Exploring and validating innovative methods for detection and localization of head and neck squamous cell carcinoma primary tumors and lymph node metastases

Early and accurate identification of head and neck squamous cell carcinoma (HNSCC) is important to improve treatment outcomes and prognosis. New optical imaging techniques may assist in both the diagnostic process as well as in the operative setting by real-time visualization and delineation of tumor. Narrow Band Imaging (NBI) is an endoscopic technique that uses blue and green light to enhance mucosal and submucosal blood vessels, leading to better detection of (pre)malignant lesions showing aberrant blood vessel patterns. Fluorescence optical imaging makes use of near-infrared fluorescent agents to visualize and delineate HNSCC, resulting in fewer positive surgical margins. Targeted fluorescent agents, such as fluorophores conjugated to antibodies, show the most promising results. The aim of this review is: (1) to provide the clinical head and neck surgeon an overview of the current clinical status of various optical imaging techniques in head and neck cancer; (2) to provide an in-depth review of NBI and fluorescence optical imaging, as these techniques have the highest potential for clinical implementation; and (3) to describe future improvements and developments within the field of these two techniques.


Introduction
Delay in diagnosis of head and neck squamous cell carcinoma (HNSCC) ultimately results in advanced stage tumors with poorer survival. Despite extensive treatments, this consequently leads to higher treatment-related morbidity. 1 Identification of early (pre) malignant lesions is generally recognized as essential in HNSCC. In the past decades, much attention has been devoted to developing optical techniques to detect lesions at an early stage and correctly diagnose suspicious lesions. Recent technological developments in optical imaging have enabled new possibilities for early detection, better classification, and intraoperative tumor margin delineation. The main advantage of optical techniques over conventional radiologic imaging techniques is the potential to detect superficial mucosal lesions before being visible on scans. As conventional imaging techniques cannot be used intraoperatively, optical techniques for visualization of tumor borders and identification of tumor tissue during surgery is desirable.
In this review we aim to provide a brief overview of current optical techniques with realtime applicability for the detection and optical visualization of HNSCC tumors intraoperatively. Moreover, Narrow Band Imaging (NBI) and near-infrared (NIR) fluorescence imaging will be discussed more extensively, since these techniques are gaining more interest in clinical use because they provide wide-field real-time visualization.

Techniques with real-time applicability
New techniques with real-time applicability have been attempted in order to better detect or delineate a primary HNSCC for the purpose of either mucosal or deep margin assessment during surgery (Fig. 1). Optical techniques can currently be divided into techniques that use surface or subsurface visualization of the target tissue, quantify optical properties or expose tissue characteristics by adding optical agents into the tissue. 2,3 The various optical imaging techniques function optimally within a certain range of the light spectrum. Wavelengths in the visible light spectrum (<650 nm) are absorbed by hemoglobin and absorption by components such as water and lipids are within the infrared range (>900 nm). 4 The optimal window for fluorescent dyes is therefore in the near infrared (NIR) spectrum (700-900 nm), in which light may travel up to millimeters rather than micrometers. 5 For tumor delineation and visualization there are several mucosal staining methods. Lugol's iodine solution stains normal mucosa brown but leaves dysplastic epithelium white or pink in color as a result of decreased glycogen in these tissues. 6 In the detection of oral SCC and dysplastic lesions an 88% sensitivity and 84% specificity were found and it was shown to reduce the incidence of dysplasia in mucosal surgical margins (Table 1). 7,8 The low specificity can be attributed to areas with inflammation, hyperkeratosis and mucus, as these cause the tissue to remain unstained. 8 Lugol has not been applied widely due to its low specificity and limited applicability in the oral cavity, since it can only be used in non-keratinized epithelium. Moreover, in other HNSCC sites it poses the risk of aspiration into the airway. 9 Toluidine blue has an affinity for nucleic acids. As dysplastic and abnormal tissues have higher RNA and DNA content, they show more intense staining than healthy tissues. 10 A recent meta-analysis found only a 75% sensitivity and 60% specificity for the detection of oral lesions. 11 Toluidine blue therefore has, like Lugol, limited clinical benefits.
Although not strictly optical imaging, Intraoral ultrasound is a method that is increasingly used for determining the mucosal and deep resection margins of oral tongue SCC with promising results. 12 In a systematic review, eight out of ten studies showed a significant correlation between ultrasonographically measured and histologic tumor thickness. Best results are obtained in vivo and in two studies comparing with MRI, ultrasound was more precise. The results, however, need to be validated in prospective studies using the same methodology. A limiting factor for the application of ultrasound intraorally is the proximity of bone or salivary glands, which compromises the interpretation of the images. As ultrasound is not an optical imaging technique, it is not discussed further here.
Optical coherence tomography (OCT), analogous to ultrasound, uses reflecting coherent light to make cross-sectional images of the underlying architecture in real-time with a penetration depth of 0.5-2 mm. [13][14][15] It is best used in the oral cavity due to its accessibility, where epithelial thickening may indicate malignancy. 16 In a healthy larynx it can identify the different layers; however, in laryngeal cancer the boundary of the basal membrane is lost and leads to unrecognizable structures, hampering identification and diagnosis. 17 In vivo sensitivity and specificity rates of 100% were found for the detection of oral SCC, compared to 93% and 69% for dysplastic lesions, respectively. 18 A limitation of OCT, however, is that it cannot differentiate between malignancy and inflammation. 16 OCT is still far from implementation in everyday clinical practice, as there is no classification to consistently diagnose lesions and since results need to be validated.
Autofluorescence can be detected due to endogenous fluorophores, such as structural proteins. Most of the autofluorescence is emitted between 450-650 nm and disappears beyond 1500 nm. 19 The endogenous fluorophores appear in green when excited by ultraviolet light or visible light radiation of certain wavelengths. 20,21 Pre-cancerous and malignant lesions appear as red-violet caused by an altered metabolism in tumor cells, less collagen in neoplastic lesions emitting fewer photons, and due to a loss of fluorescent signal resulting from epithelial thickening (blocking excitatory blue light). 20,22 In a meta-analysis of oral lesions a 78% sensitivity and 48% specificity were found and autofluorescence had a lower area under the curve than regular clinical examination (p = 0.0023). 11 The low specificity may be caused by interpretation errors of images due to variety in autofluorescence caused by scar formation, minimal amounts of blood, bacterial overgrowth and inflammation, limiting its clinical benefit compared to white light examination. 21 Raman spectroscopy is based on the fact that photons are scattered at a different wavelength (up or down wavelength shift) due to energy loss while interacting with the vibrational modes of certain molecules (inelastic scattering). 23 This shift in wavelength is measured and called the Raman scatter. In a pilot study, Raman spectroscopy was able to differentiate (pre)malignant oral lesions from normal tissue with a 100% sensitivity and 77% specificity. 24 In a small study of laryngeal biopsy specimens, a 90% sensitivity and 92% specificity was found for the detection of laryngeal SCC. 25 Although these results seem promising, analyzing an entire specimen may take up to several hours, in vivo accessibility is difficult as only rigid scopes can be used, and interpretation of results is more mathematical than visual. To analyze a specimen more quickly, the number of measurements could be reduced by sampling spectroscopy data only in specific areas of interest with a focus on spectral features associated with the presence of cancer cells (such as water content). In this manner, a 99% sensitivity and 92% specificity was found in the detection of tongue SCC, as tumor tissue had higher water content than healthy tissues. 26 Raman spectroscopy has not yet been developed for commercial applications and the first system for surgical guidance in head and neck tumors was expected to be ready for clinical trials in 2019. 27,28 Due to the rather technical nature of spectroscopy and non-visual applications, including the interpretation of results, this does not fall within the scope of this review. Raman spectroscopy, as other spectroscopies such as inelastic reflectance spectroscopies, may, however, serve as an additional tool supporting visual observations since spectroscopic techniques have the capacity to provide quantitative measurements, whereas visual techniques are at best semi-quantitative.
From the above-described optical techniques, Narrow Band Imaging (NBI) and Fluorescence optical imaging are currently most widely and continuously studied in clinical trials by a multitude of academic centers, reflecting a higher possibility of adaptation into routine clinical use in the foreseeable future. We therefore focus the review more on these two techniques.

Narrow Band Imaging
In the development of cancer, the structure and organization of vasculature may change considerably due to angiogenesis, forming intraepithelial papillary capillary loops (IPCL). 29 Enhanced visualization of these new abnormal blood vessels leads to increased and earlier detection of precancerous and cancerous lesions. 30 NBI is an endoscopic optical imaging technique that has gained increasing attention over the past two decades. Rather than using conventional white light, two narrow band filters of blue and green are used. These filters have a bandwidth of 30 nm with central wavelengths of 415 nm and 540 nm, respectively, which are maximally absorbed by hemoglobin. 9 As longer wavelengths of light penetrate more deeply into tissues due to scattering and absorption properties, the blue light at 415 nm enhances the visualization of superficial mucosal vascular patterns and the green light at 540 nm enhances submucosal IPCLs. 31 The light is captured by a charge coupled device at the tip of the videoendoscope and the image is then reconstructed, displaying the superficial vessels in brown and submucosal vessels in cyan, increasing the contrast between vasculature and surrounding mucosa. 30,32 Positive NBI findings are generally described as well-demarcated brownish areas with scattered brown dots. 33 NBI seems to be a promising technique in the detection of laryngeal, pharyngeal and oral malignancies with an overall sensitivity of 89% and specificity of 96% and was found to perform better when combined with white light imaging (WLI). 33,34 However, NBI (or comparable imaging techniques) is not considered standard of care yet, since evidence on the influence of NBI on clinical outcome or effect on health care costs has not yet been provided by large prospective studies or randomized controlled trials.
As vascular patterns associated with (pre)malignant lesions depend on the mucosa type, these alterations differ per HNSCC site and therefore different classifications are being used.

NBI in detection of laryngeal lesions
Laryngeal cancer is one of the most common types of HNSCC. 35 The most widely used classification for describing laryngeal lesions using NBI was described by Ni et al. (Fig. 2). 30 They divided lesions into five types. Types I to III represent non-malignant lesions: type I) thin, oblique, and treelike/arborescent vessels without IPCLs; type II) enlarged diameter of oblique and arborescent vessels without IPCLs; type III) a white patch on the mucosa may, depending on its thickness, obscure the arborescent vessels, without IPCLs. Types IV and V indicate (pre)malignant lesions: type IV) IPCLs are visible as scattered, regular, small, and dark brown spots; type V) a: high density of irregular IPCLs that appear hollow, brownish and speckled and in various shapes, b: IPCLs are destroyed and appear 'woven' with tortuous and line-like shapes, c: uneven density and irregularly scattered brown speckles or tortuous shapes. 30,36 As there is no significant difference in sensitivity and specificity for a cutoff value for types IV and V combined or solely type V, a more recent classification described by the European Laryngological Society (ELS) might seem more applicable. 37 It describes benign conditions to show longitudinal vascular changes with increased number and density of blood vessels and changes of direction, while precancerous or cancerous lesions are characterized by their newly formed perpendicular blood vessels, recognized as dilated IPCLs. 38 In the largest and most recent meta-analysis of NBI in laryngeal cancer, a 90% sensitivity and 90% specificity were found for NBI, compared to 78% and 77% for WLI, respectively. 39 NBI showed excellent interrater reliability with an intraclass correlation coefficient of 0.91, in line with earlier findings concluding that NBI improves inter-and intraobserver agreement for both experienced and less-experienced observers. 40 As some lesions may cause difficulties in interpretation when strictly following classifications, lesion-specific clinical features should also be taken into account. 41 Caution should be taken when interpreting results on the accuracy of NBI, as these results are highly dependent on the experience of the observers, whether clinical information was available, and on the type of evaluated and included lesions.
For instance, papillomas and post-treatment mucosa influences correct interpretation, as these lesions often appear with Ni type V vascular patterns. 36,41,42 In the ELS classification, the difficulty of discerning papillomas is overcome by describing HPV-related recurrent respiratory papillomatosis as vessel loops with a wide-angled turning point in a three-dimensional warty structure, compared to symmetrically arranged loops with a narrow-angled turning point for precancerous lesions. 38

NBI in detection of lesions in the oral cavity, oropharynx and hypopharynx
In the pharynx and oral cavity, three different types of epithelium can be found: type 1) keratinized thick stratified squamous; type 2a) nonkeratinized thin or 2b) very thick stratified squamous; type 3) pseudo-stratified ciliated columnar epithelium. 43 Consequently, the criteria described for laryngeal lesions cannot be applied. As the penetration depth of blue and green light is respectively 170 µm and 240 µm, this could limit the evaluation of some lesions in oral cavity and oropharyngeal subsites, where the mean mucosal thickness measures up to 1300 µm. 44 Consequently, the prevalence of brownish spots in early mucosal cancer was significantly higher in type 2a mucosa (floor of mouth, ventral tongue, soft palate, palatine tonsils, base of tongue, epiglottis and hypopharynx) compared to other types of mucosa in different subsites. 43 A widely used classification by Takano et al. was shown to perform significantly better than other classifications (Fig. 3). 45,46 This classification divides findings into four types: type I) normal mucosa, regular brown dots, type II) IPCL pattern dilation and crossing, type III) IPCL pattern elongation and meandering, and type IV) IPCL pattern destruction and angiogenesis following a sequence of carcinogenesis progression. 45 In a meta-analysis a sensitivity of 76% and specificity of 92% were found for the detection of oral cavity and oropharyngeal dysplastic or malignant lesions using NBI. 47 Moreover, as leukoplakia may transform into malignancies in up to 18% and erythroplakia in up to 91% of the cases, increased detection rates of these precancerous lesions by NBI is beneficial. 44 Piazza et al. showed that the performance of NBI (sensitivity 89%, specificity 85%) was significantly superior to WLI (sensitivity 78%, specificity 73%) or conventional oral examination (sensitivity 51%, specificity 68%). 44

Detection of the primary tumor in neck metastasis of an unknown primary HNSCC
SCC lymph node metastasis without any overt primary tumor, i.e., an unknown primary tumor (UPT), occurs in up to 4% of all HNSCC and is associated with a poor prognosis. 48 Conventional imaging techniques such as PET-CT, which has a 44% sensitivity and 97% specificity, may miss small superficial mucosal lesions. 49 A most recent technique, mucosectomy using an operation robot, adds transoral removal of lingual tonsils to palatine tonsillectomy, thereby increasing the detection rate of primary tumors. 50 However, this technique is invasive with a chance of complications and morbidity. NBI could overcome these limitations, and sensitivity rates of 74-91% and specificity rates of 86-95% have been reported. 33,51 Although detection rates may seem relatively low, it should be considered that these are found after a full clinical workup, sometimes even after biopsies and tonsillectomy. 33 Filauro et al. identified previously undetected lesions in 35% of patients using NBI, even after WLI endoscopy and CT, MRI or PET-CT. 51 As the lesions detected by NBI were generally small, they may still be removed by endoscopic resection, avoiding large field radiation and associated toxicity. 33

NBI in detection of recurrences
Radiotherapy or chemoradiotherapy cause changes to the mucosa which make identification of recurrent lesions more difficult. Among these changes are swelling, chronic inflammation with hypervascularization, and viscous mucus due to reduced saliva production. 52 This hampers timely identification of recurrent tumors, leading to more advanced stages, and consequently diminished therapeutic options. 48 Using NBI, however, 20% more residual or recurrent tumors could be detected than with standard WLI. 48 Attention should be paid to pronounced and irregular feeding vessels, necrosis and ulceration at the mucosal surface, since normal IPCL patterns may no longer be present and classifications for untreated primary tumors (e.g., Ni's classification) are thus inapplicable. 52 Lesions were detected at an early stage (Tcis and T1) in 92% of cases, and could therefore be treated with a conservative surgical approach. 48 For NBI, 88-100% sensitivity rates were found depending on outpatient or intraoperative setting, and 92-98% specificity rates, compared to a 66% sensitivity and 100% specificity for WLI. 48,52

Future perspectives
With the advance of technology, further improvements with respect to endoscopy may also be expected, such as endoscopes with 4 K or even 8 K ultra-high-definition (UHD) resolution. Although this may improve the visualization of the lesion, the judgment whether a lesion is benign or malignant currently remains dependent on the observer. Classifications attempt to reduce interobserver variability and standardize diagnostics, but a computer-based diagnostic aid using artificial intelligence may be more effective in standardizing diagnostics and at the same time improve diagnostic accuracy and speed. Moreover, it could assist less experienced observers (like residents in training) new to NBI and reduce their learning curve, which is considered to be approximately 6 months. 48 Using neural networks, first attempts have been made at detecting pharyngeal and laryngeal SCC. [53][54][55] In a pilot study by Mascharak et al. using only 4 OPSCC lesions for training of the neural networks, a 71% sensitivity and 69% specificity for the detection of OPSCC were found, of which the sensitivity was significantly higher compared to WLI (47%). 53 They used color and textural information of surrounding mucosa, which gave significantly better results than color information alone. 53 Addition of more image features, such as blood vessel characteristics, and a larger training dataset could possibly further increase the diagnostic accuracy of this first attempt. A neural network used by Tamashiro et al. was able to detect pharyngeal cancers, including lesions smaller than 10 mm, with an 86% sensitivity and 59% specificity for NBI, compared to a 70% sensitivity and 52% specificity for WLI. 54 Suboptimal visualization conditions such as the lesion being too distant, images showing only part of the lesion or a tangential view on the lesion hamper the diagnosis by the neural networks, and were the cause of half of the false-negative results. 54 Their system was able to analyze 50 images per second and is therefore suitable for real-time analysis. 54 Another neural network developed by Mantowski et al. was able to detect laryngeal lesions in a small set of 50 images, distinguishing papilloma/carcinoma lesions from healthy vocal cords with a 94% sensitivity and 91% specificity. 55 With improved neural networks and proper visualization of lesions during endoscopy, these first attempts hold promise for implementation of real-time assistance for the diagnosis of HNSCC in a clinical setting.

Optical molecular imaging using near-infrared fluorescence
As conventional imaging techniques only reveal anatomical structures and do not offer real-time intraoperative guidance, surgeons are dependent on their own skills and judgment to decide which tissues to resect. After resection, the specimen and wound bed are inspected macroscopically, often followed by a (time-consuming) frozen section. Assessing the whole resection margin with frozen sections is generally technically not possible, thus having the risk of sampling errors. This approach leads to positive margins in 12-30% of cases. [56][57][58] As positive margins are associated with a poor prognosis, there is a need for improved intraoperative detection and visualization of HNSCC. 59 Fluorescent agents can be used to aid in the real-time identification of primary HNSCC. In optical image-guided surgery, the tumor-to-background ratio (TBR) is important to distinguish tumor from normal tissues. 4 Ideally, all viable tumor tissue should fluoresce, while all surrounding normal tissue should be negative for fluorescence. High sensitivity rates are required in order to remove all tumor cells, as leaving any viable tumor cells behind may have significant clinical consequences. Although a high specificity is preferable, it is of lesser importance than sensitivity, as it would not affect survival but mainly the extent of surgery.
Using relatively inexpensive and mobile cameras, optical image-guided surgery can easily be incorporated into the clinical workflow to provide real-time feedback with high imaging resolution and without exposure to ionizing radiation. 19 Moreover, fluorescence imaging may prove especially beneficial in robotic or endoscopic surgery, due to the lack of tactile feedback in these procedures. 5 Potentially, if tumors could more easily be distinguished from normal tissue, complication rates as well as operating room time can be reduced. 5 Over the past decades, fluorescence imaging has become a booming research area, with a growing number of agents and targeting strategies. Tumors may be imaged in the operating theater before or during resection (open-field), while the resected specimen can be imaged in a closed-field/back-table setting, leading to better results due to more constant imaging variables (e.g., ambient light, distance between camera and specimen).
Methylene blue can, when sufficiently diluted, function as NIR dye emitting at around 700 nm and may be applied topically. 5,19 However, as it stains the entire surgical field and has also been overtaken by other fluorescent agents in terms of tissue penetration, it has not gained much attention. 5,60 ICG fluoresces at 800 nm and tumor localization can be observed as a consequence of the enhanced permeability and retention (EPR) effect that is seen in newly formed tumor tissue. In these tissues, newly formed porous blood vessels allow molecules to passively accumulate, while retention occurs due to a poorly tumor associated lymphatic system. 61 Identification and visualization of ICG may be up to 5 millimeters through soft tissue. 62 The use of 5-ALA has been approved in Europe since 2007 and eventually also in the U.S. since 2017 for the use of intraoperative imaging of gliomas. 63 5-ALA is a precursor molecule in the formation of protoporphyrin IX (PpIX). PpIX is a fluorescent molecule that mainly accumulates in gliomas, meningiomas and bladder cancer and has two emission peaks, of which one is at ~700 nm. 5,19 However, as it is dependent on high tumor metabolic rates, has great interindividual differences and has wavelengths with suboptimal tissue penetration, it has not been widely used outside of glioma surgery. 19,64 Overall, untargeted fluorescent agents may lead to non-specific fluorescence, higher background signals and faster clearance, reducing the time window for surgery. 19 Therefore, attempts have been made to bind fluorescent molecules to specific tumor-targeting agents for increased accuracy and efficacy.

Targeted near-infrared fluorescent agents
As the unbridled growth of a malignancy is associated with higher expression of growth factor receptors, angiogenesis and tumor associated metabolism, these processes form potential targets for molecular imaging. 65 Using targeted molecules can result in higher specificity and affinity, potentially lowering the required dose of the fluorescent agent and leading to better TBRs. 19 Moreover, with an increased binding time these agents could result in a prolonged time window to perform surgery, compared to ICG where peak fluorescence decreased after one hour. Targets that have been studied in HNSCC are the vascular endothelial growth factor (VEGF) receptor, epidermal growth factor receptor (EGFR), as well as tumor pH related to the Warburg effect. 66 The Warburg effect describes a change in cancer cell metabolism with higher glycolysis rates and lactate production, resulting in a lower pH. Fluorophores can be conjugated to targeting molecules, which creates fluorescent agents. Among these fluorophores, Cy5.5 and IRDye800CW have frequently been described in HNSCC. 56,64,[67][68][69][70][71][72][73][74][75][76][77] However, as IRDye800CW has higher excitation and emission wavelengths (~790 nm) for NIR imaging than Cy5.5 (~700 nm) resulting in better tissue propagation, the former has been used most in recent studies. 65 VEGF is a promotor of angiogenesis and can be targeted using the antibody bevacizumab. 68 Although fluorescence could predict presence of tumor in 17 out of 18 biopsies in xenograft mouse models using bevacizumab-Cy5.5, slightly better results were found in the same mouse model using cetuximab-Cy5.5 targeting EGFR. 67,68 As bevacizumab-IRDye800CW has proven its clinical value in other types of cancer, it may be worth further studying its potential in HNSCC. [78][79][80][81] One study on the application of bevacizumab-IRDye800CW in sinonasal inverted papilloma is ongoing (NCT03925285).
EGFR is a transmembrane protein that is overexpressed in 90% of HNSCC and has been introduced as a target for fluorescence imaging. 64,82,83 Cetuximab is a chimeric human-murine monoclonal antibody that binds with high affinity to EGFR. 83 Several studies have used cetuximab-IRDye800CW for fluorescence imaging of HNSCC. 64,70,72,77 The TBR increased over several days after infusion, suggesting that the best moment for surgery may be 2-4 days after infusion. 64 By preloading patients with unlabeled cetuximab the highest TBRs were found, requiring lower doses of cetuximab-IRDye800CW. One study showed a feasibility of only 15 mg total dose of cetuximab-IRDye800CW conjugate for clinical use when pre-dosing was applied. 70,77 The increased TBR by preloading can be explained by the fact that antibodies accumulate more easily in normal tissue than in tumor tissue, since tumor tissue has a slower uptake as a consequence of immature vascularization and elevated interstitial pressures. Without preloading, the labeled antibodies are sequestrated in non-specific tissues, resulting in a decreased TBR. 70 In a study by Voskuil et al., all cases with a tumor-positive resection margin (<1 mm) were correctly identified at back-table fluorescence imaging of the surgical specimen directly after resection, and could assist in selecting suspect areas for fresh-frozen sections. 77 There was one false-positive case, resulting in an overall sensitivity of 100% and specificity of 91%. Moreover, in this study, one satellite tumor lesion was found, underlining the potential benefit of cetuximab-IRDye800CW in the detection of HNSCC. The use of cetuximab-IRDye800CW can be considered to be well tolerated, as there were five possibly related grade 1 adverse events out of fifteen patients. 77 Currently, a trial is ongoing studying the positive predictive value of fluorescence in margin assessment of oral cancers (NCT03134846).
Following the results found with cetuximab-IRDye800CW, several studies have been performed using panitumumab-IRDye800CW. 56,69,[73][74][75][76] Panitumumab has a higher binding affinity to EGFR and improved safety profile compared to cetuximab, as it is a fully humanized monoclonal antibody. 73 An optimal dose of 50 mg panitumumab-IRDye800CW was found to lead to the best TBRs of up to 3.6. 76 Preloading patients with an unlabeled dose of panitumumab did not lead to additional imaging value, although it did prolong the fluorescence intensity of the labeled antibodies. 76 As with cetuximab-IRDye800CW, unanticipated second primary lesions may be detected by open-field imaging and improve surgical decision making. 74 Using closedfield analysis, TBRs of up to 6.5 were found for a dose of 50 mg panitumumab-IRDye800CW and resulted in a 100% sensitivity and 74% specificity for the detection of positive surgical margins. 73 A penetration depth of 6.3 mm was found; sufficient to detect surgical margins in HNSCC that are considered close or positive within 5 mm. 56 Back-table specimen mapping could assist in intraoperative decision making. When fluorescent hotspots are detected, this can be used to guide pathologists in taking frozen sections and thus reduce sampling errors. 73 Currently, studies are conducted in which intraoperative decision making based on the fluorescence is possible, without frozen sections. Optical analysis of an excised specimen typically takes <10 minutes and requires a device operator showing the data to the surgeon or pathologist.

Future perspectives
To further improve these results, development of new techniques can be predicted in different domains. With improved camera systems, targeting molecules with better biodistribution and binding affinity, and on-site activatable fluorescent markers, better results may be obtained.
Currently, differences in TBR are seen between open-field and closed-field imaging, in favor of the latter. 73 This is attributed to ambient light, camera angle, and distance between camera and patient affecting open-field imaging. 74 To reduce interference of ambient light, overhead lighting could be reduced or turned off. This, however, interrupts the surgical workflow. 73 New image acquisition techniques can potentially overcome this hindrance by pulsed image acquisition adjusted to the frequency of background lighting, thereby enabling subtraction of background lighting and resulting in images similar to those acquired in a completely dark room. 84 However, as orientation and interference of shadowing is especially difficult in HNSCC surgery using open-field imaging, the surgical value of fluorescence imaging may be greatest in closed-field back-table evaluation of the specimens. 73 Although results using whole antibodies such as cetuximab and panitumumab seem promising, they have a relatively large hydrodynamic diameter, resulting in accumulation in the liver, a longer bloodstream half-life, slower blood clearance and thus potentially increased background signal. 5,85 To overcome this disadvantage, smaller antigen-binding fragments have been developed such as nanobodies, which are derived from naturally occurring heavychain antibodies. 5,85,86 They show very specific binding and can extravasate more easily due to their small size -about one tenth of the molecular weight of monoclonal antibodies -and show rapid clearance from the body. 85,86 In a xenograft mouse model, the nanobody 7D12-IR showed optimal fluorescence contrast 2 h post-injection, compared to 24 h for cetuximab-IR. Moreover, it showed a relatively higher uptake, related to better penetration and distribution, compared to cetuximab-IR. 86 Affibody molecules are another alternative for antibodies. They are small, engineered peptides that may also bind to antibody targets. 87 Due to its rapid distribution and sufficient contrast already at 4 h after injection, it could allow for same-day surgery. Moreover, due to faster clearance, it reduces toxicity risk compared to monoclonal antibodies. 87 An Affibody trial has been completed using EGFR as a target in HNSCC and reporting of the data is expected (NCT03282461).
In addition to the currently existing fluorescent markers, quantum dots may hold promise for the future. They are small crystal nanoparticles of 2-10 nm in diameter with high signal intensity, photostability, a broad and continuous distribution of the excitation spectrum and a narrow emission spectrum that can be tuned by modifying composition and size. 65,88,89 In mice αvβ3-integrin was targeted using QD800-RGD and showed TBRs of up to 4.5 while imaging even through the skin barrier . 89 However, as the core of quantum dots is made up of heavy metals, they are potentially toxic when the cores are released. 88 Therefore, the use of quantum dots may not find their way to clinical application until quantum dots with nontoxic cores have been developed.
Another strategy of making fluorescent agents more specific, is by making them 'activatable' rather than using them in an 'always-on' state. This can be achieved by using amphiphilic polymers that form micelles in aqueous solutions and sequestrate the fluorophore, making it 'pH-activatable' and releasing the fluorophore at a certain pH threshold, as tumor is known for its acidic environment (pH <6.9). 19,66 First in vivo results using an ultra-pH sensitive amphilic polymer conjugated to ICG (ONM-100), show a 100% sensitivity and 57% specificity for the detection of HNSCC and led to the identification of 5 occult lesions that had otherwise been missed by standard of care surgery. 66

Conclusion
New imaging techniques have been developed for earlier and more accurate detection of HNSCC. Among these are NBI and near-infrared fluorescence imaging, which are already applied clinically and show the most potential for routine clinical use in the near future. NBI leads to improved and earlier detection of superficial mucosal HNSCC lesions compared to WLI. Future developments may enable real-time computer-assisted diagnosis during endoscopy. Nearinfrared fluorescence imaging targets tumor tissue and allows for intraoperative visualization, localization and delineation of primary tumor and occult satellite lesions, and offers as main clinical advantage intraoperative margin assessment. High sensitivity and specificity rates have been found, especially when using antibody-dye conjugations. Development of better camera systems, improved targeting agents with higher affinity and biodistribution and activatable agents (e.g., pH-activatable) may further increase the future potential of near-infrared fluorescence imaging. Currently, an abundance of clinical trials are being undertaken and they are expected to lead to the wide introduction of one of these compounds.