The clinical usefulness of optical coherence tomography during cancer interventions

Introduction Tumor detection and visualization plays a key role in the clinical workflow of a patient with suspected cancer, both in the diagnosis and treatment. Several optical imaging techniques have been evaluated for guidance during oncological interventions. Optical coherence tomography (OCT) is a technique which has been widely evaluated during the past decades. This review aims to determine the clinical usefulness of OCT during cancer interventions focussing on qualitative features, quantitative features and the diagnostic value of OCT. Methods A systematic literature search was performed for articles published before May 2018 using OCT in the field of surgical oncology. Based on these articles, an overview of the clinical usefulness of OCT was provided per tumor type. Results A total of 785 articles were revealed by our search, of which a total of 136 original articles were available for analysis, which formed the basis of this review. OCT is currently utilised for both preoperative diagnosis and intraoperative detection of skin, oral, lung, breast, hepatobiliary, gastrointestinal, urological, and gynaecological malignancies. It showed promising results in tumor detection on a microscopic level, especially using higher resolution imaging techniques, such as high-definition OCT and full-field OCT. Conclusion In the near future, OCT could be used as an additional tool during bronchoscopic or endoscopic interventions and could also be implemented in margin assessment during (laparoscopic) cancer surgery if a laparoscopic or handheld OCT device will be further developed to make routine clinical use possible.


Introduction
Tumor detection and visualization plays a key role in the clinical workflow of a patient with suspected cancer, both in the diagnosis and in the treatment. During the last decades, numerous imaging modalities, such as ultrasound (US), computed tomography (CT), and magnetic resonance imaging (MRI), have proven additional value in establishing the diagnosis of an oncologic patient. Nevertheless, pathologic analysis of representative tumor biopsies is often necessary for establishing the correct diagnosis.
Furthermore, intraoperative detection of the tumor margins is difficult, as surgeons currently mainly rely on visualization and palpation. Pathological techniques to examine the margins intraoperatively, such as frozen section analysis and imprint cytology, have been extensively researched for the purpose of reducing the percentage of positive margins in breast cancer surgery, for instance. However, all these methods have drawbacks, such as time-consuming and resource-intensive nature, difficulty in visualizing high-grade carcinomas, and imprecision, due to sampling errors and poor resolution (Haka et al. 2006;Kennedy et al. 2010;Revesz and Khan 2011). With their ability to image molecular and physiological changes that are associated with cancer sensitively and non-invasively, 1 3 optical imaging devices, such as optical coherence tomography (OCT), have the potential to improve intraoperative tumor detection (Frangioni 2008;Keereweer et al. 2011).
OCT is a technique that uses the interference of light to generate two-dimensional cross-sectional images. It was first described in 1991 and is often denoted as the optical analog of ultrasound; it detects back-reflected light, instead of sound, from tissues (Huang et al. 1991). In the field of cardiology and ophthalmology, it is already used as part of standard clinical care (Vakoc et al. 2012). OCT is, in contrast to other optical image modalities, able to image non-invasively and without the need for tissue preparation. The technique produces images, which are comparable to low-resolution histology. The resolution in comparison with US is 10-50 times better, and usually lies in the range of 1-20 µm in axial and transverse direction, depending on the modality used. This technique could be applied both for ex vivo and in vivo use. Moreover, in the last years, OCT was used during endoscopy or bronchoscopy, by incorporating OCT into flexible fiberoptic probes, which could be inserted in the accessory channel of the majority of standard of care scopes (Jung et al. 2004;Tearney et al. 1997a). For imaging with higher resolution and more cellular detail, high definition OCT (HD-OCT) and full-field OCT (FF-OCT) have been developed. HD-OCT is a commercially available system dedicated to skin imaging (Skintell ® , Agfa Healthcare Mortsel, Belgium and München, Germany) providing axial and transversal resolution of 3 µm over 1.8 × 1.5 mm field of view, however, with penetration depth limited to 570 µm. The penetration depth is also limited to first few hundred microns in FF-OCT that directly acquires 2D en face images (without beam scanning) by illuminating the full field of view with a white-light source, such as a halogen lamp (Boone et al. 2012;Popescu et al. 2011). In FF-OCT, three-dimensional imaging can be performed, by stepping the reference mirror and recording successive en face images resulting in a stack of images (Dubois et al. 2004). With a speed limitation, which is caused by the long acquisition times, higher resolution OCT imaging can be only applied for ex vivo imaging.
The aim of this review is to determine the clinical usefulness of OCT and its variants during cancer interventions for both preoperative diagnosis and intraoperative tumor detection, with a focus on qualitative features, quantitative features and the diagnostic value of OCT, which are described per tumor type.

Results
A total of 785 articles were revealed by our search, of which two were found by manual search on the SPIE digital library. After exclusion of 649 articles, that did not meet our eligibility criteria, a total of 136 original articles remained, of which an overview is given in Fig. 1. The included articles, which form the basis of this review, are discussed separately per cancer type. An overview of the diagnostic value of OCT for tumor detection, including relevant study characteristics, is provided in Table 1.

Skin cancer
Skin tumors are usually divided into melanoma and nonmelanoma cancer. Because of its aggressive character, the only curative treatment for local melanomas is surgical resection in combination with sentinel lymph node mapping. However, in the last years novel target therapies were developed which showed great potential in patients with unresectable or metastatic melanoma (Tripp et al. 2016). For basal cell carcinoma (BCC), which is a non-melanoma cancer and the most common type of cancer in caucasians worldwide, many treatment options are available and applied, dependent on the tumor characteristics and patient's preference (Verkouteren et al. 2017). Mohs micrographic surgery is currently performed in many clinics, to obtain free resection margins. Nevertheless, it would be preferable for both the patient and surgeon to obtain real-time feedback of the margin involvement during surgery. Many studies have determined the capacity of OCT for visualization of different types of skin cancer.

Malignant melanoma
OCT images of a malignant melanoma showed irregular structures in the lower epidermis, which corresponded to histology. The basement membrane zone was also not visible, which made these characteristics specific for malignant 1 3 melanoma (Welzel et al. 1997). Moreover, other characteristics have been investigated. In general, two characteristics were often visible: (1) the presence of horizontal highly reflective cords in the epidermis and dermis, which probably correspond to dense collagen cords of encapsulated tumor lobules and (2) the presence of large vertical icicle-shaped structures reaching the reticular dermis with the peak aspect, which corresponded to tumor cells and lymphocytes infiltration on histology (Fig. 2) (Boone et al. 2014;Gambichler et al. 2007Gambichler et al. , 2014.

Basal cell carcinoma
Several specific features for BCC were suggested, of which disruption of layering, hyporeflective rounded areas surrounded by a hyperreflective halo (honeycomb structure), peripheral palisading and dilated vessels, well circumscribed black/signal poor areas were the most common and characteristic (Alawi et al. 2013;Bechara et al. 2004;Boone et al. 2012;Coleman et al. 2013;Forsea et al. 2010;Gambichler et al. 2007Gambichler et al. , 2014Hinz et al. 2012;Jorgensen et al. 2008;Khandwala et al. 2010;Maier et al. 2013;Meekings et al. 2016;Mogensen et al. 2009aMogensen et al. , b, 2011Olmedo et al. 2006Olmedo et al. , 2007Pomerantz et al. 2011;Wang et al. 2011). Two studies evaluated the diagnostic accuracy of OCT in detecting basal cell carcinomas in vivo, showing good results with sensitivity and specificity ranging from 79 to 94 and 85-96%, respectively (Jorgensen et al. 2008;Mogensen et al. 2009a). Ulrich et al. evaluated the diagnostic value of OCT combined with clinical and dermoscopic assessment with sensitivity and specificity of 96 and 75%, respectively, which resulted in a higher diagnostic accuracy compared to clinical and dermoscopic information (Ulrich et al. 2015). However, even for experienced observers, it was difficult to distinguish BCC from actinic keratosis, which was illustrated by the 50% error rate (Mogensen et al. 2009a). Differentiation between the several BCC subtypes was difficult; however, variants of OCT like HD-OCT and multi-beam swept source OCT (MSS-OCT) showed potential for clinical use (Boone et al. 2012;Gambichler et al. 2007;Meekings (Alawi et al. 2013;Coleman et al. 2014;Cunha et al. 2011;Durkin et al. 2014;Iftimia et al. 2016;Maier et al. 2014;Pelosini et al. 2013;Pomerantz et al. 2011;Wang et al. 2013). Conventional OCT yielded to a sensitivity of 19% and specificity of 56%, whereas HD-OCT showed an improved sensitivity of 75% and a specificity of 64% (Cunha et al. 2011;Maier et al. 2014).

Conclusion
Diverse specific tumor characteristics for both melanoma and BCC were composed. In case of melanoma, no diagnostic studies were performed with OCT. OCT showed good results in BCC detection; however, margin assessment, which is clinically most relevant, was much more difficult even with higher resolution OCT.

Oral cancer
Oral cancer, of which squamous cell carcinoma accounts for 90% of the cases, is often treated by a combination of surgery and radiotherapy (Neville and Day 2002). Due to the difficult location and the surrounding vital structures, it is of outmost important to achieve complete tumor removal. OCT was utilised in ten studies to evaluate its potential use.

Oral (pre-)cancerous lesions
Several parameters were important to distinguish between benign and (pre)malignant oral lesions, such as disorganization of epithelial stratification (irregular collagen vessels), epithelial and/or keratin thickening, micro-structure invasion, heterogeneous cell distribution, and disorganization of the basement membrane (Hamdoon et al. 2016;Leeuw et al. 2015;Wilder-Smith et al. 2009). Four studies in 19-125 patients showed that dysplasia detection was possible both after training of independent reviewers and using quantitative analysis (Adegun et al. 2012;Hamdoon et al. 2013;Jerjes et al. 2010;Lee et al. 2012a). Computer analysis, using a 70% standard deviation of the epithelial thickness, yielded a sensitivity of 82% and a 90% specificity, which indicated that epithelial thickness is one of the most characterizing features of oral dysplasia (Hamdoon et al. 2012;Lee et al. 2012a (Tsai et al. 2008(Tsai et al. , 2009).

Conclusion
Several characteristics for (pre)malignant oral tissue were proposed, all showing good diagnostic accuracies. These morphological characteristics were confirmed by quantitative analysis; nevertheless, no intraoperative studies were yet performed for real-time evaluation of the surgical resection margins.

Lung cancer
The diagnosis of lung cancer, which is the most common cancer in men worldwide, is often made by CT and flexible bronchoscopy. However, bronchoscopy lacks sensitivity, especially for early stage malignancies (Andolfi et al. 2016). Five studies utilized OCT as an additional imaging tool for visualization of lung cancer both during bronchoscopy and after surgery on resected specimens.

OCT during bronchoscopy
Bronchial malignancies were generally characterized on OCT images by a thickened epithelium wall and loss of subepithelial identifiable microstructures. Tumor invasion was visible as a disappearance and/or disturbed architecture of the basement membrane (Lam et al. 2008;Michel et al. 2010;Whiteman et al. 2006). In normal lung tissue, the basement membrane and the lamina propria were visualized as highly reflective layers, due to the presence of collagen vessels. Deeper layers containing seromucinous glands, connective tissue, and cartilage, were characterized by polymorphic light and dark areas (Lam et al. 2008;Michel et al. 2010). Hariri et al. composed specific OCT criteria for the different tumor types: adenocarcinoma, squamous cell carcinoma, and poorly differentiated carcinoma (Hariri et al. 2015). Adenocarcinomas were characterized by round or angulated signalpoor to signal void structures, which were typically small and secondly by lack of signal-intense (bright) nests. Squamous cell carcinoma could be recognized by the presence of signal intense nest (brighter than surrounding tissue), which were round or irregularly shaped. These nests may have variably sizes, and sometimes, also areas of necrosis were visible as signal-poor areas. Lack of round/angulated signal-poor structures and lack of signal-intense nests were specific for poorly differentiated carcinomas. These criteria were applied by Hariri et al. in a prospective validation cohort, in which three readers evaluated 153 OCT images acquired from five patients, divided over two assessments (separated by 7 months) preceded by a training session (Hariri et al. 2015). The overall accuracy improved from 81.8 to 83.3% after the second assessment.

Surgical resected specimens
FF-OCT provided high-resolution images of both normal and malignant resected lung specimens up to a depth of 5-15 µm. Normal lung tissue was recognized by the typical leace-like pattern, which was formed by the alveoli and their septal walls, visible as signal-void dark areas and bright areas, respectively. Other lung components could also be identified, such as the pleura (bright signal), blood vessels, and bronchi (dull grey signal). Adenocarcinomas, characterized by their predominant lepidic growth pattern, could be really well identified. Tumor cells were also larger than normal cells, although they appeared to have a similar signal (dull grey) as normal cells (Jain et al. 2013).

Conclusion
Although the diagnostic accuracy was fairly high and OCT during bronchoscopy seems feasible in above-mentioned studies, OCT is yet not adequate as a complete replacement for tissue biopsy. However, it has the potential to be implemented in bronchoscopy procedures for diagnosis of lung tumors. Intraoperative use for margin assessment of tumor detection is yet not evaluated.

Breast cancer
Breast cancer, which is the most common type of cancer in women, is responsible for 14% of the cancer-related deaths annually (Siegel et al. 2018). In breast cancer surgery, it is extremely important that borders of the excised specimen do not contain any tumor cells, since these positive margins are associated with a higher risk of local recurrence of the primary tumor (Pleijhuis et al. 2009). Not only margin assessment, but also intraoperative staging by sentinel lymph node mapping is often performed in breast cancer patients (Lyman et al. 2005). Eight studies evaluated the use of OCT in resected surgical specimens and five studies evaluated the use of OCT for lymph node analysis.

Tumor detection surgical specimens
Diverse tumor-specific criteria were developed based on histological features. Invasive ductal adenocarcinomas, which showed infiltrating tumor cells in surrounding tissue and surrounding fibrous tumor stroma, were clearly visible (Assayag et al. 2014;Yao et al. 2017;Zhou et al. 2010). Mucinous carcinomas could be recognized by mucin with floating tumor cells, which were reflected in the OCT image. Assayag et al. proposed three FF-OCT-specific criteria for malignancy in addition to macroscopic characteristics, such as the absence of normal breast tissue structures and the presence of stellate lesions: (1) the presence of adipocytes with irregular size (fat infiltration); (2) highly scattering trabecula aspect of fibrous tissue; (3) the presence of dark grey areas surrounded by white fibrous structures; (3) (Assayag et al. 2014). Especially, white fibrous structures were suspect for tumor stroma, whereas grey fibrous structures were associated with scar fibrous tissue in benign breast lobules. Furthermore, ex vivo analysis of resection margins was performed by Nguyen et al. (2009). Thirty-seven breast cancer specimens were used for analysis, divided into a training set and a study data set. Of each specimen, 5-10 images were taken, resulting in 210 images used for the study data set and pathologic analysis by one researcher. Analysis showed a sensitivity and specificity of 100 and 82%, respectively, in tumor detection compared to histology, which is the current golden standard. Feasibility of multimodal imaging, by combining OCT with ultrasound and dye-enhanced wide-field polarization imaging, was demonstrated by Patel et al. (2013) and Curatolo et al. (2012). Computer analysis showed.
Recently, a handheld OCT camera has been developed, which has been used in two studies, that evaluated the diagnostic accuracy of the camera after obtaining the ex vivo final margins in 46 and 35 patients, respectively (Erickson-Bhatt et al. 2015;Zysk et al. 2015). In vivo imaging was feasible, although ex vivo images of the margins could be directly correlated to the histology slices, which consequently were used in both cohorts (Fig. 3). After comparison of histology with readers' interpretation, they showed a diagnostic accuracy in tumor detection at the surgical margins varying between 58 and 88%. The authors stated that this variability could be explained by the minimal training preceded by image evaluation and by the experience in studying OCT images of the reader.

Lymph nodes detection
Normal lymph nodes were characterized by a distinct capsule that was highly scattering, in comparison with the lower scattering cortex. The lymphoid follicles, which were visible as a circular texture on the OCT image, could also be clearly observed in the cortex (McLaughlin et al. 2010;Nguyen et al. 2010). Lymph node invasion was characterized by loss of normal tissue architecture, the presence of highly heterogeneous tendril-like structures, and the presence of areas with highly backscattering areas, possibly due to changes in size and texture of the cell nuclei. One study compared the diagnostic accuracy of OCT to histology, after training of three reviewers, which resulted in an overall sensitivity and specificity of 58.8 and 81.4%, respectively (Nolan et al. 2016).
To improve the capability to distinguish benign and malignant invasion of lymph nodes, parametric imaging of the local attenuation coefficient was applied in OCT images and showed promising results in two feasibility studies (McLaughlin et al. 2010;Scolaro et al. 2012). FF-OCT showed a more detailed view of the lymph nodes, of which the strong stromal reaction, caused by tumor invasion, was one of the most characterizing morphological features for lymph node invasion on FF-OCT images. FF-OCT showed an improved sensitivity of 89% and specificity of 87% compared to regular OCT after training of two independent reviewers (Grieve et al. 2016).

Conclusion
OCT was used for both tumor detection and sentinel lymph node detection in breast cancer patients. Diverse specific criteria were applied and showed high diagnostic accuracy in margin assessment compared to histology. A handheld OCT camera, which could be used intraoperatively, was also applied for margin assessment and showed promising results. This indicates that with more training and further development, OCT could be used as an additional tool for intraoperatively tumor detection. For lymph node detection, especially, FF-OCT was able to distinguish malignant invasion of lymph nodes from benign lymph nodes with high sensitivity and specificity.

Hepatopancreaticobiliary (HPB) tumors
The diagnosis of pancreaticobiliary cancers is often made by taking biopsies during endoscopic intervention. However, the current diagnostic accuracy for both pancreatic and biliary malignancies lacks sensitivity (Burnett et al. 2013;Chen et al. 2012). Moreover, in liver and pancreatic surgery, tumor positive resection margins defined as ≤ 1 mm tumorto-margin distance, are found up to 24 and 75%, respectively Verbeke and Menon 2009). Consequently, some progress in diagnostic accuracy of HPB tumors could be made. Nine studies evaluated the use of OCT both during endoscopy and in resected specimen.

OCT during endoscopy
OCT was used to distinguish malignant and benign pancreatic duct strictures both in vivo and ex vivo during routine endoscopic retrograde cholangiopancreatography (ERCP) procedures (Testoni et al. 2005(Testoni et al. , 2006a(Testoni et al. , b, 2007. Using disturbance of normal three-layer architecture with heterogenous backscattering as marker for the presence of tumor, ex vivo analysis of 100 OCT images of 10 patients showed an overall sensitivity and specificity for tumor detection of 78.6 and 88.9%, respectively (Testoni et al. 2005). Moreover, a concordance between OCT and histology for detection of a pancreatic adenocarcinoma was seen in 97.6% of the 126 images (Testoni et al. 2005). In vivo analysis resulted in a 100% accuracy for detection of neoplastic pancreatic ductal strictures (Testoni et al. 2007). Two criteria for malignant strictures were used: (1) unrecognizable layer architecture and (2) heterogeneous backscattering of signal.
Biliary duct imaging using OCT was performed in 2009 by Arvanitakis et al. (Arvanitakis et al. 2009). They used the above-mentioned criteria for detecting of malignant biliary strictures, which was accurate in 84% of the included 37 patients. OCT seemed favorable in preoperative detection of unknown biliary strictures compared to randomly taken biopsies, which resulted in a 67% accuracy in the same cohort.
Besides ERCP, endoscopic ultrasound-fine needle aspiration (EUS-FNA) is often used for taking biopsies for establishing the diagnosis of pancreatic masses. Grieve et al. (2015) evaluated the feasibility of FF-OCT in evaluation of FNA specimens acquired during EUS. Three images of pancreatic ductal adenocarcinomas (PDAC), two images of neuroendocrine pancreatic tumors, and two images of pancreatic metastases from renal cell carcinomas were included in the analyses and compared to the histology. PDAC was recognized by intense dark malignant cell clusters with irregular borders and high nuclear density. Glandular differentiation was indicated by atypical tall columnar epithelium and the presence of luminal spaces. Neuroendocrine pancreatic tumors were also easily identified by areas with neoplastic endocrine tumor cells, which appeared darker than normal pancreatic tissue. Pancreatic renal cell metastases showed a fair matching with histology. One of the two images showed good correspondence and was recognised by sheets of large cells, which compressed the vessels. Iftimia et al. (2011) used OCT for detection of several types of pancreatic cystic tumors: mucinous cystic neoplasm (MCN), intrapapillary mucinous neoplasm (IPMN), and serous cystadenoma (SCA). After developing OCT criteria for differentiating between MCNs, SCAs, and IPMNs, the investigators (a gastroenterologist, a radiologist and a pathologist) underwent training based on 20 OCT images of fresh-resected pancreatic specimens. After that, they were independently asked to evaluate 46 OCT images, resulting in a high sensitivity in distinguishing mucinous vs nonmucinous cystic lesions (95.6% for the gastroenterologist and 100% for the radiologist and pathologist).

Surgical resected specimen
Van Manen et al. (2017) evaluated the accuracy of FF-OCT in detecting pancreatic tumors in resected surgical specimens. Two pathologists were asked to evaluate 100 FF-OCT images after a training set, which resulted in a combined sensitivity and specificity of 72 and 74%, respectively, compared to histologic diagnosis. Moreover, they developed specific criteria for different types of pancreatic tumors. Especially, in case of pancreatic ductal adenocarcinoma, disorganization of glands and the presence of tumor stroma were really well visible (Fig. 4). However, due to low endogenous contrast, cell nuclei could not be visualized, whereas sometimes the collagen dominated the field of view due too much backscattering, which was mistaken for tumor stroma. Zhu et al. (2015) evaluated the feasibility of FF-OCT in resected liver specimens. Normal liver structures, such as blood vessels, bile ducts, and sinusoidal spaces, could be really well identified. Hepatocellular carcinoma was recognized by the presence of nuclear atypia and large tumor nodules separated by thick fibrous bands.

Conclusion
The role of OCT was evaluated both during endoscopy and on resected specimens of both cystic and solid tumors. OCT during ERCP showed high accuracy in detection of pancreatic or biliary strictures. Mucinous cystic lesions could be really well identified and distinguished from non-mucinous lesions. FF-OCT was feasible in ex vivo EUS-FNA biopsies, pancreatic, and liver specimens. Especially, in liver and pancreatic specimens, diverse tumor characteristics were found.

Oesophageal cancer
Oesophageal cancer, one of the most lethal cancers in the western world, is usually divided in adenocarcinoma and squamous cell carcinoma (SCC) (Pennathur et al. 2013). One of the independent risk factors for an oesophageal adenocarcinoma is a Barrett's oesophagus (BE), which is a transition of normal squamous mucosa into columnar (gastric) epithelium (= metaplasia), which could be considered as a pre-malignant stadium. Currently, most patients with BE undergo endoscopic surveillance, which is controversial. Moreover, development into dysplastic or neoplastic tissue could only be detected by taking biopsies, frequently accompanied with a sampling error (Falk et al. 1999). Thirteen  Bouma et al. performed the first in vivo study in 32 patients, who underwent routine endoscopy, and developed some characteristics of BE on OCT images (Bouma et al. 2000). Due to high scattering of the metaplastic epithelium, there was a loss of normal layered architecture. In normal oesophageal tissue, the five oesophageal wall layers (squamous epithelium, lamina propria, muscularis mucosae, submucosa, and muscularis propria) could easily be recognised by their relative difference in reflection (Bouma et al. 2000;Hatta et al. 2010;Jackle et al. 2000;Li et al. 2000). Together with the presence of inhomogeneous tissue contrast and abnormal and disorganised glands below the epithelial surface, visible as pockets of low reflectance, it is called BE. Especially, patients with BE without dysplasia or low grade dysplasia, the muscularis mucosae and submucosal layers often could be preserved (Chen et al. 2007;Cobb et al. 2010). Poneros et al. (2001) applied these criteria in a validation cohort in patients, who underwent routine upper endoscopy, which resulted in a sensitivity and specificity of 100 and 93% for BE detection, respectively. OCT was also used for detection of BE before and after radiofrequency ablation treatment. Unfortunately, in the minority of the patients (7.7%), OCT was capable to distinguish normal glands from buried Barrett's glands (Swager et al. 2016). Another study showed a 81% sensitivity and 66% specificity in detection of BE, indicating that OCT is currently not accurate enough compared to histology (Evans et al. 2007).One study evaluated the capacity of OCT for detection of oesophagus dysplasia (Isenberg et al. 2005). Normally, dysplasia is divided in low-grade and high-grade dysplasia, which consequently results in different clinical approach, i.e., resulting in oesophageal resection or not. Dysplasia was detected on OCT by reduced light scattering and loss of tissue architecture, which are currently the only known criteria. Evaluation of 314 OCT images of 33 patients by four endoscopists resulted in a sensitivity and specificity (compared to histology of the biopsies) of 68 and 82%, respectively (Isenberg et al. 2005). However, the authors did not make any difference between low-grade dysplasia, high-grade dysplasia, or neoplasia. Chen et al. (2007) more specifically described high-grade dysplasia as the presence of irregular and distorted glandular architecture and closely packed glands.

Barrett's oesophagus and adenocarcinoma
Adenocarcinomas were characterized on OCT images by a neoplastic epithelium, which contains large pockets of mucin surrounded by fibrotic and hypervascular tumor stroma (Bouma et al. 2000). Sometimes, infiltration of heterogeneous structures into the muscular layers could be seen as a feature of tumor invasion. Irregular shape and crowd of submucosal glands also advocated the presence of adenocarcinoma (Fig. 5) (Chen et al. 2007;Evans et al. 2006). Detection of adenocarcinoma in patients who underwent upper endoscopy for several reasons showed potential, with a detection rate of 95% (Zuccaro et al. 2001). Hatta et al. (2010) compared the capability of OCT for SCC invasion detection in the different layers of the oesophagus to histology. Superficial invasion into the epithelium layer was difficult to distinguish from normal oesophagus tissue. However, the researchers were able to detect the tumor invasion level with a high overall accuracy: 92.7%. Furthermore, they compared the diagnostic accuracy of OCT with high-frequency EUS in a larger cohort of 123 patients (Hatta et al. 2012). OCT showed a significantly higher accuracy than high-frequency EUS (90.1 vs 77.1%; P = 0.0046).

Conclusion
OCT has been evaluated for establishing the diagnosis of oesophageal SCC, BE, and adenocarcinoma. Diverse characteristics of BE, dysplasia, adenocarcinoma, and SCC were applied, which showed promising results. However, differentiating between normal glands and Barrett's and tumor glands was really difficult.

Colon cancer
Screening for colon cancer is currently mainly performed using fecal occult blood testing, in suspected cases followed by endoscopy (Ali et al. 2003). Adenomatous polyps, which carry risk of cancer development, are often found during colonoscopy. They can macroscopically be divided based on their growth pattern into tubular and villous polyps and microscopically they are generally classified as low-grade or high-grade dysplasia. Currently, no studies have described the use of OCT for cancer detection; however, six studies assessed the use of endoscopic OCT as a screening tool for premalignant polyp detection. Tearney et al. (1997b) first described the feasibility of OCT during colonoscopy for premalignant tissue detection. Normal colonic wall features were described by Westphal et al. (2005). The mucosa was visible as a reflective layer, including an underlying thin hyperreflective line, whereas the submucosa appeared highly variable probably caused by compression related to the colonoscopy procedure. Pfau et al. (2003) described the different features visible on OCT concerning the most common forms of polyps and showed that adenomatous polyps were significantly more disorganized and more hypo-reflective than hyperplastic polyps. Colorectal neoplasms were described as polyps with an uneven surface. On OCT, areas of malignant cells appeared bright, whereas non-cellular regions were less reflecting (Jackle et al. 2000). Quantitative analysis was performed for colorectal cancer in two studies, which showed that malignant tissue has lower scattering properties and less variation of signal intensity from the surface yielding a 78% sensitivity and 74% specificity (Ashok et al. 2013;Zhang et al. 2012).

Conclusion
The use of OCT during colonoscopy for tumor detection has not been frequently analysed. Currently, presumably quantitative analysis could be performed for both tumor and polyp detection; however, the diagnostic accuracy lacks sensitivity.

Urological cancer
Most of the time surgery is not the first choice of treatment for urological cancers.. Establishing the correct diagnosis before chemoradiotherapy is of outmost important. Therefore, OCT has been mainly used as an additional tool during biopsy taking in both five studies for prostate, six studies for renal and seven studies for bladder cancer detection, respectively.
Prostate cancer D'Amico et al. (2000) first described the feasibility of OCT for detection of tumor in prostate biopsies. The first large cohort study, performed by Dangle et al. (2009) p included 100 samples of which 10 had positive surgical margins. Using homogeneity of the prostate epithelium and the presence of prostatic stroma as assessment criteria, OCT yielded a sensitivity of 70% and specificity of 84%. More detailed information about malignant features were visualized using FF-OCT, such as the presence of irregular gland contours, infiltration of variably sized glands between benign glands, and the presence of cribriform patterns (Lopater et al. 2016). These characteristics were used during analysis by three independent pathologists, resulting in a 63% sensitivity and 74% specificity. However, a learning curve was observed, leading to a diagnostic accuracy of 82% at the end of the assessment. More recently, quantitative analysis of prostate biopsies showed significant differences in attenuation coefficient between normal and malign prostate tissue (Muller et al. 2016(Muller et al. , 2017. Linehan et al. performed the first ex vivo study on 20 resected renal specimens (Linehan et al. 2011). In general, in malignant tissue, there was a lack of normal structures; however, in some cases, tumor samples had heterogeneous areas, which mimicked normal kidney tissue. Furthermore, high sensitivities and specificities (up to 100%) for renal tumor detection were achieved in another study by Lee et al. (2012b). Three pathologists evaluated 35 OCT images of both normal and tumor resected specimens. Renal tumors were characterised by increased scattering tumor nests, which were separated by hyposcattering stroma (Lee et al. 2012b). More specific, distinguishing renal cell carcinoma and renal oncocytoma (benign renal tumor) was possible by both morphological characteristics and after quantitative analysis of the attenuation coefficient (Barwari et al.

3
2012; Lee et al. 2012b;Wagstaff et al. 2016). Computer analysis also showed the attenuation coefficient of normal renal parenchyma to be significantly differed from malignant tumors (Barwari et al. 2011(Barwari et al. , 2012. Moreover, several signatures of different kidney tumors could be recognized as described by Jain et al. (2015). After training, the uropathologist was able to distinguish benign and malignant tumors (67 FF-OCT images) with a 100% accuracy and achieved a 80% accuracy for correct subtyping of these tumors.

Bladder cancer
Seven studies evaluated the diagnostic value of OCT for the detection of cancer in bladder biopsies or resected specimens in populations varying between 21 and 116 patients (Gladkova et al. 2011;Goh et al. 2008;Hermes et al. 2008;Karl et al. 2010;Manyak et al. 2005;Montagne et al. 2017;Ren et al. 2009). Disorganized tissue layers and sub-epithelial nests of tumor cells were often found in tumorous biopsies. Using these criteria, sensitivities between 84 and 100%, and specificities between 65 and 89% for tumor detection could be achieved. Using FF-OCT, more details of tumor cells, such as the presence of large nuclei and newly formed blood vessels, appearing as small bright spots, could be visualized. After training of the reviewers, a diagnostic accuracy up to 96% could be obtained, as demonstrated by Montagne et al. (2017).

Conclusion
OCT was capable to distinguish urological tumors from benign tissue with varying accuracies. For quantitative analysis, the attenuation coefficient seems the most informative parameter.

Ovarian cancer
Ovarian cancer is a disease with a dreadful prognosis, making it the most lethal gynaecological malignancy (Siegel et al. 2018). Treatment mainly consists of surgery, especially in early-stage disease. Eight studies, evaluated the use of OCT for detection of ovarian cancer, of which two during surgery.

Resected specimen
Two studies evaluated the tumor characteristics based on qualitative analysis using OCT and FF-OCT, respectively (Peters et al. 2016;Yang et al. 2011a). Ovarian malignancies were characterized by the presence of hyperintense regions with irregular patterns, which turned out to be disorganised collagen fibers. Ovarian metastases could also be detected as shown by Stegehuis et al. (Peters et al. 2016). Metastatic tumor cells appeared light grey in a web-like organisation, resulting in a distorted ovarian cortex architecture. Furthermore, other studies analysed tumor images quantitatively, focussing on optical coefficients, which were derived from normalized histograms (Nandy et al. 2016;St-Pierre et al. 2017;Yang et al. 2011bYang et al. , 2012. After building a logistic classifier model, Nandy et al. (2016) were able to achieve a 91.6% sensitivity and a 87.7% specificity for tumor detection based on FF-OCT images. Boppart et al. (1999) first described the use of OCT, integrated in a laparoscopic system, in one resected ovary specimen, containing a serous papillary cystadenocarcinoma. The presence of cysts, which are surrounding by papillary structures, were characterizing for this tumor type. In vivo application of laparoscopic OCT was performed by Hariri et al. in 30 ovaries of 17 patients, who underwent surgical resection (Hariri et al. 2009). They described different features of ovarian tumors, such as papillary serous cystadenoma or adenocarcinoma. An adenoma was characterized by large, simple cystic structure with a sharp, regular, welldemarcated cyst lining. Complex, multicystic signal void structures surrounding by tumor stroma, which have a more heterogeneous signal intensity than normal stroma, were suspect for adenocarcinomas.

Conclusion
Several subtypes of ovarian tumors could be recognized using OCT both ex vivo and during laparoscopic surgery. However, until today no large patient cohort studies have been reported.

General conclusion and future perspectives
The introduction of OCT shows new opportunities during oncologic interventions. In this review, the role of OCT was evaluated for different tumor types. We have shown that several characteristics for both normal tissues and the different tumor types were established. In case of Barrett's oesophagus, even metaplasia could be detected. OCT could be integrated during bronchoscopic and endoscopic procedures. However, the diagnostic value of OCT was evaluated in limited studies, of which an overview is provided in Table 1. In general, it can be concluded, based on this overview, that use of OCT for guiding of biopsy location during bronchoscopy, intraoperative margin assessment in case of breast cancer surgery and for visualization of pancreatico-biliary strictures during ERCP might be useful. Especially, in case of breast cancer, a developed handheld device showed promising diagnostic accuracy, which could be valuable during surgical intervention.
The great advantage of OCT is its noninvasive, reproducible and well-tolerated characteristics for in human use. However, some drawbacks could be mentioned. First of all, OCT is a relatively new modality, which needs extensive training and setting-up new clinically relevant features by building extensive atlases for instance, before adopting into clinical practice. That also states the need for automatic feature extraction and even automatic tumor detection, which was recently performed by Marvdashti et al. (2016) who integrated an automatic classifier based on known characteristics and new features, resulting in a high diagnostic accuracy of 95.4%. However, due to the availability of several OCT systems classification is difficult, although matching of optical properties is possible (Liu et al. 2017). Standardization for signal intensities for each device, which has been done after the introduction of CT devices using Hounsfield units, would make analysis easier (Davis et al. 2018). Secondly, the imaging depth is restricted up to maximum 1-2 mm in OCT and to less than 1 mm in its high resolution variants (HD-OCT, FF-OCT). Conventional OCT with resolution between 7 and 10 µm showed less microscopic details compared to the current gold standard, histology. Nevertheless, further developing of new variants of conventional OCT with increased resolution, such as FF-OCT, could improve the quality of the images, which make tumor detection feasible into the clinical practice. However, even with FF-OCT, which provides images with resolution comparable to histology, the endogenous contrast of for instance cell nuclei is not good enough for visualization. Techniques to improve the subcellular contrast, such as dynamic FF-OCT, are being developed, enhancing both structural and functional information through the detection of the intracellular activity (Apelian et al. 2016). Recently, also a handheld FF-OCT rigid endoscope has been developed, with a faster and more sensitive camera integrated than current FF-OCT systems, which paves the way for intraoperative use (Benoit a la Guillaume et al. 2016). Finally, sometimes, signal of highly backscattering tissues, such as collagen vessels or tumor stroma, disturbed the OCT images and made distinguishing benign and malignant tissue from each other complicated. On the other hand, it was shown that stroma alignment is significantly different between benign and malignant tissue for pancreatic cancer (Drifka et al. 2015) and breast cancer (Bredfeldt et al. 2014), and could, therefore, contribute to better detection of tumors on OCT images. Such information can be possibly obtained using polarization-sensitive OCT, which in addition to tissue microarchitecture can provide images of tissue birefringence (Marvdashti et al. 2016). Hariri et al. showed that Polarization Sensitive OCT (PS-OCT) has a potential to help in differentiation between lung tumor and fibrosis and Kiseleva et al. used PS-OCT to diagnose mucosal pathologies in in-vivo human bladders (Hariri et al. 2013;Kiseleva et al. 2015). In addition, other functional OCT variants, such as Doppler-OCT that visualizes tissue vasculature and spectroscopic OCT that differentiates tissue types based on signal attenuation, as well as multimodality approaches are currently a topic of research (Barwari et al. 2011(Barwari et al. , 2012Mavadia et al. 2012;Welge and Barton 2017).
In conclusion, OCT showed promising results in tumor detection and with the development of novel probes allowing integration in bronchoscopy, flexible or rigid endoscopy, needles (Lorenser et al. 2013), handheld cameras and tethered capsules (Gora et al. 2013;Liang et al. 2016) could add important value during both preoperative diagnosis as well for intraoperative use in obtaining tumor free resection margins in the nearby future.

Compliance with ethical standards
Conflict of interest Claude Boccara is one of the founders of LLTech and is a part-time LLTech employee. Emilie Benoit is an employee of LLTech. Labrinus van Manen, Jouke Dijkstra, Alexander L. Vahrmeijer, Michalina J. Gora, and J. Sven D. Mieog declare that they have no conflict of interest.
Ethical approval This article does not contain any studies with human participants or animals performed by any of the authors.
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creat iveco mmons .org/licen ses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.