Dual-threshold adaptive irradiation strategy for vascular-targeted photodynamic therapy based on hyperspectral-photoacoustic dual-modality imaging

Abstract

Conventional vascular-targeted photodynamic therapy (V-PDT) typically relies on empirically chosen fixed irradiance and exposure times, without accounting for substantial inter-subject variability in photosensitizer kinetics and oxygen supply. As a result, treatment often proceeds under severe hypoxia, where additional light delivery mainly contributes to inefficient overdosing rather than productive photochemistry. To address this limitation, we established a hyperspectral-photoacoustic dual-modality imaging platform. Hyperspectral imaging simultaneously acquires diffuse reflectance and fluorescence spectra, enabling real-time quantitative monitoring of in vivo photosensitizer photobleaching dynamics and blood oxygen saturation, while photoacoustic microscopy provides micrometer-scale visualization of vascular architecture and V-PDT-induced damage. Building on this capability, we propose a physiology-guided, dual-threshold adaptive light-dosing strategy for individualized light delivery. Instead of adhering to a single preset protocol, the system uses both a tissue oxygenation threshold and an in vivo photosensitizer bleaching plateau as biologically meaningful criteria for real-time light modulation and treatment termination. In preclinical studies, this closed-loop, imaging-guided control scheme improved the balance between V-PDT efficacy and treatment time and provided a technical foundation for future patient-specific dose optimization in V-PDT.

1. Introduction

Photodynamic therapy (PDT) is a minimally invasive treatment that combines a photosensitizer, specific-wavelength light, and molecular oxygen to produce cytotoxic reactive oxygen species (ROS) for destroying target tissues [1–5]. A specialized modality, vascular-targeted photodynamic therapy (V-PDT), uses photosensitizers that preferentially localize in the vasculature and, upon activation, induce vascular injury and occlusion [6–8]. By selectively damaging the abnormal vasculature and thereby reducing blood supply to the diseased tissue, V-PDT has demonstrated substantial clinical promise, particularly for benign superficial vascular malformations/proliferative disorders such as port-wine stains (PWS), and has also been explored in certain solid tumors [9–11].

Importantly, because therapeutic outcome in V-PDT is tightly coupled to vascular physiology—particularly to blood oxygen availability—there is increasing interest in optimizing PDT dosimetry to maximize treatment efficacy while maintaining safety [12, 13]. In fact, determining an optimal light dose for each patient remains a significant challenge, as current methods struggle to account for dynamic tissue factors (oxygenation, photosensitizer levels) that critically influence V-PDT outcomes [14–16]. This has motivated intensive research into personalized, dosimetry-guided illumination strategies for V-PDT.

Conventional V-PDT protocols often use fixed drug and light doses without adapting to individual patient physiology [17,18]. Such one-size-fits-all dosing lacks physiological feedback, leading to several well-recognized challenges. First, high-irradiance or total light doses can rapidly consume available oxygen, driving the treatment zone into hypoxia and drastically impairing ROS generation [19–21]. Once oxygen is depleted, continued illumination provides diminishing therapeutic benefit and effectively wastes both the delivered light and the remaining photosensitizer capacity. Another critical issue is substantial interpatient variability in photosensitizer uptake, tissue optical properties, and baseline oxygenation. These factors can cause substantial discrepancies in the actual V-PDT dose delivered to tissues, even under identical nominal treatment settings [22,23]. As a result, standard dosing can lead to unpredictable outcomes: some patients are over-treated, risking unnecessary normal tissue damage, while others are under-treated, potentially resulting in incomplete ablation of proliferative/abnormal vessels or tumor recurrence. The lack of real-time adjustment means current regimens cannot accommodate dynamic changes such as photobleaching-induced sensitizer loss or blood flow reductions during therapy. In short, traditional V-PDT dosing does not balance photochemical demand with physiological supply, leading to oxygen depletion, inefficient sensitizer usage, and inconsistent patient responses.

Researchers have pursued several strategies to address these limitations. One approach is to enhance oxygen availability during V-PDT so that ROS production can continue efficiently [24]. For example, various oxygen carriers (e.g., hemoglobin-based nanocarriers) and in situ oxygen-releasing compounds have been explored to counteract treatment-induced hypoxia [25–27]. While these methods can temporarily raise local oxygen levels, they have practical limitations such as rapid oxygen release kinetics, limited tissue penetration, and lack of sustained delivery throughout the treatment. An alternate strategy is to modulate the light delivery itself to reduce oxygen consumption rates [28]. Reducing the fluence rate or using fractional (intermittent) illumination provides time for oxygen diffusion and resupply, preventing severe hypoxia. For instance, de Haas et al. reported that a preset fractionated illumination scheme with a dark interval significantly improved clinical response in ALA-PDT for superficial basal cell carcinoma, highlighting the therapeutic value of allowing physiological recovery during PDT [29]. Nevertheless, the optimal fractionation parameters (e.g., number of fractions and dark-interval duration) can vary with local tissue physiology and are often chosen empirically, which may limit generalizability across subjects.

Previous studies have demonstrated that low-irradiance or metronomic light delivery protocols can preserve tissue oxygenation and, consequently, enhance singlet oxygen generation and therapeutic efficacy [30]. However, these conservative dosing strategies significantly prolong the illumination time, imposing logistical burdens on patients and clinicians. More fundamentally, both the oxygen-supply augmentations and fixed light-modulation schedules represent open-loop interventions, as they are applied uniformly rather than adaptively tuned to real-time patient feedback. Consequently, these strategies lack the flexibility to respond to moment-to-moment changes in tissue conditions, and thus cannot reliably ensure an optimal balance between treatment efficiency and safety.

Recognizing the need for real-time dosimetry, researchers have developed methods to monitor key V-PDT parameters during treatment [31,32]. On the photosensitizer side, fluorescence and absorption-based techniques enable in situ quantification of photosensitizer concentration and its photobleaching within tissue [33]. Photobleaching (the light-induced depletion of photosensitizer) is often used as a dosimetric indicator, under the assumption that the ROS dose is proportional to the amount of sensitizer consumed. Several clinical studies have leveraged real-time photosensitizer fluorescence monitoring to guide therapy [34,35]. However, such univariate guidance schemes fail to account for oxygen dynamics. A sufficient fluorescence decrease does not necessarily indicate adequate ROS production if oxygen was exhausted in the process, nor does it prevent overtreatment in well-oxygenated but low-sensitizer conditions.

Complementary to photosensitizer monitoring, significant effort has also been directed at measuring tissue oxygenation during V-PDT [36,37]. Techniques such as phosphorescence quenching, diffuse reflectance oximetry, and photoacoustic imaging have been employed to capture real-time oxygen levels in the target tissue [38–41]. For example, Neuschmelting et al. used handheld multispectral optoacoustic tomography (MSOT) to spectrally unmix hemoglobin and longitudinally quantify oxygenation changes associated with V-PDT response in vivo [42]. Notably, oxygenation-only monitoring cannot determine whether continued irradiation remains photochemically productive once intravascular photosensitizer availability becomes limiting or bleaching kinetics approach a plateau. These studies have provided valuable insights into V-PDT-induced hypoxia and oxygen kinetics. However, most oxygen-monitoring studies to date have been observational: they track the decline in oxygenation to understand V-PDT physiology, but rarely feed this information back into controlling the light dose. In practice, clinicians still often rely on pre-set light protocols without real-time adjustment, even if an oxygen sensor indicates depletion. Thus, current “oxygen-informed” approaches generally stop short of closed-loop control.

We developed a dual-modality imaging system that enables continuous, quantitative mapping of blood oxygen saturation (sO₂) and intravascular photosensitizer (PS) concentration using hyperspectral imaging (HSI), while photoacoustic microscopy (PAM) provides high-resolution visualization of vascular architecture and treatment-induced damage for guidance and outcome assessment. Based on these inputs, we propose a dual-threshold adaptive irradiation strategy that enables closed-loop regulation of the total light dose delivered during V-PDT. In essence, the system modulates irradiance in real time according to the tissue oxygenation state. When sO₂ remains above the predefined threshold, a higher irradiance is applied to maintain effective photochemistry; once sO₂ approaches or falls below the critical hypoxia threshold, the irradiance is reduced to alleviate oxygen depletion and facilitate reoxygenation. Treatment is terminated when sO₂ stays at the hypoxic threshold and the photosensitizer photobleaching rate remains persistently low, indicating diminishing photochemical yield and avoiding unnecessary or low-efficiency irradiation. By maintaining blood oxygen above a defined safety threshold and ensuring sufficient photosensitizer utilization, this strategy aims to dynamically balance photochemical yield with physiological constraints in real time. To align with the vascular-targeted injury mechanism of V-PDT, we validated this adaptive strategy in a benign superficial vascular model, demonstrating that, compared with conventional single-factor or fixed-schedule approaches, dual-parameter personalized feedback control achieves improved therapeutic outcomes with higher efficiency.

2. Materials and methods

2.1. Optical-resolution photoacoustic microscopy

Figure 1(a) illustrates an optical-resolution photoacoustic microscopy (OR-PAM) system consisting of a pulsed laser source, a photoacoustic probe, a motorized translation stage, a data acquisition unit, and a control module. A 532 nm pulsed laser (VPFL-G-20, Spectra-Physics) was used to excite the photoacoustic signal. The laser beam was first expanded by a lens group, then focused by an objective lens and redirected by a right-angle prism into a custom-built high-sensitivity photoacoustic combiner. This combiner comprises an aluminum-coated prism and an uncoated prism, enabling coaxial alignment of optical and acoustic paths. The thin aluminum coating reflects the laser light while allowing acoustic signals to pass through. An optical correction lens mounted on the side of the combiner is used to correct optical aberrations introduced by the prism assembly.

Fig. 1.

Overview of the hyperspectral-photoacoustic dual-modality imaging system and quantitative reconstruction framework. (a) OR-PAM system schematic, illustrating the pulsed laser excitation, acoustic detection, and scanning configuration used for high-resolution vascular imaging. (b) HSI system schematic, showing the illumination, spectral selection, and detection modules for simultaneous diffuse reflectance and fluorescence acquisition. (c) Quantification of vascular damage based on OR-PAM imaging: (i) Representative raw OR-PAM image of the dorsal skinfold vasculature; (ii) Corresponding binary vascular map used for vessel segmentation and vascular shrinkage ratio (VSR) calculation. (d) Optical properties of the photosensitizer HMME: (i) Absorption spectrum; (ii) Fluorescence emission spectrum. (e) Schematic overview of the quantitative reconstruction workflow for photosensitizer concentration and oxygen saturation, illustrating the required spectral image sets and the resulting parametric maps: (i) photosensitizer concentration quantification based on dual-band normalized fluorescence imaging; (ii) sO₂ estimation based on a six-wavelength spectral second-derivative algorithm. Abbreviations: RL, relay lens; LCTF, liquid-crystal tunable filter; LG, light guide; OL, objective lens; HL, halogen lamp; CW, continuous wave; DAQ, data acquisition card; AMP, amplifier; FPGA, field-programmable gate array; UT, ultrasonic transducer; OUC, photoacoustic combiner; WT, water tank; RP, reflecting prism.

The generated photoacoustic signals are acoustically focused by an acoustic lens. By adjusting the laser beam path and the position of the objective lens, optical and acoustic foci are precisely aligned to achieve high-sensitivity photoacoustic detection, which is then captured by an ultrasonic transducer (V214-BC-RM, Olympus). The right-angle reflecting prism, right-angle prism, correction lens, and acoustic lens are bonded together using ultraviolet-curable adhesive (NOA61 glue, Norland Products Inc.). The detected photoacoustic signals are amplified by two cascaded low-noise amplifiers (48 dB total gain, ZFL-500LN+, Mini-Circuits), and sampled at 250 MHz using a high-speed digitizer (ATS9350, AlazarTech). Raster scanning of the imaging target is performed by two orthogonal linear stages (PLS-85, Physik Instrument). The laser firing, stage movement, and data acquisition are all synchronized through an FPGA-based control system (myRIO, National Instruments). The axial resolution is determined by a combination of laser pulse width, acoustic attenuation in tissue, and the frequency response of the ultrasonic transducer. Based on the acoustic bandwidth of the high-frequency ultrasound transducer used in our OR-PAM system, the axial resolution is approximately 20–30 µm. The lateral resolution is primarily determined by the optical diffraction-limited focus of the 532-nm excitation beam and the numerical aperture of the objective lens, yielding an effective lateral resolution of ∼5 µm.

2.2. Hyperspectral imaging system

Figure 1(b) shows the configuration of the HSI system, which comprises an excitation laser (MGL-N-532, Changchun New Industries), a broadband halogen light source (GCI-060101, Daheng Optics), a liquid-crystal tunable filter (VariSpec-VIS, PerkinElmer), and a scientific CMOS camera (Zyla 4.2, Andor). The LCTF sequentially tunes wavelengths over a spectral range of 420–720 nm with a tuning step of 1 nm. The sCMOS camera enables high-SNR image acquisition at short exposure times, thereby reducing the overall hyperspectral imaging duration. Although the hardware supports dense hyperspectral scanning, the closed-loop guidance in this study operates using a spectrally sparse (multispectral) subset tailored to the two quantitative endpoints (sO₂ and photosensitizer concentration), which substantially reduces data volume and update latency. To facilitate system control, image acquisition, data storage, and processing, we developed custom software using LabVIEW (National Instruments, Texas).

2.3. Photosensitizer

The photosensitizer used in this study was hematoporphyrin monomethyl ether (HMME), a chemically stable compound known for its high singlet oxygen yield [43]. HMME has been widely employed in the clinical treatment of benign vascular malformations and related proliferative disorders [44]. Figure 1(d) shows its absorption and fluorescence emission spectra.

2.4. Animal model

Female BALB/c mice (25–30 g, 6–10 weeks old) were used in this study, obtained from the Laboratory Animal Center of Xi'an Jiaotong University. All procedures involving animals, including housing, experimental operations, and euthanasia, were conducted in accordance with ethical guidelines approved by the Experimental Animal Ethics Committee of Xi'an Jiaotong University.

To enable real-time monitoring of V-PDT dose-related parameters within the target vasculature, a chronic dorsal skinfold window chamber (DSWC) model was established in rodents using a titanium alloy skinfold frame (SM100, APJ Trading Co., Ventura, USA), following established methods [45–47]. The DSWC mouse model is a well-established platform for investigating vascular responses during V-PDT, providing a stable in vivo preparation that allows continuous, high-resolution visualization of microvascular structure and dynamics during and after treatment.

2.5. V-PDT protocol

In the established DSWC mouse model, HMME was administered via tail vein injection at a dose of 10 mg/kg. V-PDT was initiated 1 min after injection. To investigate how irradiance, illumination timing, and physiology-guided feedback affect V-PDT outcome, six different V-PDT protocols were implemented. Sixty animals were randomly assigned to six groups (n = 10 per group), each corresponding to one illumination regimen. The detailed irradiance schedules, feedback variables, switching rules, and termination criteria are summarized in Table 1.

• (1)High-irradiance, short-duration group. Continuous illumination at 100 mW/cm² for 10 min.
• (2)Low-irradiance, long-duration group. Continuous illumination at 25 mW/cm² for 40 min.
• (3)50% photosensitizer-threshold group (50%-PS group). Continuous illumination at 100 mW/cm² was delivered until real-time hyperspectral imaging indicated that the intravascular HMME concentration within the target vessel had decreased to 50% of its baseline value. Irradiation was then terminated immediately.
• (4)30% photosensitizer-threshold group (30%-PS group). The same high-irradiance setting (100 mW/cm²) was used, but illumination was continued until the HMME concentration fell to 30% of its initial value. This threshold was selected to approximate the typical residual PS fraction observed at the end of conventional 100 mW/cm², 10-min V-PDT in preliminary experiments (approximately 15–30% remaining), while truncating the later, potentially redundant portion of exposure.
• (5)Dual-threshold group. Illumination was delivered continuously at 100 mW/cm² and terminated when a dual set of biologically motivated criteria was met: (i) the photosensitizer bleaching rate approached zero, indicating that the normalized HMME curve had entered a plateau phase with only minimal further decrease over time, and (ii) the local oxygen saturation remained persistently below 40% without further recovery, indicating sustained hypoxia. This stopping rule ensured that high-irradiance was stopped once additional light no longer produced effective PS bleaching and the reaction had effectively reached a photochemical plateau under oxygen-limited conditions.
• (6)Adaptive feedback group (Adaptive feedback). In this closed-loop protocol, the irradiance was dynamically modulated between 100 and 25 mW/cm² based on real-time vascular oxygenation and intravascular PS dynamics. Illumination was initiated at 100 mW/cm². Throughout treatment, sO₂ was continuously monitored and compared to a predefined safety threshold of 40%. When the local sO₂ approached or fell below this threshold, the irradiance was immediately reduced to 25 mW/cm², once sO₂ recovered above the threshold, the irradiance was switched back to 100 mW/cm². Illumination in the adaptive group was finally terminated when the same PS-oxygen dual-threshold condition as in the dual-threshold group was satisfied.

Table 1. Summary of V-PDT illumination protocols and control rules.

V-PDT Protocol	Irradiance	Feedback variable	Switching rule	Termination criterion	Total delivered fluence
High-irradiance short-duration	100 mW/cm2 continuous	None (open-loop)	N/A	Fixed time: 10 min	60 J/cm2 (fixed)
Low-irradiance long-duration	25 mW/cm2 continuous	None (open-loop)	N/A	Fixed time: 40 min	60 J/cm2 (fixed)
50%-PS group	100 mW/cm2 continuous	PS concentration	N/A	Stop at PS = 50% of baseline	Variable
30%-PS group	100 mW/cm2 continuous	PS concentration	N/A	Stop at PS = 30% of baseline	Variable
Dual-threshold	100 mW/cm2 continuous	PS kinetics (plateau) + sO2 kinetics (plateau)	N/A	Stop when PS reaches plateau and sO2 remains < 40% (sustained hypoxia)	Variable
Adaptive feedback	100 ↔ 25 mW/cm2 switching	sO2 (for modulation) + PS kinetics (for endpoint)	If sO2 ≤ 40%, switch 100→25; once sO2 recovers>40%, switch 25→100	Same dual-threshold endpoint as above (PS plateau + sustained sO2 < 40%)	Variable

The sO₂ threshold of 40% was selected based on consistent observations in our DSWC experiments that once vascular sO₂ dropped below ∼40%, the target region rapidly entered a persistent hypoxic plateau with slow or incomplete recovery; keeping sO₂ above this level better balanced oxygen supply and photochemical consumption. This choice is also supported by prior microvascular physiology and PDT studies suggesting that productive photochemistry becomes substantially impaired as oxygenation falls into the 30–40% range [48,49]. The high-irradiance level (100 mW/cm²) was chosen to match commonly used clinical fluence-rate settings for HMME-mediated PDT of port-wine stains (typically ∼80–100 mW/cm²), serving as a clinically relevant “high-irradiance” reference. The low-irradiance level (25 mW/cm²) was selected as an oxygen-preserving setting that markedly reduces oxygen consumption while maintaining non-zero photochemistry, and it also matches the fixed low-irradiance in our study, enabling a transparent comparison between fixed low-irradiance dosing and feedback switching.

2.6. Photosensitizer concentration reconstruction

To obtain accurate fluorescence intensity data of the photosensitizer, a compensation algorithm was employed to correct for fluorescence attenuation caused by the optical properties of biological tissues. Figure 1(e)(i) illustrates the dual-band normalization algorithm used to generate wide-field quantitative fluorescence maps. This method does not require explicit light propagation modeling across the entire field of view. Instead, it derives a correction factor by integrating reflectance spectra from two distinct wavelength ranges, thereby simplifying the overall data processing workflow [50].

Importantly, the PS plateau criterion in this study is determined from the compensated (reflectance-corrected) PS concentration rather than the raw fluorescence intensity. The reflectance images required for the dual-band normalization are acquired concurrently with fluorescence at each monitoring time point, so the compensation is updated throughout V-PDT to account for therapy-induced optical-property changes (absorption/scattering). The quantitative robustness of this dual-band normalization method against variations in tissue optical properties has been systematically validated in our prior work [19].

In our implementation, each PS quantification update is performed using concurrent reflectance sampling in the two correction bands together with fluorescence sampling, resulting in an acquisition time of ∼2.6–3.0 s per update (depending on exposure settings). This temporal resolution is sufficient because PS kinetics is used as a treatment-termination cue (photobleaching plateau) rather than a high-frequency control input, while the faster sO₂ updates drive irradiance modulation.

2.7. Oxygen saturation reconstruction

Figure 1(e)(ii) presents the spectral second-derivative algorithm used in this study to reconstruct sO₂ from hyperspectral measurements [51]. The quantitative accuracy and robustness of this six-wavelength second-derivative oximetry method have been systematically validated in our prior work, with detailed phantom and in vivo verification reported previously [19]. Reflectance spectral images were acquired at six specific wavelengths: 544 nm, 552 nm, 568 nm, 576 nm, 592 nm, and 600 nm. With an LCTF switching time of ∼50 ms and a short exposure time (typically 80–100 ms per wavelength), one sO₂ update can be obtained in ∼0.8–0.9 s, enabling near real-time feedback for irradiance modulation during V-PDT. The primary advantage of this algorithm lies in its ability to suppress interference from endogenous chromophores such as melanin in the skin, which could otherwise affect measurement accuracy. This is achieved by processing the raw tissue spectra into second-order derivatives, thereby enhancing the specificity of the oxygen saturation estimation.

2.8. Quantification of vascular damage

Figure 1(c) displays the workflow by which pre- and post-treatment PAM images of the dorsal skinfold window were converted into binary masks for quantitative vascular damage analysis. The vascular shrinkage ratio (VSR, %) was then calculated based on the change in the number of white pixels between the two binary images. The VSR is defined as:

$$\text{VSR}=\frac{N_{\text{before}}-N_{\text{after}}}{N_{\text{before}}}$$ (1)

where N_before and N_after represent the number of foreground pixels (i.e., vessel area) in the binary PAM images before and after treatment, respectively.

2.9. Quantitative analysis of photobleaching and oxygen dynamics

The oxygen depletion rate (K_sO2, %/min) was defined as the absolute average slope of the sO₂ time course from the beginning of irradiation to the moment when sO₂ first dropped to 30%:

$$K_{\mathrm{s}{\mathrm{O}}_2}=\left|\frac{\mathrm{s}{\mathrm{O}}_2(t_{start})-\mathrm{s}{\mathrm{O}}_2(t_{30\mathrm{\%}})}{t_{30\mathrm{\%}}-t_{start}}\right|$$ (2)

where t_start denotes the onset of PDT illumination, and t_30% is the time when sO₂ first dropped to 30%. This metric reflects the rate at which oxygen is consumed during V-PDT.

The time-to-hypoxia (T_hypoxia, s) was defined as the elapsed time from irradiation onset to the moment when sO₂ first dropped below 30%:

$$T_{hypoxia}=t(\mathrm{s}{\mathrm{O}}_2<30\mathrm{\%})$$ (3)

where t(sO₂ < 30%) denotes the time point when sO₂ first drops below 30%. A longer T_hypoxia indicates slower oxygen depletion and improved oxygen availability during V-PDT.

The PS plateau was defined using a relative-slope criterion. PS concentration was first normalized as C_norm(t) = C(t)/C(0), and an initial bleaching slope S_init was obtained by linear fitting over the first 60–120 s after light onset. During irradiation, a sliding time window was used to estimate the instantaneous bleaching slope S(t). When the relative slope |S(t)|/|S_init| fell below 0.1 and remained below this threshold for a predefined duration (60 s), the PS dynamics were considered to have reached the plateau, and the corresponding time was recorded as T_plateau. In the dual-threshold PDT protocol, individual treatment was terminated once this PS plateau criterion was met concurrently with a sustained local sO₂ below 40%.

The platform-bleaching rate (C_platform, %/min) was defined using the normalized photosensitizer concentration C_norm(t). From the onset of illumination to the PS plateau time T_plateau, the absolute value of the average slope of C _norm (t) was taken as the platform bleaching rate.

The half-bleaching rate (C_half, %/min) was calculated from the normalized photosensitizer curve C_norm(t) between the onset of illumination and the time point T_1/2 at which C_norm(t) first decreased to 50%. The absolute value of the average slope of C_norm(t) over this interval was taken as the half-bleaching rate, and the corresponding time T_1/2 was recorded as the half-bleaching time.

2.10. Statistical analysis

All data are presented as mean ± standard deviation (SD). Statistical analysis was performed using GraphPad Prism software. Data were first tested for normality (Shapiro-Wilk test). For datasets following a normal (Gaussian) distribution, between-group comparisons were made using one-way analysis of variance (ANOVA) followed by Tukey’s honest significant difference (HSD) post hoc test for pairwise comparisons. If the assumption of normality was violated, a non-parametric Kruskal-Wallis test was used in place of ANOVA, with Dunn’s multiple-comparison test for post hoc analysis. A p < 0.05 was considered statistically significant for all tests.

3. Results

3.1. High-irradiance V-PDT induces rapid photosensitizer bleaching and early oxygen limitation

Figure 2 illustrates the representative temporal evolution of intravascular HMME and vascular oxygenation under the conventional high-irradiance, short-duration V-PDT group (100 mW/cm², 10 min continuous illumination). Once irradiation begins, the photosensitizer concentration in the target vessels falls rapidly: the normalized HMME signal drops to ∼50% of its initial value at approximately 210 s and approaches a plateau of ∼25% around 460 s.

Fig. 2.

Spatiotemporal dynamics of photosensitizer bleaching and vascular oxygenation under high-irradiance, short-duration V-PDT (100 mW/cm², 10 min). (a) Photosensitizer kinetics: (i) spatiotemporal distribution maps and (ii) quantitative concentration curves from two ROIs. (b) Oxygen saturation dynamics: (i) spatiotemporal distribution maps and (ii) quantitative sO₂ curves from the same ROIs.

In parallel, the sO₂ maps reveal a pronounced and rapid decline in vascular oxygenation during the early phase of high-irradiance exposure. Quantitative region of interest (ROI) curves show that sO₂ falls below the 30% hypoxia threshold at roughly 280–380 s. After hypoxia is established, both PS bleaching and oxygen dynamics reach a plateau: beyond ∼6 min of illumination, HMME levels change very little despite continued exposure, while sO₂ remains depressed until the end of the 10 min treatment. These observations collectively demonstrate that conventional high-irradiance PDT rapidly depletes both photosensitizer and oxygen, driving the tissue into an early and persistent hypoxic state. Consequently, a significant portion of the delivered light dose is under oxygen-limited conditions, where the incremental photochemical yield is inherently low.

In the 50%-PS group, illumination was interrupted when real-time HSI indicated that intravascular HMME had fallen to half of its baseline value. Figure 3(a) shows that the PS signal within the ROI declined steeply during the first few minutes, and treatment typically ended at ∼200 s, leaving exactly 50% of the PS unbleached by design. Because the exposure window was truncated so early, tissue oxygenation in most animals never fully evolved into a stable hypoxic phase, and sO₂ often remained above the 30% threshold when the light was turned off. Consequently, standard oxygen consumption metrics (K_sO2 and T_hypoxia) could not be reliably quantified and appear as “N/A” in Table 2.

Fig. 3.

Spatiotemporal dynamics of photosensitizer bleaching and vascular oxygenation under PS-threshold based high-irradiance V-PDT (50%-PS and 30%-PS groups, 100 mW/cm²). (a) 50%-PS group photosensitizer kinetics, showing (i) spatiotemporal distribution maps and (ii) quantitative concentration curves from two ROIs. (b) 50%-PS group oxygen saturation dynamics, showing (i) spatiotemporal distribution maps and (ii) quantitative sO₂ curves from the same ROIs. (c) 30%-PS group photosensitizer kinetics, showing (i) spatiotemporal distribution maps and (ii) quantitative concentration curves from two ROIs. (d) 30%-PS group sO₂ dynamics, showing (i) spatiotemporal distribution maps and (ii) quantitative sO₂ curves from the same ROIs.

Table 2. Quantitative V-PDT kinetic and vascular response metrics across six illumination strategies.

V-PDT Protocol	Chalf	T 1/2	PS rem.	Cplatform	T plateau	K sO2	Thypoxia	VSR0h	VSR24h	Ttotal	Total Fluence
	(%/min)	(s)	(%)	%/min	(s)	(%/min)	(s)	(%)	(%)	(s)	(J/c m2)
50%-PS group	14.08 ± 1.60	221.2 ± 23.6	50	N/A	N/A	N/A	N/A	9.63 ± 3.00	19.75 ± 3.76	221.2 ± 23.6	22.12 ± 2.36
30%-PS group	13.27 ± 2.31	230.3 ± 34.9	30	N/A	N/A	11.60 ± 1.87	286.5 ± 40.4	17.21 ± 5.07	31.38 ± 8.67	396.0 ± 38.0	39.60 ± 3.80
Dual-threshold group	13.59 ± 2.35	233.5 ± 36.5	26.4 ± 4.53	9.47 ± 0.88	466.2 ± 32.5	10.83 ± 1.49	293.0 ± 48.9	20.15 ± 4.78	34.82 ± 6.48	466.2 ± 32.5	46.62 ± 3.25
High-irradiance group	14.59 ± 2.12	209.6 ± 24.0	25.1 ± 4.28	9.70 ± 1.07	463.2 ± 44.0	10.96 ± 1.18	290.0 ± 65.0	22.52 ± 6.91	37.79 ± 8.52	600	60
Low-irradiance group	3.09 ± 0.25	1006.8 ± 94.1	18.3 ± 3.46	2.99 ± 0.71	1639.0 ± 32.4	2.93 ± 0.36	1521.5 ± 214.4	53.23 ± 10.94	64.81 ± 12.02	2400	60
Adaptive feedback group	7.05 ± 0.80	421.5 ± 47.9	21.8 ± 5.74	3.84 ± 0.29	1221.5 ± 25.4	3.86 ± 0.55	1071.0 ± 103.4	48.73 ± 11.55	64.72 ± 11.18	1221.5 ± 25.4	53.23 ± 2.21

In the 30%-PS group, the same high fluence rate was maintained, but illumination continued until the PS concentration had dropped to 30% of its initial value. Figure 3(c) indicates that achieving this deeper level of photobleaching required a substantially longer exposure of approximately 400 s. Over this extended time window, oxygen dynamics were fully expressed: the average oxygen depletion rate rose to 11.60 ± 1.87%/min, and the 30% sO₂ hypoxia threshold was crossed at 286.5 ± 40.4 s (Table 2), closely mirroring the behavior of the fixed high-irradiance regimen.

In designing the PS-threshold regimens, the choices of 50% and 30% residual photosensitizer were not arbitrary, but grounded in V-PDT dosimetry and our preliminary data. The 50% level corresponds to the classical half-bleaching point, a well-established kinetic landmark in V-PDT, and therefore serves as an intentionally conservative early stopping rule that is conceptually analogous to “moderate illumination” and, methodologically, defines a clear lower bound for under-dosing at high fluence rates. In contrast, the 30% threshold was motivated by our observation that conventional 100 mW/cm², 10 min V-PDT typically ends with an intravascular HMME fraction of roughly 15–30%; choosing 30% was thus a pragmatic attempt to approximate the effective terminal PS level of standard dosing while truncating the potentially redundant tail of irradiation. Nevertheless, because inter-subject variability in pharmacokinetics, light distribution, and microvascular oxygenation, any single fixed PS percentage (50% or 30%) used as a standalone stopping rule remains intrinsically non-personalized and may correspond to premature termination in some subjects and unnecessary over-exposure in others.

In contrast, if the PS endpoint is defined kinetically rather than purely by percentage—i.e., a PS plateau is declared when the instantaneous bleaching rate approaches zero over a sustained interval—and this rate-based PS plateau is jointly combined with an oxygen criterion (for example, when sO₂ remains persistently below ∼40% without meaningful recovery), the stopping point becomes physiology-informed and intrinsically adaptive. In this dual-threshold framework, illumination is terminated only when two conditions are simultaneously met: (i) the PS is no longer being appreciably consumed, indicating that therapeutically meaningful photochemistry has effectively run its course, and (ii) the vasculature has settled into a stable hypoxic state below the predefined sO₂ threshold. This shift from a fixed “residual percentage” rule to a rate-based PS plateau plus oxygen-threshold rule allows the end of treatment to self-adjust to each subject’s own drug-light-oxygen balance, capturing the true individual therapeutic endpoint rather than imposing an arbitrary residual PS fraction on all V-PDT courses.

In the dual-threshold group, this concept was pushed one step further by continuing illumination at 100 mW/cm² until the rate of PS bleaching itself fell to nearly zero while sO₂ remained depressed. As shown in Fig. 4(a), the intravascular HMME signal initially declined steeply and then progressively approached a plateau, ultimately stabilizing at approximately 26% of its baseline level. One-way ANOVA confirmed that K_sO2 and T_hypoxia did not differ significantly among the 30%-PS, dual-threshold, and high-irradiance groups (p > 0.05 for both metrics), indicating that prolonging high-irradiance illumination beyond the onset of severe hypoxia—and even until the PS bleaching rate collapses—adds little in terms of additional PS consumption or oxygen utilization, while substantially extending the time spent in an inefficient, oxygen-limited regime.

Fig. 4.

Photosensitizer and oxygen dynamics in the dual-threshold high-irradiance V-PDT group. (a) Photosensitizer dynamics: (i) spatiotemporal maps of normalized intravascular HMME distribution; (ii) quantitative HMME time courses in two selected ROIs. (b) Oxygenation dynamics: (i) spatiotemporal maps of sO₂; (ii) quantitative sO₂ time courses in the same ROIs.

The downstream impact of these different stopping rules on vascular injury is summarized in Fig. 5, which compares PAM images of target vessels acquired before PDT, immediately after treatment, and 24 h later for the high-irradiance, 50%-PS, 30%-PS, and dual-threshold groups. The seemingly more pronounced capillary-level “damage” mainly reflects the vascular mechanism of V-PDT in our study: intravascular HMME-mediated photochemical insult to the endothelium can trigger endothelial swelling, platelet aggregation, and microthrombotic occlusion, and because capillaries/post-capillary venules have very small lumens, even modest edema or microthrombi can rapidly block flow, resulting in a marked loss of perfusion signals in PAM. In the 50%-PS group, the target vessel shows only mild narrowing at the end of irradiation, and the damage deepens slightly by 24 h, but remains limited overall (VSR₀h = 9.63 ± 3.00%; VSR₂₄h = 19.75 ± 3.76%). The 30%-PS group exhibits a moderate degree of acute constriction followed by partial reperfusion, corresponding to intermediate VSR values (VSR₀h = 17.21 ± 5.07%; VSR₂₄h = 31.38 ± 8.67%). The high-irradiance group induced somewhat stronger immediate damage (VSR₀h = 22.52 ± 6.91%) and achieved a modest but clearly greater level of persistent occlusion at 24 h (VSR₂₄h = 37.79 ± 8.52%). The dual-threshold group achieved vascular damage outcomes comparable to those of the high-irradiance group (VSR₀h = 20.15 ± 4.78%; VSR₂₄h = 34.82 ± 6.48%).

Fig. 5.

Vascular responses to high-irradiance, 50%-PS, 30%-PS, and dual-threshold V-PDT protocols. (a) Narrow-band (550 nm) optical images acquired before irradiation. (b) OR-PAM images of the target vessels obtained before treatment, immediately after PDT, and 24 h post-V-PDT for each protocol.

Tukey’s post hoc tests showed that the 50%-PS regimen was significantly inferior to all other high-irradiance schemes at both 0 h and 24 h (all p < 0.01), highlighting the clear under-dosing associated with an early 50% stopping rule. By contrast, the 30%-PS, dual-threshold, and high-irradiance groups clustered at a similar, moderate efficacy level, with no significant differences in VSR among these three conditions (p > 0.05). Taken together, these findings indicate that using a fixed residual PS fraction of 50% or 30% as the sole termination criterion under high irradiance does not fully exploit the therapeutic potential of PDT: an early 50% threshold yields overt under-treatment, whereas lowering the threshold to 30% mainly prolongs illumination in a severely hypoxic and kinetically inefficient regime, with only limited incremental treating effect in long-term vascular shutdown.

Notably, the dual-threshold regimen achieved a therapeutic outcome that was virtually indistinguishable from the conventional high-irradiance, short-duration. Despite a significantly shorter period of effective high-fluence exposure, the dual-threshold group produced a 24 h vascular shrinkage ratio that was almost identical to that of the high-irradiance group, with a comparable residual photosensitizer fraction at the end of treatment (both ≈ 25%). These data indicate that, under continuous 100 mW/cm² irradiation, the latter part of the conventional exposure is delivered when the PS bleaching rate has already fallen to near zero and the photochemical reaction is essentially exhausted. Beyond this point, additional light produces negligible further PS consumption or vascular damage, yet continues to accumulate as unnecessary dose. This supports using the in vivo PS bleaching rate as a biologically grounded, real-time stopping criterion for individualized V-PDT: once HSI monitoring shows that PS kinetics have entered a stable plateau under persistently low oxygen, illumination can be terminated without compromising vascular occlusion. In practice, such a PS-kinetics–based termination rule preserves the therapeutic effect of long high-irradiance V-PDT while shortening treatment time, reducing patient burden, and offering a practical route to efficiency-optimized, personalized PDT dosing.

3.2. Low-irradiance V-PDT preserving oxygen at the expense of extended treatment time

Figure 6 shows the spatiotemporal evolution of sO₂ and HMME under the low-irradiance, long-duration group (25 mW/cm², 40 min continuous illumination). After light onset, oxygen saturation in both ROIs decreased only gradually, with a markedly slower slope than in any high-irradiance regimen. Quantitatively, the oxygen depletion rate in this group was limited to 2.93 ± 0.36%/min, which was significantly lower than in all continuous 100 mW/cm² protocols (one-way ANOVA, p < 0.001; Tukey post hoc, all pairwise comparisons vs. low-irradiance, p < 0.001). The 30% sO₂ hypoxia threshold was not reached until approximately 25 min into treatment (T _hypoxia = 1521.5 ± 214.4 s), again significantly later than in any high-irradiance group (all p < 0.001).

Fig. 6.

Oxygen and photosensitizer dynamics under the low-irradiance, long-duration V-PDT protocol. (a) Photosensitizer kinetics: (i) Spatiotemporal maps of normalized HMME concentration during continuous low-irradiance illumination (25 mW/cm², 40 min); (ii) Quantitative PS bleaching curves from two selected ROIs. (b) sO₂ dynamics: (i) Spatiotemporal maps of sO₂ evolution throughout illumination; (ii) Quantitative sO₂ curves from the corresponding ROIs.

The PS kinetics in Fig. 6(a) reveal a similarly “stretched” photochemical window. Throughout the entire 40-min exposure, the intravascular HMME signal declined slowly but persistently. The half-bleaching time in the low-irradiance group was 1006.8 ± 94.1 s, almost five times longer than in the high-irradiance protocols, and the PS curve did not approach a stable plateau until ≈ 1639.0 ± 32.4 s. Beyond this time point, the HMME signal changed only minimally despite continued illumination, indicating that the system had effectively reached a PS plateau. By the end of the 2400 s exposure (60 J/cm² total fluence), the residual PS fraction had been reduced to 18.3 ± 3.46%. Thus, the low-irradiance strategy achieved a high degree of photosensitizer utilization while keeping oxygen consumption at a moderate level, but at the cost of the longest treatment time among all protocols.

The corresponding vascular response is shown in Fig. 7. Representative PAM images reveal pronounced constriction of the target vessels immediately after low-irradiance PDT, followed by an extensive loss of perfused microvasculature at 24 h. Quantitatively, the low-irradiance group achieved one of the strongest therapeutic responses among all protocols, with VSR₀h = 53.23 ± 10.94% and VSR₂₄h = 64.81 ± 12.02%.

Fig. 7.

Representative vascular responses under the low-irradiance, long-duration PDT protocol. (a) Narrow-band (550 nm) optical images acquired before irradiation. (b) OR-PAM images of the target vessels obtained before treatment, immediately after V-PDT, and 24 h post-V-PDT.

When these vascular outcomes are interpreted together with the kinetic metrics in Table 2, a coherent mechanistic picture emerges. By operating at a low fluence rate, this “conservative” protocol slows oxygen consumption, keeps sO₂ closer to the near-threshold range for a prolonged period, and avoids the deep, sustained hypoxia that characterizes continuous high-irradiance PDT. At the same time, PS bleaching is not curtailed but merely redistributed over a much longer time window: PS is steadily consumed under conditions where oxygen is still available to support ROS generation, and blood flow is maintained long enough to deliver both oxygen and drug to the target microvasculature. This combination—sustained PS utilization, preserved perfusion, and controlled oxygen depletion—appears to maximize the cumulative photodynamic insult to the vessel wall and favor the formation of stable thrombotic occlusion, thereby explaining the superior and durable VSR in the low-irradiance group. The primary drawback, however, is practical rather than mechanistic: the requirement for a 40 min continuous exposure makes this regimen time-consuming and may limit its direct applicability in routine clinical workflows.

3.3. Adaptive feedback V-PDT balances therapeutic efficacy and treatment efficiency

Although the dual-threshold high-irradiance group refined the illumination endpoint, it failed to enhance biological efficacy; the resulting long-term VSR was comparable to conventional fixed-duration treatments. Optimizing termination at a single high fluence rate may remove excess light, but it does not fundamentally reshape the spatiotemporal landscape of oxygen and photosensitizers. Consequently, we evolved the framework beyond a PS-driven rule into an adaptive feedback loop. This approach synergizes a PS-plateau termination criterion with real-time, oxygen-informed irradiance cycling.

In the proposed protocol, the HSI system continuously monitors both sO₂ and intravascular HMME in the target vessels. Illumination is initiated at 100 mW/cm² to rapidly engage photochemistry. Whenever the local sO₂ in the ROI falls below 40%, the system automatically switches the fluence rate down to 25 mW/cm², allowing partial reoxygenation while maintaining a lower but non-zero level of PS bleaching. Once sO₂ recovers above the 40% threshold, the irradiance is restored to 100 mW/cm². This 40% cutoff was chosen as a pre-emptive control point rather than a strict definition of hypoxia. Our previous observations, together with classical Type II PDT kinetics, indicate that when sO₂ drops into the ∼20–30% range, singlet-oxygen generation and PS bleaching efficiency decline steeply and the system rapidly enters a regime of severe, treatment-limiting hypoxia. By setting the control threshold at 40%—slightly above the 30% hypoxia definition used elsewhere in this study—we establish a physiological “safety margin” that prevents the tissue from dwelling in this deeply oxygen-starved state. Functionally, the adaptive controller keeps the reaction operating in a near-optimal oxygen window, where supply and consumption remain approximately balanced.

Figure 8 illustrates the practical consequence of this design. Under the adaptive feedback protocol, high-irradiance segments drive fast PS bleaching and transient sO₂ declines, but each time sO₂ approaches the 40% setpoint, the system automatically transitions to low irradiance, permitting oxygen to rebound. As a result, the sO₂ trajectory exhibits a characteristic oscillatory pattern around the control threshold, rather than the monotonic collapse seen under continuous 100 mW/cm² illumination. In parallel, PS kinetics show alternating fast and slow bleaching phases, ultimately converging to a deep but oxygen-supported depletion before the PS-plateau criterion triggers treatment termination. By integrating a rate-based PS plateau as the stopping rule with an sO₂-based 40% threshold as the switching rule, this dual-threshold adaptive PDT paradigm unifies the strengths of Sections 3.1 and 3.2—PS-guided termination and low-irradiance oxygen preservation—into a single, physiology-informed, individualized dosing strategy.

Fig. 8.

Oxygen and photosensitizer dynamics under the adaptive feedback V-PDT protocol. (a) PS kinetics under dual-threshold adaptive irradiation. (i) Spatiotemporal maps of normalized HMME concentration within the dorsal skinfold window during adaptive switching between 100 and 25 mW/cm². (ii) Quantitative PS bleaching curves from two representative ROIs. (b) sO₂ dynamics under the same protocol. (i) Spatiotemporal maps of sO₂ evolution during adaptive PDT. (ii) Quantitative sO₂ curves from the corresponding ROIs, highlighting the near-threshold oscillatory oxygen pattern induced by feedback-controlled irradiance switching.

At the cohort level, the adaptive group exhibited an oxygenation profile that lay between the high- and low-irradiance extremes, but in a way that was clearly more favorable for V-PDT. The mean oxygen depletion rate was 3.86 ± 0.55%/min, and hypoxia was reached at 1071.0 ± 103.4 s. Both metrics differed significantly from all continuous 100 mW/cm² regimens (one-way ANOVA with Tukey post hoc, both p < 0.001), indicating slower oxygen consumption and a substantially delayed onset of hypoxia. By contrast, there was no significant difference in K_sO2 between the adaptive and low-irradiance groups (p > 0.05), suggesting that, despite intermittent bursts of high-irradiance, the adaptive scheme kept the overall oxygen consumption within the same moderate range as the purely low-irradiance protocol.

The PS kinetics in Fig. 8 show a corresponding biphasic bleaching pattern. During high-irradiance segments, intravascular HMME decreased rapidly, whereas during low-irradiance intervals the bleaching rate slowed but did not cease, while sO₂ was allowed to recover. On average, the adaptive protocol reached half-bleaching at 421.5 ± 47.9 s. The final residual PS fraction was 21.8 ± 5.74%, statistically indistinguishable from both the low-irradiance and fixed-high regimens (pairwise comparisons, all p > 0.05). Importantly, the total treatment time in the adaptive group was 1221.5 ± 25.4 s, roughly half that of the low-irradiance protocol (p < 0.001), with the cumulative delivered fluence remaining well within clinically acceptable limits.

The vascular outcomes of the adaptive feedback protocol are shown in Fig. 9. PAM images acquired before V-PDT, immediately after illumination, and at 24 h post-treatment demonstrate a pronounced and sustained collapse of the target vasculature within the irradiated window. Quantitatively, this impression is confirmed by the VSR metrics: the adaptive group achieved an immediate VSR₀h of 48.73 ± 11.55% and a 24 h VSR₂₄h of 64.72 ± 11.18%.

Fig. 9.

Representative vascular responses under the adaptive feedback V-PDT protocol. (a) Narrow-band (550 nm) optical images acquired before irradiation. (b) OR-PAM images of the target vessels obtained before treatment, immediately after V-PDT, and 24 h post-V-PDT.

Table 2 provides a direct comparison between the planned fixed-dose regimens and the delivered dose under feedback control. Notably, the adaptive feedback protocol produced a substantially higher VSR₂₄h while delivering a lower total fluence than the conventional fixed high-irradiance regimen, indicating improved dose efficiency rather than dose escalation. This supports the clinical value of imaging-guided dosing as a decision-support overlay, where the standard prescription can serve as an upper bound while physiological feedback reduces redundant low-yield illumination. One-way ANOVA on VSR₂₄h identified the adaptive and low-irradiance groups as the top-efficacy tier: their 24 h VSR values were significantly higher than those of the 30%-PS, dual-threshold, high-irradiance, and 50%-PS groups (all p < 0.001). These findings indicate that the adaptive feedback scheme preserves the long-term therapeutic benefit of a conservative low-irradiance protocol, but with substantially improved temporal efficiency.

4. Discussion and conclusion

Under comparable baseline intravascular HMME levels (approximately 67–69 μg/mL), the four high-irradiance protocols exhibited similar early-phase PS kinetics. Their half-bleaching rates clustered tightly around 13–15%/min, with half-bleaching times on the order of 210–235 s. This indicates that once irradiation is initiated at 100 mW/cm², the first 3–4 minutes of photochemical reaction proceed in essentially the same way, regardless of how or when the treatment is ultimately stopped. In other words, simply changing the stopping rule does not fundamentally reshape the “front end” of the bleaching window; it primarily alters the morphology of the late tail of the PS decay curve.

In contrast, the low-irradiance continuous protocol and the adaptive feedback protocol do not merely bleach “more or less” PS, but actively reconfigure the temporal architecture of PS consumption. In these two groups, both the half-bleaching time and the PS plateau time were markedly prolonged, and HMME concentration declined more slowly but more persistently throughout the entire treatment period. Rather than rapidly exhausting the photosensitizer within a few minutes, effective photochemistry was distributed over an extended time window. This temporal stretching of PS utilization provides additional degrees of freedom for oxygen replenishment and for dynamic adjustment of the light dose, a behavior that is inherently difficult to achieve with conventional, single-intensity high-irradiance strategies.

Combined analysis of the oxygen depletion rate and time-to-hypoxia highlights two key mechanistic insights. First, rapid induction of deep hypoxia is not synonymous with effective V-PDT. All fixed 100 mW/cm² regimens in which oxygen dynamics could be tracked drove sO₂ below the 30% threshold within roughly 5 min, with depletion rates on the order of 11%/min. Although this pattern of oxygen consumption appears “aggressive”, these protocols achieved only moderate levels of long-term vascular shutdown. Once tissue is pushed into a state of sustained, severe hypoxia, a substantial fraction of the subsequent light delivery occurs under conditions of combined oxygen and PS scarcity, where the incremental photochemical yield per additional joule of light becomes very low.

By contrast, the low-irradiance and adaptive protocols maintained the tissue in a more favorable operating regime for a much larger portion of the treatment. Rather than keeping sO₂ perpetually high, these strategies allowed oxygen saturation to spend prolonged periods near—but not far below—the hypoxic threshold. Functionally, this means that a large share of the light dose was delivered within a “near-threshold oxygen window,” where oxygen is continuously replenished and remains available for ROS generation, while deep, long-lasting hypoxia is largely avoided. Such fine-tuned control of the oxygen-dose relationship is difficult to achieve with single-intensity, non-adaptive illumination schemes.

When PS kinetics, oxygen dynamics, vascular endpoints, and treatment time are considered together, each protocol occupies a distinct position in a multidimensional efficacy-efficiency-safety space. The continuous low-irradiance regimen sits at one corner of this space: it delivers near-maximal and durable vascular shutdown, but only at the cost of the longest illumination time, which limits its practicality in a clinical workflow. In contrast, the various fixed high-fluence and single PS-threshold schemes cluster in regions characterized either by clear under-dosing (insufficient efficacy at short exposure) or by inefficient over-dosing (only moderate efficacy despite additional time and dose delivered under hypoxic conditions).

Within this landscape, the dual-threshold adaptive protocol emerges as a near-Pareto-optimal solution. Quantitatively, it achieves a 24 h VSR that is essentially indistinguishable from that of the low-irradiance “gold standard”, while cutting the treatment time by roughly half. At the same time, it avoids the prolonged deep hypoxia seen in fixed high-fluence regimens and does not depend on arbitrary PS-percentage stopping rules that tend to push the system toward either under- or over-dosing. Instead, by jointly enforcing an oxygen threshold and a PS-plateau criterion, the adaptive strategy keeps the photochemical reaction operating in a physiologically favorable regime and terminates illumination once the response clearly enters a diminishing-returns phase.

Although this study was validated in a benign superficial vascular model, focusing on benign vascular lesions does not imply that the proposed platform is irrelevant to tumor-associated vascular-targeted PDT. Rather, the goal of the present work is to establish and validate a generalizable closed-loop dosing methodology, in which real-time sO₂ is used to adapt irradiance to avoid oxygen-limited photochemistry, and real-time photosensitizer kinetics are used to determine an efficient treatment termination point and reduce low-yield irradiation. These control principles are expected to be transferable to other vascular-targeted contexts, including tumor V-PDT, where treatment outcome is also strongly coupled to oxygen availability and vascular response.

At the same time, we acknowledge that tumor settings introduce additional complexities that may require protocol refinement and further validation, such as baseline hypoxia, spatially heterogeneous microenvironments, and post-treatment reperfusion/angiogenesis that may contribute to recurrence; therefore, long-term vessel regrowth cannot be ruled out by an acute endpoint alone and should be evaluated with extended follow-up. Future studies will therefore evaluate the proposed closed-loop framework in tumor models and investigate how tumor-specific physiological constraints can be incorporated into the feedback policy and endpoint definitions.

In clinical V-PDT, light delivery is typically prescribed a priori using fixed irradiance and a planned treatment fluence. Here, our dual-threshold adaptive strategy is positioned as a decision-support overlay: the clinician defines the initial settings and conservative safety limits, while the system adaptively modulates irradiance based on real-time sO₂ and terminates illumination when PS bleaching reaches a plateau under sustained hypoxia, indicating diminishing photochemical returns. This physiology-informed, dual-criterion control is well suited for superficial vascular diseases where wide-field optical monitoring can be readily integrated into existing PDT setups, and it provides a practical framework for individualized dosimetry in which key control parameters (e.g., oxygen threshold, PS-plateau definition, and maximum treatment time) can be tuned for different disease entities or patients to achieve a better efficacy-efficiency balance than a fixed one-size-fits-all prescription.

In this study, we used a hyperspectral-photoacoustic dual-modality imaging platform to systematically compare six representative V-PDT illumination strategies in vivo and to quantitatively dissect how different light-delivery schemes reshape photosensitizer kinetics, oxygen dynamics, and microvascular damage endpoints. Building on these mechanistic insights, we introduced an “oxygen-PS dual-threshold” adaptive V-PDT strategy that continuously monitors sO₂ and intravascular PS bleaching, dynamically switches between 100 and 25 mW/cm², and actively terminates illumination when the PS signal enters a true plateau. This feedback-guided protocol markedly alleviated prolonged deep hypoxia while preserving efficient PS utilization, achieved 24 h vascular shrinkage ratios comparable to those of the low-irradiance “gold-standard” regimen, and simultaneously reduced the total illumination time by approximately half. In the combined efficacy-efficiency-safety space, the dual-threshold adaptive strategy outperformed all fixed high-fluence and single PS-threshold protocols, which either under-dosed or wasted dose under oxygen-limited conditions.

Overall, our findings demonstrate that using both sO₂ threshold and PS-bleaching plateau as physiologically meaningful feedback criteria enables closed-loop, adaptive control of PDT dose that is better suited to approach an optimal therapeutic window on a per-subject basis than conventional “fixed fluence plus fixed time/fixed PS percentage” paradigms. The HSI-PAM dual-modality system and the accompanying suite of quantitative metrics established here provide experimental evidence and a methodological framework for rational, disease- and patient-specific V-PDT light-dosimetry design. More broadly, this work lays the groundwork for incorporating real-time physiological feedback into future clinical V-PDT workflows to support truly personalized, efficiency-optimized vascular-targeted therapy.

Funding

National Natural Science Foundation of China https://ror.org/01h0zpd94 ( U22A2092, 62175198); Xi'an Jiaotong University Doctoral Student “Interdisciplinary Training Support Program” ( IDT2217); Open Science Foundation of the Key Laboratory of Hepatobiliary Technology of Mengchao.

Disclosures

The authors declare no conflicts of interest.

Author contributions. R.Z. designed the experiments and coordinated the overall study. R.Z., J.L. (JING LIU), and J.L. (JING LV) performed the experiments and drafted the manuscript. J.L. (JUNDUO LIU) conducted experiments and carried out imaging data analysis. Y.M. and Y.W. provided technical support and experimental advice. J.Z. and L.F. contributed to experimental preparation and system operation. W.Z. critically revised and edited the manuscript. Q.R. and C.Y. conceived the study, provided overall supervision and resources, and finalized the manuscript.

Data availability

The data generated in the present study are available from the corresponding author upon reasonable request.