Bridging preclinical and clinical fluorescence-guided surgery with advanced cancer vision goggles

Abstract

Near-infrared (NIR) fluorescence-guided surgery (FGS) is limited by operator-dependent acquisition and non-uniform datasets, hindering quantitative comparison between users, devices, and institutions. To address these limitations, we evaluated an advanced wearable Cancer Vision Goggles (CVG) platform that standardizes imaging via dual green-pointer alignment, enabling reproducible acquisition geometry. The preclinical component benchmarked imaging standardization, quantitative robustness, and agreement with established systems, whereas the clinical arm assessed feasibility and performance relative to an FDA-approved system. Performance was evaluated using quantitative endpoints, including tumor-to-nontumor ratio (TNR), normalized intensity maps (NIMs), and Sørensen-Dice (Dice) coefficient spatial overlap. CVG achieved comparable or superior tumor contrast with high spatial overlap, as confirmed by these quantitative analysis metrics. Unlike handheld systems, CVG maintained stable fluorescence detection with no significant change in tumor-to-nontumor ratio from 10 to 60 cm, enabling reproducible quantitative imaging over a broad working-distance range. Extension to human tumors from patients injected with an NIR molecular probe (ClinicalTrials.gov: NCT05576974, 04/08/2025) demonstrated performance equivalent to that of an established FGS system with a substantial footprint in the operating room. In addition, CVG provided practical advantages through standardized single-operator acquisition, reduced operator-dependent variability relative to handheld or cart-based imaging, and quantitative real-time threshold-based visualization. These findings establish a quantitatively validated wearable platform that standardizes FGS from preclinical benchmarking to clinically relevant tumor assessment.

Introduction

Cancer remains one of the leading causes of morbidity and mortality worldwide, and surgical resection continues to serve as a primary curative treatment for many solid tumors^1–3. Despite significant progress in preoperative imaging and adjuvant therapies, incomplete tumor removal and positive surgical margins are common occurrences, strongly correlating with disease recurrence and poor patient outcomes^4–6. These challenges underscore the need for intraoperative strategies that enhance tumor delineation and facilitate more precise cancer resections.

Fluorescence-guided surgery (FGS) is an effective intraoperative technique that enables real-time visualization of tumor tissue and surrounding structures^7,8. Near-infrared (NIR) fluorescent dyes, such as indocyanine green (ICG) and tumor-targeted molecular probes currently under clinical investigation, enable improved contrast between malignant and healthy tissues while minimizing interference from ambient light^9–12. Several commercial FGS platforms have been approved for clinical use, and they have demonstrated benefits in sentinel lymph node mapping, vascular perfusion assessment, and tumor margin detection^13,14.

Despite these advances, FGS platforms remain fundamentally qualitative and operator-dependent, limiting their capacity to generate radiometrically standardized, reproducible datasets suitable for quantitative analysis or cross-platform comparison¹⁵. In clinical imaging, uncontrolled acquisition geometry and variable excitation conditions introduce systematic uncertainty, undermining measurement reliability. Many clinically available FGS systems require dimming the operating room lights to enhance fluorescence contrast, which can compromise the surgeon’s direct visualization of the operative field^15–17. In addition, reliance on remote display monitors necessitates repeated shifts of attention between the patient and the screen, disrupting surgical workflow^18,19. These constraints are further compounded in handheld and cart-based systems, where variability in camera-to-tissue distance, viewing angle, and exposure settings leads to user-dependent image quality, non-uniform signal scaling, and inconsistent background normalization. These factors complicate quantitative interpretation, limit reproducibility across operators and institutions, and hinder the development of standardized metrics for evaluating FGS performance.

Parallel to clinical advances, preclinical FGS research has expanded rapidly, with thousands of laboratories worldwide utilizing small animal models to evaluate new NIR fluorescent molecular probes and optimize imaging strategies^18,20–22. Yet, a correlative translational gap remains. Preclinical imaging systems provide high sensitivity and reproducibility but are not designed to mimic intraoperative conditions, while clinical FGS devices are rarely adaptable to routine small-animal studies^16,23. This disconnect complicates the correlation of experimental findings with surgical outcomes in patients. The motivation for a unified imaging platform is not limited to direct clinical translation. Many investigators use murine models exclusively to study tumor biology, therapeutic response, or to develop and screen fluorescent molecular probes and drugs, without immediate plans for human application^24–28. In this context, murine models provide a controlled and scalable platform to evaluate imaging standardization, working-distance robustness, spatial localization accuracy, and reproducibility during FGS tumor visualization. In these settings, the lack of open-field, quantitative fluorescence imaging tools limits the ability to perform longitudinal studies, specimen imaging, and probe characterization under conditions that resemble surgical exposure rather than enclosed benchtop imaging^27,29–31. A wearable platform such as CVG can therefore serve as a preclinical open-field fluorescence imaging tool, supporting standardized, quantitative imaging for mechanistic studies and probe development while remaining compatible with downstream translational workflows.

Wearable platforms, such as the Cancer Vision Goggles (CVG), have the potential to bridge this gap^18,22,32. This ergonomic, portable head-mounted device provides co-registered color and NIR fluorescence images directly in the surgeon’s view, enhancing workflow efficiency in the operating room and eliminating the need to divert attention to external displays. Importantly, the platform can be adapted for preclinical studies, creating a cross-platform benchmarking for data acquisition and quantitative analysis across laboratory and clinical environments. By enabling hands-free operation and consistent imaging geometry, wearable systems offer the opportunity to reduce operator dependence while supporting reproducible, quantitative fluorescence imaging.

In this study, we determined whether a single, standardized wearable platform can support both experimental preclinical and clinical research. We evaluated an advanced CVG platform that ensures a reproducible working distance through a dual green-pointer alignment strategy, thereby reducing user-dependent variability in image acquisition. By standardizing excitation geometry and enabling operator-independent acquisition, this approach supports radiometrically reliable fluorescence imaging under open-field surgical conditions. The device was benchmarked against commercial small-animal and clinical fluorescence imaging systems, with an emphasis on real-time feedback, quantitative accuracy, reproducibility, and usability. Our cross-platform evaluation in preclinical and clinical settings demonstrates that CVG addresses key workflow and reproducibility limitations of existing FGS systems, thereby supporting broader clinical adoption.

Results

Features and workflow of commercial imaging systems and the advanced CVG

To determine whether wearable fluorescence imaging can deliver standardized quantitative performance, we conducted a head-to-head evaluation of CVG against established preclinical and clinical reference systems (Fig. 1). Preclinical fluorescence imaging systems such as the Pearl Small Animal Imager (LI-COR Biosciences) and IVIS SpectrumCT (Revvity) acquire data in fully light-tight chambers to minimize ambient-light interference and provide integrated support for anesthesia, heating, and animal positioning. The Pearl uses a drawer-type imaging bed with a field of view of approximately 11.2 × 8.4 cm. It supports dual-channel NIR fluorescence (e.g., ~700 nm and ~800 nm) in combination with white-light imaging. The IVIS integrates an ultrasensitive cooled CCD for 2D/3D fluorescence and bioluminescence imaging with low-dose micro-CT in a light-tight enclosure. Although the IVIS offers multiple excitation/emission filter combinations from visible to NIR wavelengths, we used only NIR channels compatible with ICG and LS301^33,34 for the preclinical.

Fig. 1. Preclinical and clinical imaging systems were used in this study.

Small animal fluorescence imaging systems (a) Pearl and (b) IVIS SpectrumCT equipped with enclosed, light-tight imaging chambers to minimize ambient light. c Advanced CVG platform for preclinical and clinical NIR cancer imaging. Cart-based clinical FGS systems (d) Quest Spectrum and (e) Stryker SPY-PHI equipped with a handheld camera/light tethered by power and network cords.

We also evaluated the Quest Spectrum (Quest Medical Imaging) and Stryker SPY‑PHI (Stryker Corporation) clinical systems in this study. The Quest Spectrum is a handheld intraoperative fluorescence imaging device that combines visible-light imaging with one or two dedicated NIR fluorescence channels, enabling real-time overlay and navigation during surgery. Use of this device was restricted to handheld imaging in animal studies because institutional guidelines did not permit its use in both animal and human operating rooms. Similar to the Quest Spectrum, the Stryker SPY‑PHI is a cart-based fluorescence imaging system used to guide surgical decisions and assess intraoperative tissue perfusion and lymphatic flow. It supports multiple visualization modes, including white light, fluorescence overlay, grayscale contrast, and color-segmented fluorescence, displayed on a dedicated monitor.

The advanced CVG platform includes several new features designed to improve usability and quantitative imaging in preclinical and clinical settings. The system incorporates automated contrast adjustment, laser safety interlocks, and real-time thresholding to enhance reproducibility, reduce operator variability, and simplify image acquisition. A dual green-laser module projects two visible beams that converge when the device is positioned at the standardized 50 cm working distance, ensuring consistent excitation and minimizing inter-user variability that often affects handheld systems. This feature is particularly important for enabling reproducible data collection across different operators and institutions. Data management was optimized for translational use by enabling Digital Imaging and Communications in Medicine (DICOM)-compliant fluorescence acquisition, which facilitates integration with existing hospital workflows and supports multi-institutional analysis. Ergonomic refinements include a compact, light-weight headset with a waist-mounted power supply, allowing surgeons and researchers to operate without being tethered to a cart-based system.

These CVG design features provide a unified and practical workflow for laboratory and clinical operating room settings. Importantly, they address major limitations of existing enclosed preclinical imaging systems and clinical cart-based FGS platforms by simplifying operation and enabling standardized data collection for experimental and clinical studies. Key imaging system configurations and user-adjustable features are summarized in Supplementary Table 1, with full acquisition parameters provided in the Methods section.

Comparative preclinical tumor imaging with CVG and commercial small animal imaging systems

The preclinical study was used to benchmark the standardization, reproducibility, and spatial localization accuracy of CVG imaging under controlled conditions, comparing it with established imaging systems. LS301 was selected for these studies because its absorption and emission spectra closely overlap those of indocyanine green (ICG), which is widely used to evaluate many clinical imaging systems (Fig. 2a). Each imaging platform was operated using its standard NIR fluorescence excitation and detection settings, without system-specific spectral tuning. This ensured that observed differences in imaging performance primarily reflected system design and image-processing characteristics rather than spectral mismatch. Preclinical fluorescence imaging systems such as Pearl and IVIS are commonly used for noninvasive NIR fluorescence imaging of tumors in mice. To benchmark the imaging performance of the advanced CVG platform with established static small animal imaging systems, we administered the tumor-targeted NIR fluorescent molecular probe, LS301³⁴ intravenously (IV) for side-by-side imaging of breast tumors (4T1) in mice (Fig. 2a). Noninvasive imaging was performed through intact skin, after skin incision and reflection, and the ensuing tumor cavities following surgical excision (Fig. 2b–e). Regions of interest for tumor and nontumor tissues were selected using instrument-specific analysis software for fluorescence intensity measurements and calculation of the tumor-to-nontumor ratio (TNR).

Fig. 2. NIR fluorescence imaging of tumors in mice comparing CVG with benchtop and clinical platforms.

a Excitation and emission spectra of LS301 and preclinical FGS workflow (created with Biorender). b–e, Representative composite images showing pseudocolored NIR fluorescence overlaid on grayscale brightfield images of BALB/c mice (n = 3) bearing subcutaneous 4T1 tumors ~24 h after intravenous injection of LS301-HSA (30 µM, 100 µL demonstrating FGS for tumor resection acquired with: (b) CVG – inset numbers indicate real-time quantitative fluorescence values at various signal thresholds; (c) Pearl – color scale [min = 4.70 × 10⁻², max = 3.22 × 10⁻¹]; (d) IVIS – color scale [min = 350, max = 2000]; (e) Quest Spectrum (Quest) – non-quantitative handheld imaging of dorsal flank regions. Panels from left to right show pre-surgery, exposed tumor after skin deflection, the post-resection, and after incision closure. Ex vivo imaging includes intact tumor (superficial—front; deep—back side), as well as 1.5-mm thick tumor slices. Tumor regions were delineated freehand based on fluorescence intensity and designated by dotted orange lines. f TNRs for CVG, Pearl, and IVIS were not significantly different, while CVG showed significantly higher TNR compared to Quest Spectrum (P = 0.016). The red-circled area indicates the tumor ROI, while the white-circled area indicates the nontumor ROI in (b–e), which were selected for the TNR analysis. Data represent mean ± SD of TNRs from three mice. P values were calculated using one-way ANOVA with multiple comparisons in Prism 8.0.1. Scale bar: 5 mm.

Compared to Pearl and IVIS, the CVG provided equivalent contrast for noninvasive tumor visualization while demonstrating superior confinement of fluorescence to tumor regions. Image analysis shows no significant difference between measured TNR for CVG (2.74 ± 0.22), Pearl (2.70 ± 0.28, P = 0.09), and IVIS (2.40 ± 0.59, P = 0.39) systems (Fig. 2f). Following tumor resection, the surgical cavity and closed incision, CVG revealed only background fluorescence, whereas IVIS and Pearl frequently detected significant nontumor-associated signals from surrounding tissues. While reduced background signals highlight the effectiveness of CVG’s auto-thresholding feature, part of the observed difference may stem from the greater fluorescence-detection sensitivity of these fully enclosed preclinical imaging systems. The real-time thresholding values for the top 10%, top 5%, and the maximum intensity displayed within CVG during FGS clearly showed a high signal of the tumor (53, 57, 71), which dropped to background levels after surgery (35, 36, and 37), indicating complete resection (Fig. 2b).

Neither the IVIS nor the Pearl system is designed for FGS, prompting us to compare the CVG with Quest Spectrum, a clinical cart-based FGS device. The handheld camera of the Quest Spectrum enabled real-time feedback for FGS in preclinical models, but required interrupting the surgical workflow for visualization. In addition, the fluorescence signal intensity observed on screen varied significantly with changes in the camera-to-subject distance. To enable standardized comparison, we fixed the Quest Spectrum imaging head position at 20 cm above the surgical platform, matching the CVG used at 50 cm for FGS (excitation power 10 mW/cm²). To compare distance-dependent fluorescence intensity and TNR, we varied the working distances for the CVG and Quest Spectrum from 10 cm (measured excitation power density ≈ 35 mW/cm²) to and acquired NIR fluorescence images from the same LS301-injected cohort. Representative images in Fig. 2e were obtained at a 20 cm working distance. Across all tested conditions, LS301 tumor fluorescence contrast measured by Quest Spectrum (TNR: 1.82 ± 0.17) was consistently lower than that obtained with other systems (P = 0.016), limiting tumor margin delineation (Fig. 2f). This discrepancy likely arises from differences in excitation/emission parameters and image-processing algorithms contributing to reduced sensitivity ¹⁶.

Our data demonstrates that the low-cost CVG matched or exceeded the imaging performance of more expensive nonsurgical preclinical systems and the Quest Spectrum FGS device under our experimental conditions. By combining real-time visualization with enhanced tumor-to-nontumor discrimination, the CVG enables better representation of intraoperative conditions for FGS than benchtop platforms, improving its adoption as a preclinical FGS and routine small animal fluorescence imaging system.

Quantitative comparison and reproducibility analysis

Visual assessment confirmed that the CVG effectively delineated tumors in preclinical FGS studies, with performance comparable to commercial preclinical and clinical imaging systems. However, qualitative fluorescence overlays alone can be misleading because each imaging platform applies proprietary intensity scaling, colormaps, and dynamic range compression. To quantitatively evaluate imaging performance, we conducted correlation analyses across CVG, IVIS, Pearl, and Quest Spectrum imaging platforms. Because raw fluorescence intensity values are reported and displayed differently by each system, we generated normalized intensity maps (NIMs) by scaling pixel intensities relative to background, defined as the mean of the lowest 10% of signal pixels within each tissue slice. This normalization enables objective, cross-platform comparison that is independent of instrument-specific display parameters or fluorophore brightness.

Furthermore, to avoid reliance on a single threshold and to preserve spatial information, we compared tumor-associated fluorescence regions across multiple intensity levels, ranging from the top 50% to the top 10% of pixels. We also applied an absolute threshold of NIM > 1.5 to capture high-contrast tumor regions (Fig. 3 and Supplementary Fig. 1). This multi-threshold approach enables assessment of both high-intensity tumor cores and more diffuse tumor-associated fluorescence patterns using TNR³⁵ rather than raw intensity values. In CVG images of tumor slices, a clear separation between tumor and background fluorescence was observed, with normalized values of 3.95 ± 1.71 (lowest 10%) versus 40.95 ± 3.70 (top 10%) (Fig. 3a). Similar patterns of high-fluorescence localization were noted across other platforms, with signal differences reflecting platform-specific characteristics: Quest Spectrum (35.98 ± 11.58 vs. 106.25 ± 7.23) (Fig. 3c), IVIS (768.53 ± 98.19 vs. 1880 ± 129.02) (Fig. 3d), and Pearl (0.05 ± 0.02 vs. 0.22 ± 0.02) (Fig. 3e). These values represent relative contrast rather than absolute signal strength, which cannot be directly compared across systems used in this study. Supplementary Fig. 1e-f further illustrates Dice coefficient comparisons using various high-intensity pixel thresholds (10%–50%).

Fig. 3. Spatial correlation of tumor-selective fluorescence signals across imaging systems.

Normalized intensity maps (NIMs) were generated from fluorescence images acquired with different platforms and compared using Dice coefficient analysis to evaluate tumor localization consistency. Representative images of 1.5 mm tumor slices were acquired with the (a–b), CVG. c–e, Quest Spectrum (Quest) (c), IVIS (d), and Pearl (e). c1–e1, by each ‘system’s software. c2–e2, Corresponding visible-light and NIR images transformed to match the spatial orientation of CVG images. c3–e3, Overlay of the top 10% high-intensity fluorescence pixels from CVG (red) and established systems (green); yellow indicates regions of fluorescence overlap: CVG vs. Quest Spectrum (c3), CVG vs. IVIS (d3), and CVG vs. Pearl (e3). c4–e4, NIMs of the transformed NIR images for Quest (c4), IVIS (d4), and Pearl (e4). c5–e5, Overlays of high-intensity NIM regions (NIM > 1.5) comparing CVG with Quest (c5), IVIS (d5), and Pearl (e5). f Dice coefficients comparing the top 10% pixel overlap between CVG and Quest Spectrum, IVIS, and Pearl: P = 0.423 for Quest Spectrum vs. IVIS, P = 0.649 for IVIS vs. Pearl, P = 0.255 for Quest Spectrum vs. Pearl; all not significant (ns). g Dice coefficients comparing NIM regions with intensity > 1.5: P = 0.070 for Quest Spectrum vs. IVIS, P = 0.027 for IVIS vs. Pearl, P = 0.084 for Quest Spectrum vs. Pearl; P values calculated using paired t-tests in MATLAB R2023a, with P < 0.05 considered statistically significant. All top 10% and NIM > 1.5 Dice overlay images are displayed in RGB (1,2,0) format, with the CVG image shown in red and images from other devices (Quest Spectrum, IVIS, and Pearl) shown in green and the resulting overlap is displayed in yellow. Data represent mean ± SD of n = 5 tissue slices.

To benchmark spatial agreement independent of signal magnitude, we computed Dice coefficients between CVG and each established system. The Dice coefficient quantifies the degree of spatial overlap between fluorescence-positive regions and does not reflect intensity or contrast. Strong spatial agreement based on the overlap of the top 10% of high-intensity pixels was observed (mean ± SD): Quest Spectrum (0.61 ± 0.13), IVIS (0.67 ± 0.09), and Pearl (0.70 ± 0.10) as indicated by yellow regions in Fig. 3c5–e5. Despite differences in acquisition geometry between the head-mounted CVG and vertical fixed-camera systems, greater than 60% overlap of tumor-associated regions was achieved. These results confirm that CVG reliably captures the same spatial tumor fluorescence patterns as established preclinical and clinical platforms. Further Dice coefficient analyses using multiple pixel thresholds (10%-50%) and background normalization ranges demonstrate that CVG maintains high spatial agreement over varying signal levels (Supplementary Fig. 1a-d). Importantly, CVG exhibited a higher proportion of high-contrast pixels (NIM > 1.5) than Quest Spectrum, IVIS, or Pearl, consistent with improved tumor-to-background delineation. In addition, the reproducibility of TNR across repeated acquisitions highlights CVG’s reliability for preclinical applications. These analyses demonstrate that CVG provides high spatial localization and quantitative contrast, establishing a standardized framework for cross-platform FGS research.

CVG enables consistent fluorescence imaging across variable working distances

Unlike handheld FGS systems, the CVG platform is designed to acquire fluorescence images across a broad working distance range expected when worn during open surgery (10–60 cm), while maintaining consistent image quality and quantitative performance (Fig. 4). Because fluorescence intensity is directly proportional to excitation power and imaging geometry, standardized excitation conditions are essential for reproducible quantitation. To support this requirement, CVG integrates a dual-green laser alignment module that provides real-time visual guidance to maintain a consistent, pre-calibrated working distance (50 cm). When the two green laser spots converge into a single point, the excitation field geometry is fixed, enabling standardized data acquisition and improving real-time quantitative interpretation across users and imaging sessions.

Fig. 4. Comparison of working distance-dependent fluorescence imaging performance between CVG and Quest Spectrum systems.

A BALB/c mouse bearing a 4T1 tumor received an intravenous injection of LS301-HSA (30 μM, 100 μL), followed by imaging at various working distances. a–c Representative overlay (top row) and NIR fluorescence images (bottom row) acquired using CVG and Quest Spectrum (Quest) at working distances of 15 cm, 30 cm, and 50 cm. Working distance refers to the separation between the animal and the camera/light source. Red arrows indicate the tumor location. d, Fluorescence intensities of tumor and nontumor regions in CVG and Quest Spectrum images at working distances ranging from 10 cm to 60 cm. e Tumor-to-nontumor fluorescence ratios (TNRs) from CVG and Quest systems across working distances from 10 cm to 60 cm. For CVG, TNR remained consistent across all distances, including the recommended 50 cm, with no significant differences (P > 0.5). For Quest Spectrum, TNR at 15 cm was not significantly different from 10 cm (P = 0.9599), but significantly different with increasing working distance: P = 0.0224 at 20 cm vs. 10 cm, P = 0.0006 at 25 cm vs. 20 cm, and P < 0.0001 for the rest of the working distance comparisons. P values were calculated using one-way ANOVA (Prism 8.0.1). Data represent mean ± SD (n = 3), where three images were acquired at each distance, and the corresponding measurements were calculated independently.

To clarify the effect of working distance on signal magnitude and contrast, we evaluated relative fluorescence intensity in tumor and nontumor regions in parallel with TNR. CVG features effective noise suppression while maintaining high sensitivity. As working distance increased, fluorescence intensity from both tumor and adjacent nontumor regions decreased proportionally (Fig. 4e). Because tumor and background signals decreased in parallel, TNR remained stable, showing no significant difference across working distances (e.g., 2.32 ± 0.07 at 10 cm vs. 2.44 ± 0.17 at 60 cm, P > 0.5; Fig. 4e).

In contrast, the Quest Spectrum handheld FGS system exhibited substantial variability in both image quality and quantitative performance as a function of working distance (Fig. 4a–e). At 10 cm and maximum sensitivity settings, Quest Spectrum detected tumor fluorescence (210.86 ± 2.73 vs. 109.85 ± 2.08 for tumor vs. nontumor, P = 6.78 × 10⁻⁴), but elevated background signal limited contrast, resulting in a TNR of 1.92 ± 0.06 (Fig. 4d, e). As the working distance increased, tumor fluorescence intensity decreased sharply, particularly between 15 cm and 30 cm, compared with that observed with CVG. In contrast, fluorescence intensity in adjacent nontumor regions did not decrease proportionally, reflecting the system’s high inherent noise level. Beyond 35 cm, the decrease in nontumor signal was marginal. Because of this asymmetric signal loss, TNR decreased significantly with increasing working distance, falling to 1.71 ± 0.01 at 20 cm (152.78 ± 2.70 vs. 89.10 ± 1.10, P = 9.01 × 10⁻⁴). At a working distance of 30 cm, the vendor-recommended working distance^16,32, both tumor and background signals dropped markedly (79.95 ± 0.61 vs. 50.30 ± 0.36, P = 2.24 × 10⁻⁵), yielding a suboptimal TNR of 1.58 ± 0.02 (Fig. 4d-e).

At low injected LS301 concentrations, Quest Spectrum could reliably detect tumors only at very short working distances. For example, the system produced a strong tumor signal (90.64 ± 1.90) relative to nontumor tissue (45.56 ± 2.44) at 5 cm, yielding a TNR of 1.99 ± 0.078 (P = 1.45 × 10⁻⁵; Supplementary Fig. 2a-f). However, increasing the working distance to 20 cm caused both signals to decrease substantially (14.06 ± 0.46 vs. 10.07 ± 1.11, P = 0.0046), resulting in a TNR of 1.41 ± 0.13, which is below the minimum threshold (TNR = 1.5) generally required for effective real-time FGS³⁶. In contrast, CVG maintained high contrast at extended working distances, achieving a TNR of 3.27 ± 0.34 at 50 cm (58.14 ± 2.66 vs. 17.65 ± 2.35, P = 4.66 × 10⁻⁷; Supplementary Fig. 2g–i).

These results demonstrate that signal intensity alone does not fully describe imaging performance under variable working distances. Therefore, contrast-based metrics such as TNR are essential for evaluating practical FGS performance, especially when raw intensity values are system-dependent or inaccessible. By stabilizing excitation conditions, suppressing working distance-dependent variability, and standardizing data acquisition working distance, CVG enables reproducible quantitative imaging. This enhances reliability for multi-institutional preclinical studies and strengthens its potential as a bridge between benchtop imaging and clinical FGS.

Clinical evaluation of CVG use in excised human tumors

The clinical studies aim to assess the feasibility of CVG and its equivalence to an FDA-approved FGS system when imaging excised human tumor specimens (human tumor). To evaluate the performance of CVG fluorescence detection in the OR setting, we compared the CVG platform to the FDA-approved Stryker SPY-PHI handheld FGS device, imaging resected tumor tissues from patients undergoing head and neck cancer surgery (Fig. 5). Patients received IV injected Pegsitacianine³⁷, a NIR pH-sensitive fluorescent nanoprobe, 24–72 h before surgery. The excised tumors were immediately imaged using both systems. For CVG, visible and NIR grayscale images were acquired at a fixed 50 cm working distance (Fig. 5a, e). Real-time composite maps (Fig. 5b) were generated using a dynamic thresholding algorithm with a rainbow color scale to enhance visual contrast. The quantitative intensity metric shown in the CVG display complements the dynamical contrast-enhancing colormap while providing a quantitative readout based on raw pixel values for the top 5%, top 10%, and maximum pixel intensity (Fig. 5b). For comparison, SPY-PHI images (1080 × 1920, height × width resolution) were acquired using the system’s standard rainbow overlay on grayscale brightfield images without the benefit of quantitative fluorescence intensity feedback (Fig. 5c, d, f). Selected SPY-PHI images were cropped and rescaled to match the CVG output for side-by-side evaluation.

Fig. 5. Assessment of CVG performance against the FDA-approved Stryker SPY-PHI using human tumor tissue.

a Representative visible-light image captured by CVG. b Pseudocolored NIR fluorescence overlay generated using a dynamic thresholding algorithm, overlaid on the visible image. The three boxed values represent fluorescence intensity thresholds for the top 10%, top 5%, and maximum intensity pixels. CVG images were acquired at a resolution of 480 × 640 (height × width) pixels from a fixed 50 cm working distance. c Visible image of the same tissue captured by the handheld Stryker SPY-PHI system (SPY-PHI). d NIR fluorescence (blue-to-red pseudocolor) overlaid on grayscale white-light image obtained by the Stryker SPY-PHI system in SPY-CSF mode. e Raw NIR fluorescence image from CVG. f, NIR fluorescence image obtained by the Stryker SPY-PHI system in SPY contrast mode. g Overlay of the top 15% fluorescence intensity pixels from the CVG NIR image (red) and the transformed Stryker SPY-PHI NIR image (green) in an RGB composite; yellow indicates overlapping regions. h Dice coefficient comparing the top 15% pixels from both systems: 0.84 ± 0.04 (n = 6), with two outliers shown as brown dots (0.50 and 0.59). i Normalized intensity map (NIM) derived from the CVG NIR grayscale image, showing a mean of 19.84 ± 14.23 and a maximum of 54.93. j NIM from the transformed SPY-PHI image, with a mean of 2.76 ± 1.28 and a maximum of 4.77. Both NIMs were normalized by the mean intensity of the lowest 10% of pixels in the tissue sample. k Overlay of high-intensity NIM regions (NIM > 1.5) from CVG (red) and Stryker SPY-PHI (green) in RGB format, where yellow indicates overlap. l, Dice coefficient for NIM > 1.5 comparison: 0.91 ± 0.02 (n = 6). Scale bars, 5 mm.

Across multiple patient specimens, CVG-generated intensity maps correlated strongly with SPY-PHI data, confirming the system’s equivalence as a potential clinical-grade imaging platform. While grayscale images from both devices appeared nearly identical (Fig. 5e, f), the CVG’s real-time color-coded thresholding provided improved delineation of subtle intra-tumoral heterogeneity (Fig. 5b; Supplementary Fig. 3), thereby enhancing the visualization of tumor regions that were less distinct on conventional grayscale fluorescence intensity images. In particular, CVG’s threshold maps closely matched the SPY-PHI’s green-fluorescence overlay, highlighting their clinical similarity (Supplementary Fig. 3; original SPY-PHI images shown in Supplementary Fig. 4). Quantitatively, CVG showed higher contrast, with a maximum TNR value of 54.93 and a mean of 19.84 ± 14.23, compared to the SPY-PHI’s maximum TNR of 4.77 and a mean of 2.76 ± 1.29 (Fig. 5i,j). Dice coefficient analysis revealed high correlation between CVG and Stryker SPY-PHI fluorescence localization for both the top 15% of fluorescence pixels (0.84 ± 0.04) and for pixels with NIM > 1.5 (0.91 ± 0.02) (Fig. 5g,h,k,l). Importantly, the green laser distance guide enabled CVG data acquisition at a fixed working distance, ensuring consistent and reproducible quantitation across samples. As with the Quest system, SPY-PHI image quality varied with changes in working distance and acquisition settings. A key limitation of SPY-PHI is its tendency to produce saturated fluorescence images under SPY contrast mode, which compromises the overlay accuracy for the top 15% fluorescence regions and results in lower Dice scores (0.5 and 0.59) when saturation exceeds 15% of the total pixels (Fig. 5h).

Discussion

This study demonstrates that a head-mounted, binocular NIR fluorescence platform (advanced CVG) can deliver tumor-constrained signaling and quantitative readouts during small-animal surgery and in human tumor assessments, while maintaining agreement with widely used reference instruments. This performance was evaluated as a standardized imaging measurement problem under open-field surgical conditions. Despite differences in optics and image orientation, pixel-level comparisons using NIM and Dice coefficients on tissue sections confirmed substantial spatial overlap between CVG and clinical and preclinical reference devices. In human tumors, the CVG provided detailed fluorescence overlays and thresholded masks that were at least comparable to, and in some views more spatially specific than, a clinical handheld system. These results demonstrate that CVG produces radiometrically interpretable fluorescence measurements that are consistent with established systems, extending its use beyond qualitative visualization. This capability supports the use of CVG as a single imaging platform for preclinical discovery and clinical evaluation.

In addition to quantitative equivalence with preclinical and clinical FGS devices, the CVG introduces workflow-level innovations that directly address known limitations of FGS in the laboratory and the OR^17,28,38. The CVG software, implemented through an intuitive graphical user interface, integrates fluorescence visualization, real-time thresholding, and image and video capture directly into a see-through display. This design minimizes the need for handheld devices and reliance on external display monitors, both of which have been shown to disrupt surgical workflow and contribute to operator-dependent variability^18,19. The addition of a visible green laser distance guide represents a key engineering contribution, as fluorescence signal intensity scales with excitation geometry and working distance. Incorporating a fixed acquisition working distance standardizes fluorescence measurements across different users and sites, enabling real-time quantitative fluorescence readouts and reducing reliance on relative color maps alone. Beyond demonstrating equivalence to an FDA-approved clinical imaging system, CVG offers additional advantages, including real-time multi-thresholding, enhanced tumor delineation, reduced system cost, and improved portability through its wearable design.

Numerous reviews and consensus reports have emphasized the clinical value of NIR FGS and the necessity for reproducible, quantitative performance across different systems and sites^17,28,39. Most prior work has focused either on feasibility demonstrations or on system-specific performance, without establishing cross-platform comparison using common tracers, acquisition geometry, and analysis metrics^15,17,40. Our cross-platform benchmarking directly addresses this gap by co-registering images from preclinical and clinical instruments to CVG data using a shared workflow, tracer, and pixel-wise analytical platform. Prior goggle-style or wearable FGS reports have primarily demonstrated feasibility in animal models^22,32 or in early clinical settings⁴¹, but have not provided a quantitative comparison between preclinical and clinical systems under a shared experimental and analytical workflow. Our approach addresses this gap through standardized acquisition geometry and cross-platform normalization, reducing user-dependent variability that commonly arises from uncontrolled working distance in handheld FGS imaging.

A further practical contribution of CVG is the maintenance of imaging continuity during small-animal procedures. Benchtop systems, such as IVIS Spectrum and LI-COR Pearl, are optimized for high-sensitivity imaging in light-tight enclosures, which necessitates removing the animal from the surgical field. This can interrupt sterile workflow or alter physiologic conditions⁴². These interruptions fragment the procedure and complicate longitudinal quantitation. By contrast, CVG enables continuous, in-field visualization with a fixed, guided working distance, enabling immediate tumor imaging, cavity inspection, and imaging of excised tissue without transport.

Small animal FGS performed under stereomicroscopes presents additional challenges, including photobleaching from microscope illumination and variability in the field of view over time^43,44. Incident laser power delivered by CVG is much less than that used by microscopes⁴⁵. The system also preserves hand-eye coordination and maintains a consistent geometry, improving reproducibility of signal-to-background measurements. Beyond workflow and ergonomics, the translational significance of CVG lies in its ability to harmonize fluorescence imaging data across biological scales and experimental contexts. Preclinical imaging systems differ substantially from clinical devices in geometry, sensitivity, and data output, complicating quantitative comparison across studies⁴⁶. By providing a common, head-mounted platform that captures raw grayscale and overlaid views at a programmed distance and reproducible metadata, CVG aligns with ongoing efforts in the community to standardize FGS evaluation, calibration, and reporting^15,47,48.

We also compared preclinical and clinical imaging devices with clear NIR imaging capabilities, but that vary substantially in design. The preclinical arm was intended to establish standardized open-field imaging performance under controlled conditions (working-distance robustness, spatial localization agreement, and reproducibility), whereas the clinical arm was designed to assess feasibility and comparative performance in human tumors relative to an FDA-approved system. Findings of this study support translational readiness of the imaging approach rather than immediate replacement of established clinical platforms. From an implementation perspective, CVG is designed to reduce FGS workflow challenges through head-mounted visualization, guided acquisition distance, and real-time quantitative overlays. The DICOM-compatible outputs can be incorporated into existing clinical data environments.

Some translational barriers remain. First, prospective intraoperative clinical studies are needed to quantify effects on operative workflow, decision confidence, margin assessment, and time efficiency^49–51. Second, multi-institutional evaluations are required to confirm reproducibility across diverse users, OR environments, and acquisition protocols¹⁵. Third, implementation with other clinically used imaging agents and tumor types is needed to assess generalizability beyond the probes and tumor contexts studied here⁵². Finally, implementation considerations, including user training, institutional integration, and regulatory requirements for specific clinical use cases, will influence adoption in routine practice. We also note that earlier generations of the Pearl and Quest systems were used in this study, which may not reflect the capabilities of their updated models.

In summary, CVG provides a practical bridge between preclinical and clinical FGS by enabling continuous, quantitative imaging with standardized data acquisition. In small animal models, CVG matched the performance of established preclinical systems while providing more tumor-confined fluorescence and reduced background signal. In human tumors, CVG produced results comparable to those of an FDA-approved handheld device, while offering advantages in portability, cost efficiency, and operator independence. Separating preclinical performance benchmarking from clinical feasibility assessment allowed us to establish a structured translational pathway for evaluating wearable FGS imaging systems. CVG supports reproducible probe development, multi-user investigations, and quantitative FGS, thereby advancing this platform from qualitative visualization toward a more rigorous, scalable, and reproducible imaging paradigm.

Methods

Animal models

All animal experiments were performed in accordance with protocols approved by the Institutional Animal Care and Use Committee (IACUC) at the University of Texas Southwestern Medical Center (APN 103253). Mice were acclimated for a minimum of 2 days before any procedures. For tumor-bearing mouse studies, female BALB/c mice (Charles River Laboratories, Texas, USA) were subcutaneously implanted in the right flank with 5 × 10⁵ 4T1-GFP-Luc breast tumor cells. Tumor size was monitored weekly by caliper measurements and enrolled in studies as described below.

Preclinical in vivo FGS and ex vivo imaging

Tumor-bearing mice were enrolled in the preclinical FGS comparison study when maximum diameter reached 15–20 mm (n = 3). For FGS, mice received an intravenous injection of LS301-HSA (30 µM in 1% HSA/PBS; 100 µL) about 24 h prior to surgery. Sustained-release buprenorphine was administered subcutaneously (1.0 mg kg⁻¹) and mice were anesthetized with isoflurane using a SomnoFlo digital vaporizer (Kent Scientific, CT, USA; 2-3% in room air) and maintained continuously by nose cone during FGS and imaging. Primary tumor resection was performed under fluorescence image guidance by a surgeon wearing the CVG. Images of mice, tumors, and surrounding tissues were acquired at multiple stages: pre-surgery, intraoperative exposure, post-resection cavity, and post-closure. Immediately after surgery, mice were euthanized under anesthesia by cervical dislocation. Front (superficial) and back (deep) surfaces of resected tumors were imaged ex vivo, followed by sectioning into 1–2 mm slices for additional imaging. At each stage, imaging was also performed using Pearl (LI-COR Biosciences, Lincoln, NE, USA), IVIS SpectrumCT (Revvity, Waltham, MA, USA), and Quest Spectrum® (Quest Medical Imaging, Middenmeer, The Netherlands) systems. The study utilized the same mouse model to mimic FGS surgery, enabling comparison of CVG with Pearl, IVIS, and Quest Spectrum. A total of three mice were included for the FGS study, providing enough tumor slices for a comparative analysis using the Dice coefficient across all four instruments. There were no predefined inclusion or exclusion criteria, and no animals were excluded from the analysis.

Working distance experiment

To evaluate the effect of working distance on measured fluorescence signal and TNR, mice (n = 3) with tumors approximately 10–15 mm maximum diameter were imaged approximately 24 h after LS301 administration as described above. For working distances ranging from 10 cm to 60 cm, three independent measurements were acquired at 5 cm increments using CVG and QUEST. As the goal of this study was to compare instrument function, data from one mouse were selected for comparative analysis based on fluorescence intensity within an appropriate dynamic range that allowed clear discrimination between the tumor and adjacent nontumor regions using both systems. Due to biological and/or pharmacologic factors, variations in fluorescence intensity led to unsuitable comparisons due to signal loss at long working distances for the Quest (supplementary Fig. S2), while those with excessively high-fluorescence led to signal saturation at short working distances.

Ex vivo imaging of resected human tissues

The study was conducted in accordance with protocols approved by the Institutional Review Board of the University of Texas Southwestern Medical Center (IRB # STU-2022-0460). Imaging of suspected tumor tissues resected from human subjects diagnosed with head and neck cancer and enrolled under the clinical trial (NCT05576974) was performed in the operating room using CVG and Stryker SPY-PHI⁵³. ILLUMINATE is a non-randomized clinical trial led by Dr. Baran Sumer at UT Southwestern Medical Center to assess safety and imaging feasibility study of Pegsitacianine, an intraoperative fluorescence imaging agent, and was registered at ClinicalTrial.gov on April 8th, 2025. Written informed consent was obtained from all participants before study participation. Patients were intravenously injected with Pegsitacianine (1 mg/kg) 24–72 h before the surgery. Full information, including the study plan, intervention, and comparator, outcomes, as well as participation criteria, is available at https://clinicaltrials.gov/study/NCT05576974. Data collection for this study, which aims for a quantitative data and validated comparison of the CVG with other clinical FGS imaging systems, was performed prior to pathologic diagnosis and all patient information regarding specific tissues was withheld.

Image acquisition settings

CVG imaging system: Details of the CVG instrumentation have been previously described⁵⁴. Fluorescence images were acquired using the Cancer Vision Goggles (CVG) at a fixed working distance of 50 cm for data analysis. The dual-green laser alignment module was calibrated to converge at a working distance of 50 cm, corresponding to the standardized excitation geometry used for all quantitative fluorescence measurements. This design ensures that fluorescence intensity is measured under consistent excitation conditions, independent of operator experience. The NIR camera exposure time was 33 ms. Images were captured at three resolutions (height × width): 480 × 640, 720 × 1280, and 1080 × 1920 pixels. ISO settings were 800 for the NIR camera and auto for the visible camera. White balance gain (WB_gain) was set to [1, 1] for the NIR camera and [1, 6] for the visible camera. Analog digital gain (AD_gain) was set to [1, 6] for the NIR camera and [1, 1] for the visible camera. Composite pseudocolor fluorescence overlays were generated using the CVG user interface software, which provides four thresholding methods for fluorescence signal delineation: “Simple” uses a fixed default value of 128; “MEAN2” automatically calculates the threshold as the mean intensity of the grayscale images plus two standard deviations; “Yen” threshold is derived using Yen’s algorithm⁵⁵ followed by subsequent image processing, whereas the “Yen Simple” applies the “Simple” thresholding approach but replaces the fixed threshold with one calculated by the Yen algorithm.

IVIS SpectrumCT: NIR fluorescence images were acquired using an excitation filter at 745 nm and an emission filter at 820 nm. The imaging field of view (FOV) was 13.2 × 13.2 cm². Exposure time was set to 1 s, with medium binning (8) and f/2 aperture. Images were processed using Living Image Software v4.8.0 (64-bit). The system generated grayscale visible images (960 × 960, 16-bit) and fluorescence images (240 × 240, 16-bit). Composite pseudocolor fluorescence overlays were automatically produced by the software.

Pearl Imaging System: Fluorescence images were captured using the 800 nm channel (785 nm laser excitation, 820 nm emission) with a spatial resolution of 85 µm. The system employs a thermoelectrically cooled, low-noise CCD camera and uses a high dynamic range acquisition method with floating point pixel values. Raw visible and fluorescence images were exported in half-floating-point format at 964 × 1300 (height × width) pixel resolution. Image analysis was performed using ImageStudio Software v5.10.

Quest Spectrum: Images were captured with both visible and indocyanine green (ICG) fluorescence light levels set to 100%. Color gain was set to 6.5 dB, and ICG gain to 34 dB. Exposure settings included 1000 µs for color and 38 ms for ICG. The system operated in “Strong” overlay mode and “High” sensitivity mode. All images were acquired at 720 × 1080 (height × width) pixel resolution. Each acquisition generated four image formats: RGB, standard color image; SLD, green-fluorescence overlay on RGB; GRD, fluorescence heatmap (blue or red) on grayscale; and ICG, grayscale fluorescence image. Images were exported in JPEG format with identical 8-bit intensity values across RGB channels. Raw data export was not available; therefore, the red channel was extracted and used as grayscale input for fluorescence quantification.

Stryker SPY-PHI: Ex vivo human tumor images were acquired in the operating room using the following SPY modes: SPY-CSF: fluorescence overlaid in color scale on grayscale visible image; SPY overlay: fluorescence overlaid in green on white-light image; SPY contrast: fluorescence in grayscale; and SPY Color Maps: enhanced pseudocolor visualization under SPY-QP mode. Images were acquired at 1080 × 1920 (height × width) resolution and exported in JPEG format with uniform 8-bit values across all RGB channels (excluding software display icons). As raw image data could not be exported, the red channel was extracted and used as the grayscale representation of NIR fluorescence for quantitative analysis.

Image-processing pipeline and post hoc adjustments

For cross-platform spatial analyses, non-CVG images were geometrically aligned to CVG orientation using predefined steps (cropping, resizing, and affine transformation) implemented in MATLAB 2023a. These operations were applied to harmonize the field-of-view geometry and did not alter the underlying fluorescence ranking within each image. ROI selection was predefined and applied consistently for tumor and nontumor regions. Quantitative analyses (NIM, TNR, Dice) were then computed on the transformed grayscale fluorescence data using fixed formulas and threshold definitions specified in this section and figure legends.

Image acquisition optimization for cross-platform comparison

Acquisition settings were prospectively defined for each platform and are summarized in Supplementary Table 1. For CVG experiments, working distance, exposure settings, and output format were fixed according to a predefined protocol to support reproducible quantitative analysis. For commercial systems, acquisition was performed using manufacturer-supported clinical/preclinical modes relevant to each use case. Because many commercial platforms do not export raw radiometric data and apply proprietary processing pipelines, direct comparison of absolute intensity values across devices is limited. Therefore, cross-platform quantitative comparisons were based on normalized metrics (NIM, TNR) and spatial overlap metrics (Dice), as detailed below.

Quantitative TNR analysis

Quantitative TNR analysis was performed to assess fluorescence contrast between tumor and adjacent nontumor regions. Regions of interest (ROIs) for tumor and nontumor tissue were manually selected for each image using ImageJ and MATLAB R2023a. For working distance-dependent imaging experiments, ROI locations were adjusted to account for spatial shifts as imaging working distance varied, while maintaining comparable pixel areas across all working distances to ensure consistency.

TNR values were calculated as:

$$\mathit{TNR}=\frac{\mathit{mean}\mathit{intensity}\mathit{of}\mathit{tumor}\mathit{ROI}}{\mathit{mean}\mathit{intensity}\mathit{of}\mathit{nontumor}\mathit{ROI}}$$ 1

Dice coefficient analysis

Although each imaging system was individually calibrated for visible and NIR imaging, direct comparison across platforms was challenging due to inherent differences in pixel value representations, resolution, field of view, and object orientation. As a result, original visible and NIR images from the IVIS, Pearl, Quest, and Stryker SPY-PHI systems required preprocessing before quantitative comparison with CVG data. To enable pixel-wise analysis, images from each comparator system were cropped, resized, rescaled, and spatially transformed to align with the CVG image reference frame. Image registration and preprocessing were performed using MATLAB R2023a. NIMs were generated for each image by dividing the fluorescence intensity at each pixel by the mean intensity of the lowest percentage of pixels within the ROI, defined manually on visible-light images:

$${\mathit{NIM}}_{(\mathit{low}\%)}=\frac{\mathit{NIR}\mathit{fluorescent}\mathit{intensity}\mathit{in}\mathit{all}\mathit{ROI}\mathit{pixels}}{\mathit{mean}\mathit{intensity}\mathit{of}\mathit{the}\mathit{lowest}\%\mathit{pixels}\mathit{in}\mathit{ROI}}$$ 2

Dice coefficient analysis was then performed in MATLAB R2023a using the built-in dice function, which calculates the Sørensen–Dice coefficient between binary images:

$$\mathit{Dice}\mathit{coefficient}=\frac{2\left|{\mathit{BW}}_1\bigcap {\mathit{BW}}_2\right|}{\left|{\mathit{BW}}_1\right|+\left|{\mathit{BW}}_2\right|}$$ 3

where $B{W}_1$ and $B$W₂ represent the binary images of high-intensity pixels from CVG and the reference system, respectively. Dice values were computed for specific intensity thresholds (e.g., top 10%, 15%, or NIM > 1.5) to evaluate spatial overlap in tumor fluorescence signal across systems. The Dice coefficient analysis was conducted on mouse tumor slices (n = 5) and human tumor tissue (n = 6). The human tumor tissue selected for this analysis consisted of samples with saturated intensity pixels not exceeding 15%, which exhibited similar imaging characteristics between the CVG and Stryker SPY-PHI systems.

Quantitative analysis framework

Imaging performance was evaluated using three quantitative endpoints: (i) TNR for fluorescence contrast, (ii) normalized intensity maps (NIMs) for cross-platform intensity normalization when raw radiometric outputs were unavailable or non-comparable due to proprietary image processing, and (iii) Dice coefficients for spatial overlap of tumor-associated fluorescence regions. For NIM computation, the tissue/specimen ROI was defined on the corresponding visible image and applied to the NIR image. Pixel intensities were normalized by dividing each pixel by the mean intensity of the lowest 10% of pixels within that ROI; analyses using percentile ranges from 10% to 50% are provided in Supplementary Fig. 1. Dice coefficients were computed for both percentile-based high-intensity masks (top 10–50% pixels) and threshold-based masks (NIM > 1.5), as specified in various figures and legends.

Statistical analysis

Final image processing and quantitative analyses were performed using MATLAB R2023a. Data are presented as mean ± standard deviation unless otherwise noted. Group comparisons were conducted using either a paired t-test in MATLAB or a one-way analysis of variance (ANOVA) followed by multiple comparisons, based on group mean effects. ANOVA-based statistical analyses were performed using Prism version 8.0.1 (GraphPad Software, USA). A P value < 0.05 was considered statistically significant.

Author contributions

S.A. conceived wearable CVG; H.Z., X.X., C.T., I.Z., W.J.A., and S.A. designed the research; H.Z., X.X., C.T., and I.Z. performed device development research; H.Z., C.H., N.B., performed cell and animal imaging research; H.Z. performed CVG clinical imaging; B.S. performed human surgery; S.V. and N.M.R. performed non-CVG clinical imaging; K.S.G. and J.G. prepared NIRF agents; H.Z., X.X., W.J.A, S.A. analyzed data; W.A., X.X., B.S., J.G., and S.A. supervised research; H.Z., X.X., W.J.A., and S.A. wrote the manuscript; all authors reviewed or edited the manuscript.

Data availability

All data supporting the findings of this study are available within the paper and its Supplementary Information.

Competing interests

The Authors declare no Competing Non-Financial Interests but the following Competing Financial Interests: SA is the primary inventor of CVG and LS301, which are used in this study. Other authors declare no competing financial interests. B.D.S. and J.G. are inventors of Pegsitacianine and scientific co-founders of OncoNano Medicine, Inc.

Supplementary information

The online version contains supplementary material available at 10.1038/s44303-026-00170-x.