2.1.

Ethics

This systematic review was conducted according to Preferred Reporting Items for Systematic Reviews and Meta-Analysis guidelines. The study was registered on PROSPERO (CRD42021286487). No ethical approval was required.

2.2.

Strategy for Identification and Selection of Studies

Embase, MEDLINE, Web of science, and Scopus were systematically searched for all articles published before December 2022. The search was conducted using the following Medical Subject Headings (MeSH): “fluorescence” AND “breast cancer” AND “surgery” AND “fluorescence imaging.” For different databases, the search terms were adjusted as required. Additional reports were identified using Google Scholar and the CLEARER database (of FGS used in cancers) through citation tracking. The full search strategy can be found in the Appendix.

Covidence systematic review software (Veritas Health Innovation, Melbourne, VIC, Australia)²⁶ was used for duplicate removal, title and abstract screening, full-text review, and data extraction.

2.3.

Eligibility Criteria

Studies were included in this review if: (1) a fluorescence camera system was used to assess breast cancer and/or surgical cavities; (2) contrast agents were utilized; and (3) the full text was available in the English language. Studies were excluded if: (1) the optical imaging system was for pre-operative cancer diagnosis; (2) spectroscopy (but not imaging) was used; (3) the studies included only benign breast tissue lesions, only sentinel lymph nodes, or non-breast cancers; or (4) the report was a review, case report, poster, abstract, project proposal, expert opinion, animal study, or cell line study.

2.4.

Study Selection and Data Synthesis

Data were screened and extracted by two authors independently, MK and HC. Disagreements were resolved with the senior authors (DRL and DE). Sociodemographic variables including sample size, age, and body mass index (BMI) were collected. With regards to cancer characteristics, genotype (e.g., invasive ductal carcinoma), immunophenotype (ER, PR, HER2), and use of pretreatment (i.e., neoadjuvant chemotherapy or hormonal therapy) were determined. Elements describing the imaging systems themselves and the contrast agents they were paired with (including dosage, route of administration, excitation, and emission wavelengths) were identified. Lastly, outcomes such as tumor to background ratio (TBR), positive margin assessments, diagnostic accuracy (including sensitivity, specificity, positive predicted value), re-excision rate, and any adverse events were recorded.

3.1.

Search and Selection of Articles

1182 articles were identified, of which, 372 studies were removed after de-duplication. This resulted in 810 studies undergoing title and abstract screening, of which, 692 studies failed to meet the inclusion criteria. An additional 40 studies were identified through bibliographic cross-referencing, and out of the 157 reports that were assessed in detail for eligibility, only 21 studies met all criteria for inclusion in the review.

3.2.

Patient Demographics

Overall, there were 12 prospective clinical trials.^{14,19–23,25,27–32} The nine remaining studies were either case series, feasibility trials, or cohort studies.^24,33–39 Overall, these studies encompassed 894 patients receiving optical imaging in conjunction with contrast agents in breast cancer tissue assessment. Table 1 summarizes patient demographics and cancer subtypes studied.

Table 1

Summary of study type, patient demographics, and cancer clinicopathological data.

Four studies described ethnicity,^14,24,27,28 but only one reported on menopausal status.²³ Five studies included patients who underwent neoadjuvant chemotherapy (NACT) prior to BCS or mastectomy. Unkart et al.²⁵ included five pretreated patients out of 27, Veys et al.³⁷ encompassed eight pretreated patients, and Kedrzycki et al.¹⁴ and Leiloglou et al.²⁸ reported 2 out of 40 patients that had received NACT. However, only Zhang et al.³² investigated the impact of patients with NACT compared to those with primary surgery using a custom-built camera system. They reported a significant difference ($p<0.05$) in fluorescence detection rate and strength of signal, whereby only 30% of NACT cases were detected with a TBR of 1.63 in contrast to 80% of primary cases with a TBR of 1.94.³²

3.3.

Imaging Systems

Table 2 summarizes the imaging systems and their diagnostic accuracy. A total of 11 different imaging systems were reported. Studies using Food and Drug Administration approved camera systems permitted for purposes other than breast cancer included: two studies which exploited the Photodynamic Eye™ (PDE) camera system (Hamamatsu Photonics, Shizuoka, Japan),^24,30 two employed Fluobeam 800™ imaging system (Fluoptics, Grenoble, France),^37,41 two utilized the Artemis™ fluorescence imaging system (Quest Medical Imaging, Middenmeer, The Netherlands),^27,38 two capitalized on the mini-FLARE^TM (Beth Israel Deaconess Medical Center, Boston, Massachusetts),^31,36 and one study used the Visual Navigator^TM camera system (SH System, Gwangju, South Korea).³⁴

Table 2

Comparison of diagnostic accuracy between different fluorescence imaging systems.

The remaining studies included two that deployed the LUM fluorescence imaging system (Lumicell, Inc., Newton, Massachusetts),^23,24 one that used the synchronized infrared imaging system (SIRIS) (Teal Light Surgical, Inc., Seattle, Washington),¹⁹ two that capitalized on the EagleRay-V3 (Technical University of Munich, Munich, Germany),^20,40 and finally one that utilized portable real-time optical detection identification and guide for intervention (PRODIGI) handheld fluorescence imaging system (SBI-ALApharma Canada Inc., Toronto, Canada).²² Two studies used an unspecified camera system (system by SurgVision, Harde, The Netherlands),²¹ and the remaining studies^14,25,28,32 that developed in-house camera systems did not include specified model or company name.

3.4.

Contrast Agents and Tumor-to-Background Ratio

There were eleven studies which used the nonspecific passive fluorophores indocyanine green (ICG) and methylene blue (MB).^{14,27,28,30–35,37} These fluorophores take advantage of the enhanced permeability and retention (EPR) effect, whereby they leak into the tumor via porous vasculature and remain there due to impaired lymphatic outflow [Fig. 3(a)].^{14,27,28,30–35,37} Of the nine studies deploying ICG, three used the same custom built camera system^14,28,33 and six used commercially accepted imaging systems.^{27,30,34,35,37} Six studies encompassing 105 patients, administered $0.25\mathrm{mg}/\mathrm{kg}$, $5\mathrm{mg}/\mathrm{kg}$, or 12.5 mg intravenous ICG.^{14,27,28,33,37} In these studies, TBR varied from 1.72 to 3.46, irrespective of the disparity in ICG doses. Three studies administered 10 or 25 mg ICG intralesionally.^30,34,35 Of the two studies that capitalized on MB,^31,32 one study employed the Mini-FLARE³¹ and one utilized a custom-built camera system.³² Both studies administrated MB intravenously and reported a similar range of TBR ($1.94\pm 0.71$ and $2.40\pm 0.80$).^31,32

Fig. 3

Mechanism of action for targeting tumors using fluorophores. (a) Passive targeting through the EPR effect whereby the fluorophore leaks into tissue due to the porous vasculature and has impaired lymphatic outflow. Panels (b) and (c) illustrate active targeting. (b) The targeting of specific receptors overexpressed in tumor. (c) The targeting of specific enzymes present in the tumor microenvironment by requiring the probe activation by that enzyme.

Only Kedrzycki et al.¹⁴ and Leiloglou et al.²⁸ assessed the effects of two different timings of administration, one immediately prior to resection and the other at the start of the operation. Kedrzycki et al. observed that ICG administration immediately prior to resection had a statistically significant higher TBR in both ex-vivo and histopathology cut-up than the start of the operation ($2.10\pm 0.6$ and $3.18\pm 1.74$ versus $1.72\pm 0.31$ and $2.10\pm 0.92$ relatively).¹⁴ However, the differences in sensitivity and specificity were not statistically significant except for certain cases where texture metrics were applied.

One study evaluated the fluorescence of protoporphyrin IX, a metabolite of ALA, whose accumulation is caused by metabolic disruption in the heme formation pathway in breast cancer cells.²² The study by performed by Ottolino-Perry et al.²² was a phase 1 safety and comparative study of oral 15 and $30\mathrm{mg}/\mathrm{kg}$ ALA utilizing the PRODIGI custom-built camera.²² Although they did not report a TBR, they reported a statistically significant difference between patients who had received ALA and control patients ($p<0.05$).²²

There were five studies that targeted specific receptors including bevacizumab–IRDye800CW (vascular endothelial growth factor), EC17 (folate), and tozulesteride (chloride channels), all of which were administered intravenously [Fig. 3(b)]. Three studies administered 4.5, 10, 25, and 50 mg bevacizumab–IRDye800CW,^20,21,40 two exploiting the Eagle-Ray custom-built camera system^20,40 and one study utilizing a system developed by SurgVision.²¹ Among these studies, only Koch et al.²⁰ reported a TBR, ranging from 1.8 to 9.0. Only Lamberts et al.⁴⁰ compared the concentration of bevacizumab–IRDye800CW to VEGF-A levels in tumor versus healthy tissue and found a direct correlation between the two.

One study examined EC17 in combination with the Artemis camera system; however, the authors reported that there was too much background noise from the autofluorescence of normal breast tissue to enable tumor identification.³⁸ Lastly, there was a phase two comparative study evaluating 6 and 12 mg of tozulesteride in combination with the SIRIS camera system in-vivo; however, TBR was not reported.¹⁹ Although targeting calcium deposits instead of tumor receptors, one study applied PM700-Ca and PM800-SO3 on excised breast tissue and combined it with the mini-FLARE to evaluate pre-cancerous ductal carcinoma in-situ.³⁶

Three studies used enzyme targeting fluorophores [Fig. 3(c)].^23,24,42 Two studies administered LUM015,^23,24 which requires cleavage by cathepsin to be activated. Utilizing the LUM imaging system, this combination exhibited the highest TBR demonstrating $4.70\pm 1.23$ and $4.22\pm 0.96$ when receiving a 0.5 and $1.0\mathrm{mg}/\mathrm{kg}$ dose, respectively.²³ One study by Unkart et al.²⁵ capitalized on AVB-620, which requires activation via matrix metalloproteinases. They recorded a TBR of $1.85\pm 11$.²⁵

Ten studies calculated TBR using the average pixel fluorescence intensity in the entire region of interest in the tumor versus background on the ex-vivo specimen.^{20–25,27,31,32,37} Of those, three used the bisected specimen to calculate the value of the signal.^22–24 In addition, two studies compared a matched number of pixels from tumor and background in both ex-vivo and histopathology specimens for two timings.^14,28 However, none of the studies went into sufficient detail regarding the way in which tumor or healthy tissue was marked, but stressed that the findings were confirmed on histopathology with fluorescence. Only Koch et al. calculated TBR relative to each patient.²⁰

There were three studies in which the signal-to-background ratio was measured.^27,28,41 Pop et al.⁴¹ assessed the area suspicious for tumor intraoperatively, Keating et al.²⁷ examined the resection cavity, and Leiloglou et al.²⁸ compared the difference between freshly excised tissue and histopathology specimen (which had undergone formalin fixation).^27,28,41

A further four studies performed a qualitative analysis.^21,37,41,43 Both Pop et al.⁴¹ and Smith et al.²³ examined the cavity, whereas Koller et al.²¹ looked at both in-vivo and ex-vivo tissues. Conversely, Veys et al.⁴³ compared benign and malignant lesions.

3.5.

Image Processing

Four studies assessed accuracy using texture metrics.^14,20,28,33 Leiloglou et al.,²⁸ Leiloglou et al.,³³ and Kedrzycki et al.¹⁴ performed image analysis via Fourier transformation for slope and intercept. Koch et al.²⁰ capitalized on $f\mathrm{STREAM}$ to streamline the intensity and spatial correlation.

Three studies mentioned the software used to determine TBR.^20,27,37 This includes the HeatMap plugin within ImageJ (National Institutes of Health, Bethesda, Maryland),²⁷ fSTREAM,²⁰ and IC-Calc 2.0.³⁷

3.6.

Intraoperative FGS

Six studies utilized FGS intraoperatively.^{24,30,31,34,35,40} Three employed intralesional ICG which was used for both in-vivo guidance and assessing the cavity to confirm adequacy of resection. ^30,34,35 Two used the PDE system (5.4% and 12.5% PMR, respectively),^30,35 and one used the visual navigator system (10.5% PMR).³⁴ The remaining three studies that reported PMR used conventional techniques (such as guidewires or seeds), thus PMR and reoperation rate were irrelevant to FGS.^24,31,40 Of these, two provided in-vivo cavity images and ex-vivo images of BCS specimens, but surgical guidance was discretionary.^31,40 Lastly, although marker localization was used intraoperatively, Smith et al. opted for additional cavity shaves in the event of fluorescence, resulting in a PMR of 17.8% and a reoperation rate of 8.9% with the LUM system.²⁴

3.7.

Diagnostic Accuracy

Seven studies assessed diagnostic accuracy using passive nonspecific fluorophores. Kedrzycki et al. observed a sensitivity of 69% and specificity of 97% in a study utilizing a custom built camera with pixel based processing to detect ICG fluorescence.¹⁴ However, when Leiloglou et al. applied texture metrics, a sensitivity of 75% and specificity of 89% were achieved.²⁸ Liu et al. used the PDE to detect ICG signal and achieved a PPV and FPV of 100% and 0%.³⁵ Veys et al. also assessed ICG by deploying the Fluobeam 800 imaging system and obtained a sensitivity and specificity of 94% and 32%, respectively.³⁷ However, Pop et al. achieved a specificity of 60% and a PPV of 29% with the same combination.⁴¹ Zhang et al. developed their custom-built imaging system for MB signal detection and achieved a sensitivity and PPV of 63% and 79%, respectively.³² Lastly, Ottolino-Perry et al. utilized PRODIGI to detect 5-ALA’s metabolite (PpIX) signal and compared the diagnostic accuracy between the low dose ($15\mathrm{mg}/\mathrm{kg}$) and high dose ($30\mathrm{mg}/\mathrm{kg}$) cohorts.²² In the low-dose cohort, they presented a sensitivity, specificity, and PPV of 65%, 85%, and 77%, respectively.²² In the high dose cohort, they recorded a sensitivity, specificity, and PPV of 68%, 80%, and 75%, respectively.²²

There were only two studies that reported the diagnostic accuracy of targeted fluorophores. Specifically, Smith et al. deployed the LUM imaging system for LUM015 detection and achieved 84% sensitivity and 73% specificity.²⁴ Koch et al. achieved the highest sensitivity and specificity while capitalizing on the combination of bevacizumab–IRDye800CW and a custom-built imaging system, attaining 98% and 79%, respectively.²⁰ The calculations for diagnostic accuracy can be found in the Supplementary Material.

3.8.

Reproducibility

Fourteen studies provided inclusion/exclusion criteria,^{13,14,20,22–25,27,28,30,31,37,38} of which only Ottolino-Perry et al.²² had listed a minimum tumor size threshold. Six provided control samples; however, none of the studies included benign tumors for comparison.^{14,22,23,28,30,34} Thirteen studies described how TBR was calculated.^{13,14,19–25,27,28,31,37} Four studies provided a minimum TBR,^13,27,37,41 with a further two using patient-normalized thresholds.^21,24 Five compared individual cancer types,^{14,19,28,30,37} and five were able to detect DCIS.^{13,19,23,31,36} Three described the seniority of surgeons included in their study^14,28,30 but no study described whether they were trained in FGS (Table 3).

Table 3

Assessment of consistency between studies. TBR, tumor-to-background ratio; DCIS, ductal carcinoma in-situ; FGS, fluorescence guided surgery.

3.9.

Adverse Events

There were no serious adverse events relating to any of the fluorescence imaging systems. Only one patient had a hematoma attributed to the device (due to pressure applied in the cavity); however, this resolved spontaneously.²⁴

Eight studies reported adverse events due to drug-related side effects.^{21–24,27,31,32,38} These included: one patient with mild nausea successfully treated with IV diphenhydramine,²⁷ another patient with untreated nausea and one with hot flushes (both of which recovered spontaneously),²¹ five patients with mild transient pain on injection of MB (three of which were successfully treated with saline flush),^31,32 one had blue skin discoloration after extravasation of LUM015 (which resolved within 3 months),²⁴ and one with self-limiting hypersensitivity to EC17 (abdominal discomfort, itching throat, sneezing) during injection.³⁸ Furthermore, there was one case of sunburn with ALA; however, it was due to a patient not abiding by the post-operative protocol.²²

One patient experienced adverse events related to the anesthetic, exhibiting transient hypertension on induction and awakening.²³ Furthermore, there was a case of transient peri-operative hypertension and another case with peri-operative nausea; however, both were reported to be unlikely related to the trial.²⁴