Diagnostic accuracy and surgical outcomes of fluorescence-guided surgery in breast cancer: a systematic review and meta-analysis.

Introduction
Precise intraoperative margin assessment is vital for the success of breast-conserving surgery (BCS) and mastectomy [1], as margin status is a key determinant of local recurrence and reoperation rates [2, 3]. However, conventional techniques have inherent limitations. Palpation and visual inspection often fail to detect subclinical or therapy-altered disease, particularly in dense breast tissue [4]. Similarly, specimen radiography provides only two-dimensional data and is prone to errors caused by tissue deformation and clip migration [5]. Furthermore, it is ineffective for non-calcified tumors and ductal carcinoma in situ (DCIS). These challenges contribute to persistently high positive-margin rates, underscoring the need for real-time, direct detection of residual tumor.
Fluorescence-guided surgery (FGS) offers a compelling solution to these limitations. FGS utilizes fluorescent probes that accumulate in tumors via metabolic trapping, the enhanced permeability and retention (EPR) effect, or specific receptor binding [6–8]. Intraoperatively, an excitation light illuminates the surgical field, and specialized near-infrared (NIR) cameras capture the emitted fluorescence. This generates a real-time overlay quantifying the tumor-to-background ratio, thereby visualizing residual disease directly on the cavity wall or specimen. Clinically available probes include non-targeted agents, such as indocyanine green (ICG), 5-aminolevulinic acid (5-ALA), and methylene blue (MB) [9–11], as well as targeted agents like bevacizumab-IRDye800CW (Beva800CW) and pegulicianine [12, 13]. By providing wide-field, calcification-independent visualization, FGS effectively addresses the blind spots of conventional methods.
Despite the theoretical promise of FGS, existing evidence presents conflicting results [14]. This discrepancy likely stems from heterogeneity in fluorescent agents, tumor subtypes, and surgical techniques [15]. Furthermore, prior research has primarily focused on diagnostic capabilities, leaving critical surgical outcomes (specifically positive margin and reoperation rates) under-explored. Currently, no meta-analysis comprehensively synthesizes both the diagnostic accuracy and surgical impact of FGS. This evidence gap hinders the formulation of definitive clinical guidelines.
Therefore, the primary objective of this study was to quantify the diagnostic accuracy of FGS and evaluate its effect on key surgical outcomes, including positive margin and reoperation rates. By providing high-quality evidence, we aim to inform surgical decision-making, potentially reducing reoperation risks and improving long-term patient outcomes.

Method

Study protocol
This systematic review and meta-analysis adhered to the PRISMA 2020 guidelines [16], with the objective of assessing the diagnostic accuracy and surgical outcomes of Fluorescence-Guided Surgery (FGS) in the breast cancer. The study protocol was prospectively registered with PROSPERO (registration number: CRD420251165928), Available from https://www.crd.york.ac.uk/PROSPERO/view/CRD420251165928.

Search strategy
A systematic search of the literature was performed across four electronic databases: PubMed, Embase, Web of Science, and the Cochrane Library. The search covered publications from inception up to the final search date of October 3, 2025. Full search strategies for each database are provided in Supplementary Table 1. Only studies published in English were considered. Titles, abstracts, and full texts were independently assessed by two reviewers (JML and KBG) based on predefined eligibility criteria. Discrepancies were resolved through discussion or, when needed, by consulting a third reviewer (YQF).

Study selection
In this systematic review and meta-analysis, studies were selected based on the following PICOS framework:Population: Adult patients with breast cancer, focusing on FGS, fluorescence detection of excised in vitro specimen margins, and post-operative fluorescence assessment of surgical margins. Excluded were studies involving non-human subjects, pediatric populations, or benign lesions.

Intervention: FGS as the primary technique, utilizing agents like ICG. Excluded were studies where fluorescence was adjunctive to other treatments or not central to the intervention.

Comparator: Studies comparing FGS with conventional method, including before-and-after comparisons where patients initially underwent standard resection and later received fluorescence-guided detection for suspected residual lesions. Historical or internal controls were considered.

Outcomes: The primary outcome was the intraoperative diagnostic classification of surgical margins by FGS. Data were stratified into a 2 × 2 contingency table comprising true positives (TP), false positives (FP), true negatives (TN), and false negatives (FN), using final histopathology as the reference standard. Secondary outcomes included downstream surgical metrics: the final positive margin rate and reoperation rate.

Study Design: Randomized controlled trials (RCTs), prospective and retrospective cohort studies. Excluded were case reports, comments, and low-quality studies.

Data extraction
Two reviewers independently abstracted detailed data from each study: first author, publication year, sample size, study design, clinical subtype. Particular emphasis was placed on FGS, encompassed fluorophores used, detection platforms/instruments, imaging modalities, analytical units, margin assessment methods, surgical resection techniques, diagnostic accuracy metrics, and surgical outcomes.

Quality assessment
The quality assessment of the studies included in this review was conducted using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool [17]. This instrument provides a comprehensive analysis of potential biases and applicability concerns within four critical domains: patient selection, index test, reference standard, and flow & timing.

Statistical analysis
Quantitative synthesis was stratified by data availability. Studies reporting sufficient data to construct 2 × 2 contingency tables (TP, FP, TN, FN) were included in the primary diagnostic meta-analysis. Studies lacking 2 × 2 data but reporting surgical outcomes were included in the respective secondary analyses. Those lacking quantitative data were restricted to qualitative review.
Diagnostic accuracy was assessed using a bivariate random-effects model to account for the correlation between sensitivity and specificity. Summary receiver operating characteristic (SROC) curves were plotted with 95% confidence and prediction regions. Heterogeneity across studies was quantified using the I2 statistic and Cochran’s Q test; values of I2 ≥ 50% or P < 0.05 were deemed indicative of significant variability. Sources of heterogeneity were explored via subgroup analyses and meta-regression. Multicollinearity was ruled out using a Variance Inflation Factor (VIF) cutoff of < 10; covariates included fluorescence type, imaging device, margin definition, and other study characteristics. Sensitivity analyses were conducted using a “leave-one-out” approach. Additionally, due to the disproportionate sample sizes of two studies (sample sizes: 1584 and 2346) [13, 18] compared to others (sample sizes: 12–172) [9, 11, 14, 19–31], a sensitivity analysis excluding these high-leverage studies was performed. Publication bias was evaluated using Deek’s funnel plot asymmetry test, with P < 0.10 indicating significant small-study effects. In adherence to Cochrane guidelines, this assessment was restricted to analyses including at least 10 studies, as tests for asymmetry lack statistical power and reliability in smaller datasets [32].
The impact of FGS on surgical outcomes (positive margin and reoperation rates) was synthesized as the Mean Difference (MD) in rates between pre- and post-implementation phases. This metric captures the absolute risk reduction within paired cohorts.
Analyses were performed using Stata 18.0 (midas, metandi modules) and R 4.4.2 (mada, metafor packages).

Result

Study identification and characteristics
A total of 979 records were identified from Cochrane (n = 27), Embase (n = 228), PubMed (n = 641), and Web of Science (n = 83). Titles/abstracts were screened and full texts assessed against prespecified criteria, yielding 18 studies [9, 11, 13, 14, 18–31], encompassing 1283 patients for inclusion in the systematic review and meta-analysis. The study selection process is depicted in the PRISMA flow diagram (Fig. 1).

Table 1 presents the key characteristics of study participants and basic study design features. The included studies comprised a variety of designs: 2 RCTs [23, 25], 10 prospective observational studies [11, 13, 14, 18, 19, 24, 27, 29–31], and 6 retrospective studies [9, 20–22, 26, 28]. Most investigations were conducted between 2016 and 2025 across diverse geographical regions, including the United States, China, South Korea, the Netherlands, the United Kingdom, France, and Canada. Most studies included patients with invasive ductal carcinoma (IDC), often with or without DCIS. Nearly all participants underwent BCS, although a subset of studies also included mastectomy cases.
The primary intervention across studies was FGS, utilizing various imaging agents, with ICG being the most frequently used [9, 14, 19, 20, 23, 24, 26, 29–31]. Other fluorophores included pegulicianine [13, 18], Beva800CW [21, 22, 27], MB [11], and 5-ALA [25]. 10 studies employed in vivo imaging within the surgical cavity [13, 18–20, 22, 23, 25, 26, 28, 31], 6 studies utilized ex vivo imaging on excised specimens [9, 11, 14, 24, 29, 30], and 2 studies combined both in vivo and ex vivo imaging modalities [21, 27]. A range of fluorescence imaging systems such as pegulicianine FGS (pFGS), Fluobeam 800, and custom-built or commercial fluorescence cameras. Most studies focused on fluorescence analysis at the patient or margin level to assess residual tumor detection and margin status. Despite the heterogeneity in imaging modalities, fluorophores, analytical methods, and other related factors, all studies shared the common objective of enhancing intraoperative visualization and margin assessment during breast cancer surgery.

Diagnostic performance of FGS for marginal evaluation
Of the 18 studies included in this systematic review, 11 provided sufficient data in 2 × 2 contingency tables (TP, FP, TN, FN) to permit a meta-analysis of diagnostic test accuracy (DTA) [9, 11, 13, 14, 18, 19, 22, 25, 26, 29, 30]. One study (Kedrzycki et al. [14]) was stratified into two cohorts based on the preoperative timing of fluorescent agent administration, whereas two others (Yu et al. [30] and Ottolino-Perry et al. [25]) were subdivided into two cohorts according to the administered dose. Pooled estimates revealed a summary sensitivity of 0.72 (95% CI: 0.62–0.81) and specificity of 0.75 (95% CI: 0.67–0.81) (Fig. 2a). The SROC curve yielded an area under the curve (AUC) of 0.80 (95% CI: 0.76–0.83) (Fig. 2c).

Subgroup analysis and meta-regression
To investigate the potential sources of heterogeneity in the diagnostic performance of FGS for breast cancer, the following covariates were considered in the meta-regression: (1) fluorescence type (targeted fluorescent probes vs. non-targeted fluorescent probes); (2) imaging device (handheld fluorescence imaging systems vs. intraoperative near-infrared imaging systems); (3) imaging modality (ex vivo imaging vs. in vivo cavity imaging); (4) margin definition [Ink on tumor vs. Strict margin (defined as tumor cells present within < 1 mm or < 2 mm of the inked surface)]; (5) surgery type (BCS vs. BCS and mastectomy); (6) tumor type (presence of DCIS component vs. without DCIS component); (7) unit of analysis (margin-level vs. patient-level); (8) centre characteristics (single); (9) sample size (small). (Table 2)
Performance estimates for each covariate and the corresponding meta-regression outputs are summarized in Supplementary Table 2. All covariates showed acceptable collinearity (VIFs < 10), indicating no substantial multicollinearity; accordingly, none were excluded. These covariates were carried forward to the subgroup analyses, from which stratum-specific pooled sensitivity, specificity, and 95% CIs were derived.
When sensitivity was evaluated alone, non-targeted fluorophores outperformed targeted fluorescent dyes (0.83 vs. 0.65). Ex vivo imaging surpassed in vivo cavity imaging (0.81 vs. 0.68). Cohorts excluding DCIS exhibited higher sensitivity than those including DCIS (0.82 vs. 0.65). Patient-level analyses yielded greater sensitivity than margin-level analyses (0.86 vs. 0.71). NIR imaging systems outperformed handheld fluorescence systems (0.86 vs. 0.74). Sensitivity did not differ significantly between cohorts limited to BCS and cohorts including both BCS and mastectomy (0.70 vs. 0.67). Similarly, sensitivity estimates were comparable across margin definition subgroups (Strict margin: 0.79 vs. Ink on tumor: 0.72); however, this stratification notably resolved statistical heterogeneity for sensitivity within the “Strict margin” group (I2 = 0%). Regarding centre characteristics and sample size, model non-convergence precluded comparative analyses; consequently, pooled sensitivity estimates were generated restrictedly for single-center studies (0.79) and small-sample studies (0.76). Conversely, while specificity differences were minimal across most other subgroup factors (Δspecificity ≤ 0.1), margin definition impacted specificity more substantially (Strict margin: 0.87 vs. Ink on tumor: 0.72; Table 2).

Publication bias and sensitivity analysis
Including 11 studies, Deek’s test indicated funnel-plot asymmetry (P < 0.01) (Supplementary Fig. 1a). After excluding the two very large studies (sample sizes 1584 and 2346), the evidence for asymmetry attenuated and was no longer statistically significant (P = 0.16) (Supplementary Fig. 1b). For transparency, pooled sensitivity and specificity estimates excluding these two studies are provided in Supplementary Fig. 2. Extending this assessment to specific subgroups with sufficient data points, Deek’s test indicated no significant funnel plot asymmetry among small-sample studies (P = 0.20; Supplementary Fig. 3) or single-center studies (P = 0.20; Supplementary Fig. 4).
In the bivariate boxplot of logit (sensitivity) versus logit (specificity), one study (Barbara L et al.) fell outside the 95% outer ellipse (Fig. 2b), indicating a potential outlier. Consequently, a leave-one-out sensitivity analysis excluding this study was performed to quantify its influence on the pooled summary estimates. The corresponding alternative estimates are presented in Supplementary Table 3.

Quality assessment
Methodological quality of the 18 included studies was assessed using QUADAS-2. Risk of bias was as follows: patient selection: 2 high, 4 unclear, and 12 low; index test: 2 high, 3 unclear, and 13 low; reference standard and flow and timing: low in all 18 studies. Applicability concerns were generally low: in the patient selection domain, 2 studies showed high concern and 2 were unclear; in the index test domain, 2 studies showed high concern and 1 was unclear. (Supplementary Fig. 5)

Surgical outcomes of FGS for marginal evaluation
The pooled positive-margin rate with FGS was 14% (95% CI: 0.08–0.21; I2 = 53.5%) (Fig. 3a); the reoperation rate was 10% (95% CI: 0.05–0.16; I2 = 72.2%) (Fig. 3b); and the reoperation-avoidance rate was 16% (95% CI: 0.09–0.23; I2 = 47.4%) (Fig. 3c). Subgroup analyses were performed by fluorophore characteristics. With Beva800CW, the proportion avoiding a second surgery was 50% (4/8), whereas the pooled effect for pegulicianine was 15% (95% CI: 0.08–0.22; I2 = 0%).

Leave-one-out sensitivity analysis showed that heterogeneity in the fluorescence-guided positive margin rate decreased to 0 after exclusion of Kedrzycki et al. [14]. Heterogeneity in the fluorescence-guided reoperation rate was not reduced by the leave-one-out analysis. Excluding Koberg et al. [27] reduced heterogeneity for the fluorescence-guided reoperation avoidance rate to 0 (Supplementary Fig. 6). No significant publication bias was detected (Supplementary Fig. 7).

Discussion
FGS in breast cancer primarily targets two clinical areas: axillary staging and primary tumor excision. The utility of fluorescence (particularly ICG) for sentinel lymph node biopsy (SLNB) is well-established [33, 34]. Conversely, the application of FGS for intraoperative margin assessment represents a separate clinical objective. Our systematic review specifically addresses this latter domain. We focus on the diagnostic accuracy of fluorescent dyes in assessing both specimen margins and residual tumor at the surgical bed, as well as their associated surgical outcomes.
This meta-analysis demonstrates that FGS achieves moderate diagnostic performance for intraoperative margin assessment in breast cancer, with a pooled sensitivity of 0.72, specificity of 0.75, and SROC AUC of 0.80. Subgroup patterns indicate higher sensitivity with non‑targeted fluorophores, ex vivo imaging, patient‑level analyses, cohorts excluding DCIS, and NIR imaging systems. Regarding specificity, estimates remained largely consistent across most strata (Δspecificity ≤ 0.1), with the notable exception of margin definition. Clinically, FGS was associated with lower positive-margin rates (14%), reduced reoperation rates (10%), and higher reoperation‑avoidance (16%). Heterogeneity was substantial (I2 = 66% for sensitivity; 94% for specificity) under a bivariate random‑effects model.
The observed accuracy likely reflects complementary biological and technical mechanisms. Real time fluorescence enhances detection of subclinical foci at the resection surface and highlights lymphatic channels and nodal basins, thereby improving sensitivity for occult disease otherwise missed by visual inspection or palpation [35]. Non‑targeted agents such as ICG bind serum albumin, facilitating lymphatic trafficking and accumulation within regions of increased vascular permeability; this can boost sensitivity, even exceeding targeted probes when the latter are constrained by heterogeneous receptor expression, suboptimal affinity, or limited tissue penetration at the margin [36, 37]. The modest gain in specificity may arise when targeted probes enrich signal in cancerous tissue and when albumin-bound ICG exhibits preferential retention within tumor microvasculature [11, 14]; yet off-target uptake and background scatter limit large specificity improvement [38, 39].
FN plausibly reflect: (1) biological factors: low perfusion or duct-confined growth patterns characteristic of DCIS, may limit the effective delivery of probes [40, 41]; (2) optical constraints: limited NIR penetration and hemoglobin absorption that attenuate signals from deep or bleeding surfaces; (3) workflow variables: imaging before optimal wash‑in or after signal quenching at high local concentrations; and (4) analytic thresholds that trade sensitivity for specificity [11, 25]. FP may stem from non‑specific EPR-like accumulation in inflamed or hypervascular benign tissue, macrophage-rich granulation, or fat necrosis [42]; targeted agents can also bind non-malignant epithelia with low‑level antigen expression [8].
Subgroup patterns provide hypothesis. One possible explanation for the superior sensitivity observed with ex vivo imaging could be the controlled lighting conditions, minimal motion, and the opportunity to closely examine the specimen surface [29]. In contrast, in vivo cavity imaging may be more prone to challenges such as the geometry of the cavity, which could potentially obscure portions of the margin [13]. The advantage of NIR platforms over handheld systems is consistent with real time full-field imaging with a wide field of view, higher signal-to-noise ratios, and better rejection of ambient light [43]. Greater sensitivity at the patient level than at the margin level suggests unit-of-analysis effects: multiple correlated margins within a patient raise the probability that at least one true-positive focus will be detected, whereas per-margin analyses dilute per-patient performance. The similar sensitivity across surgery types (BCS vs. BCS and mastectomy) indicates that FGS benefits likely generalize beyond lumpectomy alone; however, sensitivity declines when DCIS is present, underscoring biological limits of dye delivery to non-invasive, duct-limited disease [41].
Substantial heterogeneity was observed, with the most pronounced variability in specificity (I2 = 94%). Potential sources of this heterogeneity include differences in probe chemistry (ICG vs. MB vs. targeted agents), variations in dosing and injection-to-imaging intervals, discrepancies in device sensitivity and spectral filtering, patient factors (such as molecular subtype, breast density, and prior neoadjuvant therapy), and inconsistent definitions of “positive” margin across studies [9, 11, 22]. Among these potential confounders, our analysis isolated inconsistent margin definitions as a primary driver of variability, warranting deeper examination. Stratifying by margin definition resolved sensitivity heterogeneity within the “Strict margin” group (I2 = 0%), underscoring definition consistency, though residual specificity heterogeneity (I2 = 43%) implies influences from surgical variability or dye diffusion. The superior concordance in the “Strict margin” group (Sensitivity: 0.79, Specificity: 0.87) versus the “Ink on Tumor” group (Sensitivity: 0.72, Specificity: 0.72) likely reflects the inherent physics of fluorescence imaging, due to factors such as depth penetration and blooming effects that allow for the detection of subsurface (< 1 mm) disease [35]. Such signals are clinically relevant but technically classified as false positives under strict 0 mm “Ink on Tumor” criteria, artificially depressing specificity. Conversely, wider “Strict margin” definitions align naturally with the modality’s capability to identify risk-associated close margins. Therefore, surgeons operating under standard “No Ink on Tumor” guidelines must interpret residual fluorescence cautiously, acknowledging it may signal sub-millimeter disease exceeding minimal pathological requirements yet warranting clinical attention.
Consistent with reports from other oncologic subsites, FGS improves intraoperative localization and can reduce positive margins and reoperations, for example with 5-ALA or targeted agents in head and neck cancer [44] and high-grade glioma [45]. Prior breast surgery studies have emphasized sentinel lymph node mapping and flap perfusion rather than systematic evaluation of margin-level diagnostic accuracy [34]. To the best of our knowledge, this is the first meta‑analysis to systematically evaluate the diagnostic performance of FGS for breast cancer resection margins and to relate these findings to surgical outcomes. It advances the field by (1) providing a formal diagnostic‑accuracy meta‑analysis with AUC estimation; (2) stratifying performance by probe type, imaging context (ex vivo vs. in vivo), device class, tumor histology (DCIS component present vs. absent), and analytic unit; and (3) linking diagnostic performance to positive-margin and reoperation rates.
While our findings suggest that FGS may favorably impact surgical outcomes, these results must be interpreted with caution given the limited number of studies contributing data on reoperation rates. If confirmed by larger datasets, the observed reductions in positive margins and reoperations would be highly clinically meaningful. By lowering these rates, FGS has the potential to decrease delays to adjuvant therapy, reduce anesthesia exposures, improve cosmesis by preserving healthy parenchyma, and alleviate psychological burden from unplanned second procedures.
Clinically, FGS enhances surgical decision-making by providing real-time visual feedback, enabling immediate margin correction. However, widespread implementation requires significant investment in imaging systems and specialized training to overcome the learning curve. While the initial capital outlay is substantial, the potential reduction in costly reoperations suggests long-term cost-effectiveness, though robust economic data remains limited. Future research must prioritize large-scale randomized trials focusing on standardized operative protocols, the development of tumor-specific probes, and comprehensive cost-benefit analyses to validate FGS as a standard-of-care modality.
This systematic review and meta-analysis has several noteworthy limitations that merit consideration. First and foremost, the inherent quality and design of the included literature impose constraints on the strength of our conclusions. The majority of eligible studies were observational cohorts, with a scarcity of RCTs. Risk-of-bias assessments revealed frequent methodological shortcomings, particularly the lack of blinding for surgeons or image assessors, which introduces potential performance and detection biases. Consequently, findings regarding diagnostic accuracy should be interpreted with caution. Second, substantial heterogeneity was observed across studies (I2 up to 94% for specificity), likely stemming from residual confounding factors such as variations in probe dosing, imaging timing, device sensitivity, and reader thresholds. This heterogeneity was compounded by inconsistent clinical definitions; for instance, margin assessment criteria varied significantly (“Ink on tumor” vs. “Strict margin”), as did the unit of analysis (per-margin vs. per-patient). Furthermore, the unavailability of individual patient data (IPD) precluded harmonized adjustments for critical covariates, including tumor subtype, cellularity, neoadjuvant therapy status, breast density, and the extent of DCIS. Third, data reporting variability and sparsity hindered certain quantitative syntheses. Incomplete reporting necessitated the exclusion of specific studies from pooled analyses, such as one study lacking reconstructible diagnostic contingency tables, which may introduce selection bias. Moreover, statistical constraints limited our ability to perform robust subgroup analyses. Specifically, the limited number of primary studies in multicenter and large-sample subgroups resulted in model non-convergence within the bivariate meta-analysis framework. Similarly, extreme homogeneity in study design (predominantly observational, Supplementary Fig. 8) and administration routes (mostly intravenous, Supplementary Fig. 9) precluded comparative regression analyses for these factors. Regarding publication bias, while the initial Deek’s test suggested small-study effects, sensitivity analyses indicated this was driven by sample-size imbalance and high-leverage studies rather than selective non-publication. Although excluding these outliers removed the asymmetry (P = 0.16), we present these estimates conservatively. Finally, the assessment of clinical utility remains limited by the paucity of long-term oncologic outcomes, specifically regarding local recurrence and survival benefits. Additionally, our literature search was restricted to English-language publications; thus, the exclusion of relevant studies published in other languages cannot be ruled out, potentially introducing language bias to our findings.
Recognizing these shortcomings is the first step toward improvement; therefore, we recommend that subsequent work focus on (1) probe innovation: breast‑cancer‑specific targeted dyes (HER2, EGFR, FRα) with optimized pharmacokinetics and high tumor-to-background ratios [46, 47]; (2) standardized protocols: consensus on dose, injection-to-imaging interval, device calibration, ambient-light control, and quantitative thresholds (adhering to STARD/QUADAS‑2 guidance) [17, 48]; (3) rigorous trials: multicenter, adequately powered RCTs that embed cost‑effectiveness, patient‑reported outcomes, and long‑term local control; (4) analytic advances: quantitative fluorescence with ratiometric or dual-tracer normalization [49] and AI‑assisted margin mapping to reduce reader variability [50]; and (5) focused clinical scenarios: evaluation after neoadjuvant therapy (to gauge residual disease), guidance of cavity shaving strategies, and integration with axillary staging (sentinel node mapping and targeted axillary dissection).

Conclusions
FGS provides moderate diagnostic accuracy for margin assessment and shows promise for potentially reducing positive margins and reoperations in breast cancer surgery, although current evidence for these surgical outcomes is preliminary. Given its real time, surface weighted information and favorable safety profile, FGS should be considered as an adjunct to standard intraoperative assessment, particularly in BCS and centers equipped with NIR platforms. Standardization, next generation breast specific probes, and high quality multicenter trials that include durability and health economic endpoints are now needed to define precisely where FGS delivers the greatest value and to enable its reliable incorporation into routine practice.

Supplementary Information