study used a collagenase-induced rabbit keratoconus-like model and confirmed central corneal curvature elevation, reduced shear-wave speed, and collagen disorganization, providing a disease-relevant experimental basis for evaluating post-CXL therapeutic benefit.

In this model, USMB-enhanced CXL reduced central corneal curvature by 45.85 ± 6.32% at one month after treatment, significantly greater than the reduction in the control group and with a further decreasing trend compared with the US group. The primary elastography outcome showed that $c_s$ reached 7.71 ± 0.65 m/s in USMB, compared with 6.60 ± 0.54 m/s in US and 4.99 ± 0.29 m/s in control; corresponding changes from pretreatment were 52.50 ± 12.89%, 30.47 ± 10.58%, and −1.42 ± 5.76%. Because corneal wave propagation is affected by loading direction, testing frequency, boundary conditions, thickness, anisotropy, prestress, and model assumptions, absolute material moduli should not be compared directly across methods [[52], [53], [54], [55], [73], [68], [69], [70], [71], [72]]. The present study therefore emphasizes protocol-matched longitudinal and between-group changes in $c_s$ as an in vivo treatment-response indicator. The central contribution is that microbubble-assisted pLIFU-enhanced stromal riboflavin delivery was accompanied by measurable early corneal-shape and shear-wave responses in a keratoconus-like disease model rather than merely increased ex vivo fluorescence.

The present design does not establish non-inferiority or equivalence to standard epithelium-off CXL because it lacked a contemporaneous epithelium-off arm, a prespecified non-inferiority margin, and confirmatory power. Its competitiveness is therefore considered at the preclinical proof-of-concept level. The procedure avoids deliberate epithelial removal and uses a 5-min delivery step; it increased peak stromal fluorescence by 9.27 ± 2.16-fold and penetration depth by 4.09-fold, followed by concordant early curvature and shear-wave-speed responses. At one month, $c_s$ increased by 52.50 ± 12.89% in USMB, compared with 30.47 ± 10.58% in US and −1.42 ± 5.76% in control. For magnitude calibration only, in vivo porcine epithelium-removal CXL increased shear-wave speed by 56 ± 15% [68]. The similar order of magnitude supports biological plausibility, not equivalence, because species, disease state, IOP, follow-up, and wave models differ. Clinical syntheses likewise indicate that transepithelial approaches may reduce epithelial-healing complications but have less consistent efficacy than epithelium-off CXL [[10], [74], [75]]. The findings therefore support an efficient epithelium-preserving preclinical strategy, while a powered head-to-head epithelium-off comparison remains necessary before any non-inferiority or replacement claim.

$c_s$ is the primary elastography result; Young’s modulus is retained only as a secondary textual estimate using the original relation $E\cong 3\mu \cong 3\rho c_s^2$ and is not plotted. This distinction is necessary because the cornea is a thin, curved, prestressed, anisotropic lamellar tissue, whereas that relation assumes a homogeneous, isotropic, nearly incompressible bulk medium. Our principal inference therefore concerns longitudinal and between-group changes in $c_s$ acquired with the same system, propagation orientation, analysis region, and processing pipeline. These protocol-matched changes provide a treatment-response index, not the cornea’s intrinsic direction-independent Young’s modulus [[52], [55], [73], [68], [69], [70], [71], [72]].

High-frequency acoustic-radiation-force excitation is not inherently unsuitable for dispersion analysis; the limitation is the acquisition used here. The sequence used one 600-cycle push, retained only 30 tracking frames at 10 kHz (approximately 3.0 ms), and estimated time of flight from two lateral locations separated by D ≈ 0.788 mm. It did not acquire controlled multi-frequency excitation, a dense frequency–wavenumber field, or a stable multi-distance phase-velocity curve. Contemporaneous IOP was also not recorded. A retrospective Lamb-wave or anisotropic dispersion inversion would therefore be underdetermined. Destructive tensile testing was not performed because euthanasia and strip excision were incompatible with the same-eye 1-, 2-, and 4-week design, and no separate terminal calibration cohort was available. Accordingly, $c_s$ is used as a protocol-specific longitudinal treatment-response metric. Concordant changes in $c_s$, curvature, and stromal organization support a treatment-associated mechanical response, but not an absolute tensile modulus or specimen-specific validation of Young’s modulus.

3.3. Short-term ocular tissue tolerance

When pLIFU with microbubbles is used for ocular therapy, safety evaluation cannot rely solely on delivery efficiency. Microbubble-enhanced cavitation can increase tissue permeability, but excessive transient cavitation may increase the risk of tissue damage [40], [41], [23], [24], [25], [26], [27]. In the study by Sun et al., low-frequency ultrasound delivery achieved riboflavin delivery and corneal stiffness enhancement approaching the epithelium-off protocol, but mild superficial epithelial exfoliation and ZO-1 tight-junction alteration were observed immediately after treatment and recovered within 3 to 4 days [22]. Therefore, for epithelium-on CXL, delivery efficiency and tissue safety must be evaluated together. In the present study, temperature monitoring and passive cavitation detection were incorporated from the parameter-selection stage, and corneal and retinal histology as well as serum and aqueous humor cytokine assays were performed after treatment for short-term safety evaluation. Under the final condition, the maximum corneal temperature rise did not exceed 1.5 °C; the mechanical index at the retinal position was approximately 0.076, below the FDA-recommended limit for ophthalmic ultrasound applications [56]; H&E staining showed no overt laminar disruption in the cornea or retina; and IL-1β and MIP-1 levels in serum and aqueous humor were not persistently elevated. Compared with studies that use fluorescence enhancement or penetration depth as the primary endpoint, this study jointly evaluated delivery efficiency, cavitation activity, temperature rise, posterior-segment acoustic pressure attenuation, and short-term tissue response, providing more complete evidence for safe and effective physical augmentation of epithelium-on CXL.

Topical corneal administration of SonoVue is off-label, and intravenous tolerability cannot be extrapolated to the ocular surface. The route was selected for anatomical and mechanistic reasons: the central cornea is avascular, whereas the intended target is the epithelial–stromal interface, so intravenous microbubbles cannot directly populate the superficial target compartment; the cornea is externally accessible, allowing a defined surface suspension to be insonified locally [[16], [76]]. Published evidence provides limited but relevant precedent rather than proof of safety: ultrasound–microbubble exposure has been used for gene transfer to rabbit corneal epithelium and keratocytes without apparent tissue damage [28], bubble-liposome ultrasound enhanced ocular-surface gene transfer without an apparent adverse tissue reaction [29], and microbubble-enhanced corneal riboflavin phonophoresis has been reported in an ex vivo conference study [30]. Two additional topical ultrasound-only corneal riboflavin studies support the feasibility and short-term tolerance of the surface insonation route, while not testing SonoVue itself [21], [22]. Because slit-lamp grading, fluorescein staining, tear-film breakup time, osmolarity, endothelial-cell assessment, and tight-junction staining were not performed, ocular-surface clinical safety is not claimed and must be established in a dedicated study.

3.4. Limitations and future work

This study has several limitations. First, ex vivo riboflavin delivery parameter selection and in vivo CXL evaluation under disease conditions used corneas from different species. Differences in thickness, curvature, and tissue structure may affect acoustic field distribution and delivery efficiency; the added sensitivity analysis quantifies expected redistribution but does not replace direct in vivo intrastromal dosimetry. Second, the small unequal groups, absence of an epithelium-off comparator, and lack of a formal a priori sample-size calculation restrict between-group findings to exploratory interpretation. Third, residual collagenase activity, contemporaneous IOP, final-formulation osmolality/viscosity/acoustic properties, and time-resolved microbubble size/concentration were not measured. Fourth, the bulk conversion from $c_s$ to Young’s modulus did not resolve anisotropy or dispersion, and no destructive tensile test was performed. Fifth, direct slit-lamp and fluorescein assessment, tear-film testing, endothelial-cell evaluation, and longer-term functional follow-up were not available. The collagenase-induced rabbit model captures several ectatic features relevant to keratoconus but does not reproduce the full chronic pathobiology of human disease. Future studies should incorporate real-time cavitation/temperature feedback, subject-specific acoustic modeling, a powered epithelium-off comparator, standardized anterior-segment examination, and longer follow-up.

The 1-, 2-, and 4-week observations extend beyond immediate ex vivo delivery studies [21] and the 3–4-day epithelial-recovery interval reported for low-frequency ultrasound-mediated corneal riboflavin delivery [22], providing a longer early longitudinal assessment of curvature and $c_s$ response. Nevertheless, one month does not predict long-term remodeling or durability. Standard rabbit CXL has been evaluated at 3 and 8 months [77], establishing a longer benchmark that cannot be replaced by the present dataset. We therefore restrict the conclusions to early response and short-term tissue tolerance and do not infer long-term clinical success. Overall, this study identified a microbubble-assisted pLIFU riboflavin delivery condition with a maximum temperature rise not exceeding 1.5 °C and showed greater early corneal flattening and a larger same-platform shear-wave-speed response after epithelium-on CXL in a rabbit keratoconus-like model. Together with posterior acoustic-pressure, histological, and cytokine measurements, the findings support early feasibility and short-term tissue tolerance for the assessed endpoints. They do not establish non-inferiority to epithelium-off CXL, an intrinsic tensile modulus, comprehensive ocular-surface safety, or long-term durability.

4. Materials and methods

4.1. Ex vivo experiments: integrated acoustic parameter selection and riboflavin delivery evaluation

The ex vivo experiments were designed to identify the treatment condition for microbubble-assisted pLIFU-enhanced riboflavin delivery and consisted of acoustic insonation, cavitation monitoring, and temperature monitoring modules (Fig. 1c). In the acoustic module, pulsed sinusoidal signals were generated by an arbitrary waveform generator (DG1062Z, RIGOL, Beijing, China), amplified by an RF power amplifier (AG1006, T&C Power Conversion, Inc.), passed through impedance matching, and delivered to a pLIFU transducer with a center frequency of 756 kHz, focal length of 8.5 cm, and diameter of 32 mm (Wuxi Seavo Electronic Technology Co., China). In the cavitation monitoring module, a passive cavitation detection (PCD) probe with a center frequency of 7.5 MHz and diameter of 5 mm (Wuxi Seavo Electronic Technology Co., China) detected acoustic emission signals from the focal region. Signals were acquired by a high-speed digitizer (CSE8389-2GS, Dynamic Signals, LLC) at a sampling rate of 25 MS/s and a sampling depth of 5 × 10^6 points, and were then processed offline for cavitation quantification. In the temperature monitoring module, a T-type thermocouple (YTCCFN, Sunlit, Shanghai, China) measured the temperature near the cornea in real time, and a digital thermometer (TCP-400X, Measurefine, Guangdong, China) recorded temperature changes at 1 Hz. The pLIFU transducer, PCD probe, and ex vivo eye holder were spatially confocal using a three-dimensional positioning system. This system enabled simultaneous acquisition of cavitation activity and temperature rise under different PRFs, duty cycles, focal-zone acoustic pressures, and microbubble doses, thereby identifying the treatment condition for subsequent in vivo CXL therapy.

Fig. 1.

Study design and multimodal evaluation of microbubble-assisted pLIFU–enhanced epithelium-on corneal cross-linking. a, Schematic illustration of microbubble-assisted pulsed low-intensity focused ultrasound (pLIFU)–mediated riboflavin delivery followed by UVA-induced corneal cross-linking (CXL). Microbubble activity under pLIFU insonation is designed to transiently enhance epithelial and stromal permeability through cavitation-associated mechanical effects, thereby increasing stromal riboflavin availability before photoactivation. b, Experimental timeline showing keratoconus-like model induction, pretreatment evaluation, pLIFU-assisted riboflavin delivery, CXL treatment and longitudinal follow-up. c, Passive cavitation detection and temperature-monitoring system used for acoustic parameter optimization. d, High-frequency ultrasound imaging and shear-wave elastography workflow for in vivo assessment of corneal morphology and stiffness-related changes.

To evaluate acoustic pressure propagation toward the posterior segment, a needle hydrophone (NH0200, Precision Acoustics Ltd.) was used in a degassed water tank to measure pressure approximately 2.0 cm behind the focal zone, a position approximating the retinal region when the eye was present. Measurements were repeated after placing an intact eye, and peak negative pressure changes were recorded to estimate attenuation of posteriorly propagating acoustic waves by ocular tissues.

To quantify porcine-to-rabbit acoustic translation, an axisymmetric k-Wave model was constrained by measured free-field hydrophone profiles of the 0.756-MHz transducer. The measured focal distance and axial/lateral −6-dB widths were 71.1, 39.45, and 4.22 mm, respectively; the numerical field was referenced to the 1.43-MPa nominal free-field focal-zone PNP used experimentally. Hydrophone–model normalized RMSE values were 0.0602 axially and 0.0346 laterally. A thickness-isolated analysis varied central thickness from 0.290 to 0.989 mm while holding curvature and longitudinal acoustic properties constant. A complementary species-informed analysis incorporated literature-based rabbit and porcine thickness, curvature, longitudinal sound speed, density, and attenuation. Low-frequency elastic or viscoelastic moduli were not converted to MHz longitudinal sound speed because they describe different material responses. Model inputs and outputs are reported in Supplementary Fig. S1 and Table S1.

Riboflavin sodium phosphate (Macklin, Shanghai, China; CAS No. 130-40-5) was dissolved in sterile 1× phosphate-buffered saline (PBS; Gibco BASIC PBS, Thermo Fisher Scientific; Cat. No. C10010500BT; Lot No. 8118303) to 3.47 mg/ml. The measured pH was 7.39 ± 0.04 (n = 5). The PBS vehicle had a manufacturer-specified osmolality of 280–315 mOsm/kg; final-solution osmolality was not measured. The PBS contained potassium phosphate monobasic, sodium chloride, and sodium phosphate dibasic in water, without calcium, magnesium, or phenol red. No dextran, hydroxypropyl methylcellulose, EDTA, benzalkonium chloride, viscosity enhancer, or preservative was added. The formulation was filter-sterilized through a 0.22-μm cellulose acetate membrane (25-mm diameter; Solarbio, Beijing, China; Cat. No. YA0671), protected from light, and stored at 4 °C until use.

Fresh ex vivo porcine corneas (N = 60) were used for pLIFU-microbubble parameter selection and riboflavin delivery evaluation. PRF was set to 1, 10, 100, 500, or 1000 Hz; duty cycle was set to 1%, 5%, 10%, 20%, or 30%; focal-zone peak negative pressure was set to 0.07, 0.38, 0.77, 1.43, or 2.15 MPa; and microbubble dose was controlled by adding different volumes of SonoVue suspension with a concentration of 2.0×10⁸ microbubbles/ml. The added microbubble solution volumes were 20, 50, 100, 200, and 300 μl. PCD signals and temperature changes were recorded synchronously in all experiments, and the maximum temperature rise was required not to exceed 1.5 °C as the thermal-control criterion [44]. Raw PCD radiofrequency signals were imported into MATLAB and classified using comb filters [40], [45], [46], [47]. Harmonic, subharmonic, and ultraharmonic frequencies were calculated from the pLIFU fundamental frequency, with the highest harmonic defined as the ninth harmonic. Fixed third-order bandpass filters were then constructed over the harmonic frequencies ± 200 kHz, subharmonic frequencies ± 100 kHz, and ultraharmonic frequencies ± 50 kHz. After bandpass filtering, the absolute values of signals in each band were summed to characterize stable cavitation. Similarly, fixed third-order bandstop filters were constructed over the same frequency bands to remove discrete spectral components and extract broadband noise associated with transient cavitation. The absolute values of the broadband residual signals were summed as transient cavitation intensity. This processing was repeated segment by segment over the continuous 5-min acquisition to obtain time courses of stable and transient cavitation. Peak value and area under the curve were extracted to represent maximum instantaneous cavitation intensity and cumulative cavitation effect, respectively.

Under the selected condition, stromal riboflavin delivery was evaluated. After treatment, ex vivo corneas were thoroughly rinsed with saline to remove surface residual riboflavin and then fixed with 4% paraformaldehyde. Confocal fluorescence microscopy (Leica TCS SP8 STED 3X) was performed with an excitation wavelength of 405 nm, a scanning depth of 250 μm, and a z-step of 5 μm. Three-dimensional fluorescence images were imported into MATLAB to calculate mean fluorescence intensity at each depth, yielding depth-resolved fluorescence profiles. Peak fluorescence intensity and maximum effective penetration depth were extracted. A depth was considered to have no effective riboflavin signal when the mean fluorescence intensity was below 3 a.u.

4.2. In vivo experiments: CXL treatment

The in vivo experiments followed the timeline shown in Fig. 1b and were designed to evaluate the therapeutic efficacy of epithelium-on CXL after pLIFU-microbubble-enhanced riboflavin delivery in a keratoconus-like disease state. Six consecutive observation time points were included: time point I, healthy baseline before model induction; time point II, post-collagenase model formation; time point III, pre-CXL baseline; time point IV, 1 week after CXL; time point V, 2 weeks after treatment; and time point VI, 1 month after treatment. Central corneal curvature, thickness, and shear-wave speed ($c_s$) were evaluated longitudinally. Young’s modulus was calculated only as a secondary textual estimate using Eq. (5). The one-month endpoint assessed early treatment response and short-term tissue tolerance, not long-term clinical durability.

New Zealand white rabbits were used to establish a keratoconus-like disease model. Healthy rabbits were anesthetized with isoflurane and immobilized. The corneal surface was treated with 75% ethanol for 20 s, followed by application of 5 mg/ml type II collagenase to the central cornea for 15 min to induce stromal collagen degradation [48], [49], [50]. After treatment, the enzyme solution was removed and the cornea was rinsed three times with sterile 0.9% normal saline. After model induction, high-frequency B-mode ultrasound and shear-wave elastography were used to evaluate changes in corneal curvature, thickness, and shear wave speed. The model was considered successfully established when corneal curvature increased, thickness and shear wave speed decreased [48], [49], [50]. All animal experiments were approved by the Animal Ethics Committee of the School of Life Science and Technology, Xi’an Jiaotong University (2020 No. 33), and complied with the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research.

At study initiation, New Zealand white rabbits (N = 15) entered model induction, with a planned allocation of control n = 3, US n = 6, and USMB n = 6. Because animal availability was limited, the principal intervention group (USMB) and its mechanistic comparator (US) were each assigned six animals, while the untreated reference group was assigned three. This allocation was prespecified before treatment and was not based on outcome data. One USMB-assigned rabbit died unexpectedly after model induction but before pLIFU or microbubble exposure; the cause was not definitively established. Therefore, a total of 14 rabbits with successfully established models proceeded to the treatment and analysis stages (control group, n = 3; US group, n = 6; USMB group, n = 5). The control group did not receive pLIFU or microbubbles; the US group received pLIFU-assisted riboflavin delivery; and the USMB group received SonoVue microbubbles under the same pLIFU condition. As shown in Fig. 1d, 10 ml of 3.47 mg/ml riboflavin solution was added into the coupling chamber, and 200 μl of SonoVue microbubble solution was further added in the USMB group. In vivo riboflavin delivery was performed using the selected pLIFU condition: 1 Hz pulse repetition frequency, 20% duty cycle, 1.43 MPa focal-zone peak negative pressure, and 5 min insonation. The chamber was removed after riboflavin delivery. CXL was performed with UVA irradiation at 360 nm and 3 mW/cm² using repeated cycles of 2 min light and 2 min dark intervals for a total treatment duration of 30 min. At the beginning of every dark interval, 1.0 ml riboflavin formulation was applied topically. All groups received the same replenishment volume and timing, topical handling, irradiation geometry, and UVA exposure.

Routine unaided observation before and after treatment revealed no obvious corneal surface damage or clinically apparent corneal edema. However, slit-lamp biomicroscopy and fluorescein grading were not performed; color fundus photography documented only gross posterior-segment and retinal vascular morphology.

4.3. In vivo experiments: CXL efficacy evaluation

In vivo CXL efficacy evaluation followed the high-frequency ultrasound imaging and shear-wave elastography workflow shown in Fig. 1d. High-frequency B-mode ultrasound was used to obtain the anterior corneal surface contour and thickness information, and central corneal curvature was calculated to reflect corneal morphological changes after keratoconus-like model induction and CXL treatment. The primary elastography endpoint was $c_s$. Young’s modulus was calculated only as a secondary model-dependent estimate using Eq. (5), reported in the text, and not plotted. Measurements were repeatedly acquired at the time points shown in Fig. 1b, enabling longitudinal self-comparison within the same animals from pre-modeling to post-modeling, pretreatment, and post-treatment follow-up stages.

B-mode ultrasound and shear-wave elastography were performed before and after treatment using a Verasonics Vantage 256 platform (Verasonics Inc., WA, USA) and a high-frequency linear array probe (MAT 3.0, Vermon, Tours, France). B-mode imaging parameters were as follows: frequency, 15.625 MHz; pulse length, 1 cycle; driving voltage, 20 V. Twenty B-mode frames were acquired to extract the anterior corneal surface contour and calculate central corneal curvature from curve fitting of the central region. For shear-wave elastography, acoustic radiation force excitation was delivered at 15.625 MHz with an excitation length of 600 cycles and a driving voltage of 30 V. Subsequently, 200 IQ frames were acquired at a PRF of 10 kHz to track shear-wave propagation. Offline analysis first used the two-dimensional Loupas autocorrelation algorithm to estimate axial particle velocity from complex IQ data [78]. The axial kernel length was 5 samples, the temporal ensemble length was 5 frames, sound speed was set to 1540 m/s, transmit frequency was 15.625 MHz, and temporal sampling interval was 1/10000 s. Axial particle velocity fields were smoothed using a 3 × 3 × 3 mean filter to reduce random noise in propagation-delay estimation. Directional filtering in the frequency domain was then used to separate unidirectional shear waves [79], and the rightward propagating component with higher signal-to-noise ratio and better propagation continuity was selected for shear-wave speed estimation.

The acquisition was designed for local time-of-flight shear-wave-speed estimation rather than frequency-dependent dispersion inversion: a single 600-cycle acoustic-radiation-force push was followed by 30 retained tracking frames at 10 kHz (approximately 3.0 ms). No controlled multi-frequency excitation or frequency-resolved phase-velocity curve was acquired.

Local shear-wave speed was estimated from the time delay between axial velocity curves at two neighboring lateral positions, a method based on time-of-arrival or cross-correlation that is widely used in ultrasound shear-wave elastography [[79], [80]]. For each target location in the corneal ROI, two lateral positions were selected on the left and right sides at a distance of $d=4$ pixels. The effective lateral distance between the two positions was $D$.

$$D=2d\mathrm{\Delta }x$$ (1)

where $\mathrm{\Delta }x$ is the lateral sampling interval. Based on system reconstruction grid calibration, $\mathrm{\Delta }x$ was approximately 0.09856 mm in this study, yielding $D\approx 0.78848$ mm. Axial velocity time curves $s_1\left(t\right)$ and $s_2\left(t\right)$ were extracted from the two lateral positions, and cubic interpolation was used to increase temporal sampling fivefold for subframe time-delay estimation. A Hann window was then applied to both curves to reduce truncation effects. The cross-correlation function was calculated as follows:

$$R_{12}\left(\tau \right)=\sum _t{s}_1\left(t\right)s_2\left(t+\tau \right)$$ (2)

The time delay corresponding to the maximum correlation was denoted ${\tau }_{\mathit{max}}$. Local shear-wave speed was calculated as follows:

$$c_s\left(z,x\right)=\frac{D}{\left|{\tau }_{\mathit{max}}\right|}$$ (3)

A directional filter was used to retain the dominant propagation direction and suppress opposing or reflected components. Mean $c_s$ was then calculated within the corneal region of interest.

For a secondary textual comparison only, shear modulus was estimated under the homogeneous, isotropic, nearly incompressible bulk-tissue approximation (55, 58–60):

$$\mu \cong \rho c_s^2$$ (4)

where $\rho$ is corneal tissue density and $c_s$ is shear-wave speed. Tissue density was approximated as 1062 kg/m3. Young’s modulus was then calculated as follows:

$$E\cong 3\mu \cong 3\rho c_s^2$$ (5)

4.4. Histology and inflammatory cytokine assays

At predefined time points after treatment, animals were euthanized and ocular tissues were collected. Corneal and retinal tissues were fixed, dehydrated, paraffin-embedded, sectioned, and subjected to H&E staining to assess corneal epithelium, stromal structure, and major retinal layers. Masson trichrome staining was used to evaluate corneal stromal collagen structure. Image analysis was performed at the same magnification and under consistent ROI conditions. Color segmentation and morphological processing were used to extract blue-stained collagen regions and white void regions, and collagen-occupied area fraction, void fraction, collagen orientation entropy, and angular median absolute deviation were calculated to evaluate collagen density, stromal looseness, and fiber organization [[81], [82]].

To assess short-term inflammatory response, serum and aqueous humor samples were collected before treatment and at one and seven days after treatment. Approximately 100 μl of aqueous humor and 2.0 ml of peripheral blood were collected each time. Blood samples were centrifuged to obtain serum. IL-1β and MIP-1 levels in serum and aqueous humor were measured using commercial rabbit ELISA kits (BYabscience, BY-ER772059 and 772712), with four replicate wells for each sample. Final concentrations were calculated using a four-parameter logistic standard curve.

4.5. Statistical analysis

Each final group contained at least three independent animals, permitting estimation of within-group variance and leaving 11 residual degrees of freedom for the three-group model (N − k = 14 − 3). This establishes mathematical estimability of the prespecified three-group comparison but does not establish adequate confirmatory power or a universal minimum sample-size threshold. No formal a priori sample-size calculation was performed. An approximate fixed-effects three-group omnibus-F sensitivity analysis used N = 14, α = 0.05, and 80% target power. With df₁ = 2 and df₂ = 11, Fcrit = 3.982 and λ = 12.816, yielding Cohen’s f = √(λ/N) = 0.957 (approximately 0.96). The study was therefore sensitive mainly to very large omnibus effects; this calculation did not model repeated-measures covariance, group-by-time interaction, or pairwise-comparison power. All quantitative data are presented as mean ± s.d. Two-group comparisons were performed using two-sided Student’s t tests. Comparisons among three or more groups were performed using one-way ANOVA with Tukey post hoc correction. Longitudinal data involving both group and time factors were analyzed using two-way repeated-measures ANOVA. p < 0.05 was considered statistically significant. *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001. Unless otherwise specified, the animal/eye was used as the independent statistical unit. All post-processing was performed in MATLAB R2020a (MathWorks Inc., Natick, MA, USA) on a personal computer (Lenovo, China; Intel Core i5-9500 CPU at 3.0 GHz; 32 GB RAM).

Data availability

All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data are available from the corresponding author upon reasonable request.

CRediT authorship contribution statement

Qiao Wang: Writing – review & editing, Writing – original draft, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Shifang Guo: Writing – review & editing, Supervision. Youjia Sun: Investigation. Yaojie Wu: Investigation. Ziyi Lin: Methodology, Investigation, Data curation. Tongyu Wang: Investigation. Xipeng Chen: Investigation. Qiang Su: Supervision. Yupin Zheng: Supervision. Yue Wu: Supervision. Diya Wang: Writing – review & editing, Supervision, Funding acquisition, Conceptualization. Mingxi Wan: Supervision.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article:

Supplementary Data 1

Supplementary Fig. S1 and Table S1 report the hydrophone-constrained source, thickness-isolated, and species-informed cases, along with pressure metrics and the explicit limitation that these are model-based sensitivity outputs rather than direct in vivo pressure measurements.