Assessment of Sentinel Node Biopsies With Full-Field Optical Coherence Tomography

Abstract

Current techniques for the intraoperative analysis of sentinel lymph nodes during breast cancer surgery present drawbacks such as time and tissue consumption. Full-field optical coherence tomography is a novel noninvasive, high-resolution, fast imaging technique. This study investigated the use of full-field optical coherence tomography as an alternative technique for the intraoperative analysis of sentinel lymph nodes. Seventy-one axillary lymph nodes from 38 patients at Tenon Hospital were imaged minutes after excision with full-field optical coherence tomography in the pathology laboratory, before being handled for histological analysis. A pathologist performed a blind diagnosis (benign/malignant), based on the full-field optical coherence tomography images alone, which resulted in a sensitivity of 92% and a specificity of 83% (n = 65 samples). Regular feedback was given during the blind diagnosis, with thorough analysis of the images, such that features of normal and suspect nodes were identified in the images and compared with histology. A nonmedically trained imaging expert also performed a blind diagnosis aided by the reading criteria defined by the pathologist, which resulted in 85% sensitivity and 90% specificity (n = 71 samples). The number of false positives of the pathologist was reduced by 3 in a second blind reading a few months later. These results indicate that following adequate training, full-field optical coherence tomography can be an effective noninvasive diagnostic tool for extemporaneous sentinel node biopsy qualification.

Introduction

Axillary lymph node resection commonly accompanies mastectomy or lumpectomy to determine metastasis of breast cancer. In patients with early-stage breast cancer, presence or absence of malignant deposits in axillary lymph nodes is the most important prognostic indicator. ¹ Recently, sentinel lymph node biopsy (SLNB) has replaced axillary lymph node dissection (ALND) as the preferred procedure for staging metastasis in the lymph node chain. Absence of carcinomatous deposits in the sentinel node is a strong indicator for absence of metastasis in the axillary chain, so that a single node may be removed instead of the full chain, thus reducing the severity of the surgical procedure and the incidence of lymphedema. ² Lymphedema, in which swelling at the site of removal causes lack of sensation in the arm, can occur in 20% of patients with a relative risk of 3 to 5. ^3,4 Lymphedema can be associated with pain, numbness, and sometimes infection. In rare cases, it can lead to cancers such as lymphangiosarcoma or hemangioendothelioma. Intraoperative techniques such as macroscopic analysis and palpation, frozen section (FS) analysis, cytological smears, touch imprints, or molecular pathology are performed on the SLNB to guide the surgeon in removing the axillary chain, in case metastasis should be discovered in the sentinel node.

Frozen section is currently a common intraoperative method. In Europe, a survey in 2003 revealed that 60% of the pathology units dealing with sentinel lymph nodes (145 of 240) performed intraoperative assessment, and among them, 89% used FS. ⁵ Furthermore, the American Society of Clinical Oncology guideline recommendations for SLNB in early-stage breast cancer state that, although FS carries the risk of significant destruction of potentially diagnostic tissue, it may be the most desirable intraoperative assessment for experienced surgeon/pathologist teams. ⁶ Frozen section can give a preliminary diagnosis in around 20 minutes, based on 1 to 3 levels analyzed for each SLNB. ⁷ Numerous different protocols exist; however, FS analysis on lymph nodes is challenging due to the presence of fatty tissue in case of adipous involution of the node. It tends to be highly operator dependent due to the difficulty in slicing in the presence of fat and moreover consumes tissue. Cytological examination provides a quick (10-15 minutes) diagnosis of metastasis and tissue is not consumed. However, metastatic architecture is lost and it is difficult to identify rare events against a background of lymphocytes, activated lymphoid cells, and histiocytes during the brief time permitted for intraoperative examination. Finally, molecular pathology is a highly sensitive method of discovering macro-metastasis (>2 mm) and micro-metastasis (0.2-2 mm) as well as isolated tumor cells (ITCs; <0.2 mm) in lymph node tissue, based on a correlation between the amount of products of polymerase chain reaction detected and the size of the metastatic deposit. Its sensitivity surpasses that of FS analysis and histology, but the major drawback is that it completely consumes the tissue, it is expensive, time consuming (40 minutes), and characteristics of carcinomatous cells within the sample are lost due to the destructive nature of the analysis process. The latter drawback prevents comparison of the metastasis with the primary tumor and the possibility to perform immunohistochemical testing. Besides, the prognostic value of removing lymph nodes containing micrometastasis or ITCs is debatable. ^{2 –9}

Imaging modalities such as ultrasound, positron emission tomography, computed tomography, and magnetic resonance tomography may be used to image lymph nodes in vivo. ¹⁰ However, the moderate resolution of these techniques allows measurement of the size and shape of the node without revealing any morphological detail. Tumorous nodes are indeed larger in size than normal nodes but so are benign reactive nodes due to inflammation. These techniques are not able to distinguish between metastastic and reactive nodes, resulting in removal of benign tissue. ^11,12 Optical imaging techniques offer a higher resolution with a trade-off of reduced penetration depth. Light microscopy ¹³ and confocal ¹⁴ techniques have been investigated for clinical applications. In confocal microscopy, sectioning capacity is of the order of a few micrometers but relies on objectives with a very large numerical aperture. Such optical configuration limits the field of view and the penetration depth, as it makes the system more sensitive to tissue optical defects. In addition, the technique often requires the use of a fluorescent contrast agent. For ex vivo imaging, the contrast agent modifies the tissue, precluding the subsequent use of the tissue for gold standard histological processing. For extension to in vivo use, the number of fluorescent contrast agents suitable for use by humans is limited. Studies have also investigated the use of reflectance microscopy for various medical applications yet not for the analysis of SLNB. Acquisition time (9 minutes for 12 × 12 mm) was compatible with intraoperative use. ¹⁵

Optical coherence tomography (OCT) is an imaging modality analogous to ultrasound but using light reflected from structures inside tissue. ¹⁶ Five main studies for in vivo transcapsular imaging as well as ex vivo analysis of lymph node with OCT have been published. ^{17 –21} They were carried out with commercial and laboratory setups demonstrating axial resolution ranging from 2 to 12 μm and transverse resolution ranging from 10 to 15 μm. A few samples were imaged (18 samples maximum). Normal structures were recognized and reactive or metastatic tissue could be identified with its accompanying increase in scattering. A review on the use of OCT for breast cancer, including its use in lymph nodes, was recently published. ²²

Full-field OCT (FFOCT) ²³ is a variant of conventional OCT, which relies on a Linnik interferometry configuration. Two-dimensional en face images are captured on a camera and 3-dimensional (3D) data sets may be obtained by scanning in the depth direction. This configuration and the use of broadband thermal light sources allow for higher axial and transverse resolution of the order of 1 µm. No contrast agents are required as contrast is entirely endogenous, as in conventional OCT. These characteristics make the technique better adapted to pathology applications when compared to conventional OCT, since 3D high resolution is compatible with visualization of cellular details, and since the en face acquisition geometry mimics current microscopy examination of histology slides. The FFOCT provides images resembling reflectance confocal images. However, axial sectioning is here uncorrelated from transverse resolution and allows a better balance with field of view and penetration depth. As compared to confocal systems, FFOCT further holds the advantage not to require lasers or fast scanning elements; the technique can be integrated in a simple setup. The FFOCT has already been evaluated in the pathology laboratory in various preclinical studies on human skin, ^24,25 pulmonary, ²⁶ urologic, ²⁷ and gastrointestinal tissue, ²⁸ retina and cornea, ^29,30 and brain. ³¹ Recently, a preclinical study was performed on breast tissue samples ³² and achieved sensitivity of 92% and specificity of 77%. Scanning time (less than 10 minutes for 1 cm²) was comparable to figures published for ex vivo reflectance confocal microscopy as mentioned earlier.

The aim of this study is to assess the effectiveness of FFOCT to detect metastasis in nodes, in view of a possible ex vivo application in the analysis of SNLB as an alternative to FS analysis.

Materials and Methods

A total of 38 patients with breast cancer at Tenon Hospital, Paris, were enrolled in the study. Written and oral consent was given according to the procedure at Tenon Hospital.

Forty-eight sentinel lymph nodes were examined with FFOCT. The imaging protocol is illustrated in Figure 1. Fresh sentinel nodes were bisected along the long axis, with one half destined for intraoperative FS analysis and the other half for FFOCT imaging. A further 23 fresh half lymph nodes of the axillary chain, removed after a positive analysis of the sentinel nodes, were also imaged by FFOCT after dissection of the fresh unfixed adipose tissue and bisection. All the imaged specimens were subsequently preserved and processed for routine histology in the pathology laboratory. The histological analysis was performed in the department by pathologists external to this study. In all cases, FFOCT imaging was performed minutes after excision by a technician from the pathology laboratory, who followed a training of 1 hour before the start of the study. A previous study ³² verified that the FFOCT imaging procedure does not alter the tissue in any way that affects the histological diagnosis and that it is perfectly safe to perform an FFOCT imaging step prior to histology.

Figure 1.

Imaging protocol for fresh sentinel nodes and lymph nodes from axillary lymph node dissection (ALND).

The FFOCT instrument used in this study is a commercial system from LLTech, France. It has been described previously. ³² Illumination is provided by a halogen source, whose short coherence length leads to an axial resolution of 1 µm. The full field is illuminated and images are captured on a CMOS camera. The interferometer arms hold a matched pair of microscope objectives in the Linnik configuration; X10 water immersion objectives with 0.3 numerical aperture lead to a lateral resolution of 1.6 µm. Penetration depth is highly sample dependent: in lymph nodes, it is of the order of 200-300 µm. A 1 cm × 1 cm stitched field is captured in less than 10 minutes. A 200 µm × 1 mm × 1 mm 3D stack is also captured in less than 5 minutes.

In this study, the half nodes were individually placed in the system chamber with saline water and gently pressed against the optical window provided, in order to flatten the surface and reduce optical aberrations. Immersion oil was applied on the optical window (hence not in contact with the sample), and the sample in its holder was inserted in the FFOCT instrument. The acquisition process required no manual alignment. The user followed the instructions on the acquisition software, which included acquiring a macro image, waiting for the automatic zero-position adjustment of the system, and defining the area to be acquired on the macro image. In this study, the whole lymph node sections were imaged. The depth of imaging was set to 20 μm, and image averaging for noise reduction was set at the default value of 40 averaged images per frame. Attention was paid to orientate the sections after imaging such that histological slides would be prepared from the same surface.

The first blind analysis of the images was performed by pathologist MA who was not involved in the histological analysis of the specimens. She had been involved, however, in a previous study on breast tissue ³² and was therefore familiar with FFOCT images. Lymph node images were collected into randomized and anonymized batches of 15 to 20 samples. The number of imaged specimens examined by MA was n = 65, as the image collection was not completed at the time of the end of the analysis. Pathologist MA reviewed the FFOCT images, identifying key features of the sentinel node architecture and noting the order in which these features were analyzed in order to form a diagnosis. Each sample was categorized as normal or metastatic, and notes were taken on further observations or precisions on how the pathologist arrived at a particular diagnosis. The corresponding histological slides were then observed and the diagnosis confirmed before another batch of images was blindly analyzed. Thus, pathologist MA obtained regular feedback during the blind analysis after each batch of images and could adjust the reading criteria. At the end, the diagnoses made were compared to those obtained on the histological slides on the same node half as was imaged with FFOCT. Note that diagnosis on this half did not necessarily correspond with the final diagnostic report on the whole node as the metastasis may have been contained in the other half.

A nonmedically trained person (ED), familiar with FFOCT images in other organs, was the second person to perform blind diagnosis (normal/malignant) on n = 71 FFOCT images. Training was first carried out using the FFOCT reading criteria defined by pathologist MA on random subsets of the FFOCT images. Then, during the blind diagnosis, ED was informed, following each diagnosis response, of the pathologist’s verdict and comments on the FFOCT image, the pathologist’s diagnosis from the corresponding histology slide, as well as the FS and immunohistochemical results, and the diagnosis stated in the final patient report (ie, based on all of the above-mentioned elements combined, bar FFOCT). This provided her with immediate feedback on her responses and allowed her to learn to recognize pathology in the nodes.

Finally, pathologist MA performed a second blind reading of a set of false-positive cases, at a time far removed (between 6 months to 2 years later) from the initial diagnoses, and randomized among other cases.

Results

For each node, low magnification followed by high magnification analysis of the FFOCT images was performed, searching for features of normal lymph nodes. Figure 2 shows an example of normal lymph node, where the following features are observed: perinode fat tissue, capsule, collagen trabeculae, lymphoid follicles, and hilum, such as previously described. ³³

Figure 2.

Normal sentinel node in full-field optical coherence tomography (FFOCT; A) with perinode fat tissue (arrowhead) and tangential view of hilar region with adipocytes (arrow); corresponding histological slide (B). C and E, Zoom at high magnification shows cortex containing lymphoid follicles (dark grey aspect with interspersed white collagen fibers arrow) and surrounded by capsule and collagen trabeculae (arrowhead); D and F, adipocytes in hilum. Scale bars show 1 mm (A), 200 μm (C and E), and 100 μm (D and F).

Figure 3 shows an example of a lymph node invaded with ductal carcinoma. A dense, highly scattering bright white appearance is observed, highlighting a dense collagen mesh. The tissue architecture is significantly altered and the lymphoid follicles are no longer visible. This strong stromal reaction is associated with metastasis, hence it was used as the primary reading criterion for malignancy in the FFOCT images. An especially careful examination was made of the lymph node periphery, where there is a high prevalence of metastasis. ³³

Figure 3.

Full-field optical coherence tomography (FFOCT) image (A) of a sentinel node invaded with ductal carcinoma. B, Zoom at higher magnification shows the limit of the macrometastasis (arrow), showing the bright white aspect of highly scattering dense collagen meshwork, as confirmed by corresponding histological slide (C). Scale bars show 500 μm (A) and 200 μm (B and C).

Lobular carcinomas were revealed by the different aspect of the lymphoid tissue: In this case, the node showed a homogeneous and granular aspect without well-defined lymphoid follicles (Figure 4).

Figure 4.

Full-field optical coherence tomography (FFOCT) image (A) of a sentinel node invaded with lobular carcinoma. B, Zoom at higher magnification shows the limit of the macrometastasis (arrow), with homogenous and granular appearance, as confirmed by the corresponding histological slide (C). Scale bars show 500 μm (A) and 100 μm (B and C).

Statistical analysis performed on the blind diagnosis of the readers revealed sensitivity of 92% and specificity of 83% for the pathologist and 85% sensitivity and 90% specificity for the nonmedical expert, on a first benign/malignant diagnosis (ie, on diagnosis performed while the image features were being identified and the image atlas constructed). The results are summarized in Table 1.

Table 1.

Results From First Blind Analysis of Lymph Node FFOCT Analysis.

	Pathologist MA	Nonmedical Imaging Expert ED
True positives	11	11
True negatives	44	52
False positives	9	6
False negatives	1	2
Total number of specimens	65	71
Sensitivity	92%	85%
Specificity	83%	90%

Abbreviation: FFOCT, full-field optical coherence tomography.

A second look at a set of false positives randomized among other samples, following further training on the image atlas, resulted in a reduction in false positives by a factor of 3. Although it was the same pathologist viewing the same data, the time elapsed and the fact that she views up to 20 lymph nodes per week meant that she was certain that the images were no longer individually remembered. This reduction would predict an overall improvement in the specificity from 83% to 94%.

Discussion

The FFOCT efficacy values for a benign/malignant diagnosis on the lymph nodes are 92% sensitivity and 83% specificity for pathologist MA (predicted specificity of 94% after training) and 85% sensitivity and 90% specificity for imaging expert ED.

The analysis of the results helps to identify reasons for false negatives and false positives. There were very few false negatives: 1 for MA and 2 for ED. For one of them, a micrometastatic deposit was visible on the histological slides, which was not readily visible on the FFOCT image (see Figure 5). This indicates that high magnification analysis of areas of lymphoid appearance should systematically be performed to check for fibrosis. For the other false-negative case, the readers admitted afterward it could have been corrected with a closer look at the image and the presence of the collagen mesh.

Figure 5.

False negative—area of dark gray appearance is identified as lymphoid tissue based on grey level aspect in low magnification FF-OCT (A). Subsequent high magnification zoom (B) reveals moderately scattering fibrosis (asterisk) at the location of a micro-metastasis diagnosed on the histological slide (C). Scale bars show 2 mm (A) and 500 μm (B and C).

False positives were also rare. The pathologist training during the first look analysis led the reading criteria to be refined. In particular, the fact that collagen fibers are structured around nodules is a sign of benignity, while a uniform, nonstructured honeycomb appearance reveals fibrosis. Fibrosis is generally not suspicious when near the hilum because of the presence of numerous vessels. If however the fibrosis also appears near the capsule, then it is likely that the node is invaded. ³³ Furthermore, the presence of vessels can be misleading, as their walls may appear hyper-reflective on the FFOCT images, and they may cause alteration of the surrounding tissue. These effects explain some false-positive cases, as illustrated in Figure 6. Reading a sufficient number of images was essential to gain a feel for the location, structure, and threshold of abnormal density of collagen and recognize all normal features. Indeed, the number of false positives was significantly reduced on the second blind reading. It is interesting to note that the stromal reaction was also a criterion identified for the analysis of breast tissue. ³²

Figure 6.

Full-field optical coherence tomography (FFOCT) image of lymph node with false positive diagnosis (A). Zoom (B) shows high scattering area, thought to be suspect tissue, and subsequently revealed as vessels (arrows) on histological slide (C). The surrounding tissue is homogenized and follicular structures are no longer visible. Scale bar shows 500 μm (A) and 200 μm (B and C).

It is expected that sensitivity and specificity values on a larger cohort in the context of subsequent clinical studies, that is, following adequate pathologist training, should lie close to the second-look results obtained here. This first clinical study has allowed us to create an image atlas for pathologist training on FFOCT imaging of sentinel nodes. Training on this atlas will be indispensable in the next steps of clinical studies on sentinel node imaging with FFOCT. Although differences in aspects could be visualized between lobular and ductal carcinomas, this study did not measure the performance in differentiating between tumor types. The sample cohort was too small to provide significant statistics on this measure. A larger cohort could also be used for such evaluation.

Overall, the results are very promising, in particular the second-look analysis: sensitivity and specificity over 90%. They can be compared to the performance obtained with FS analysis. The FS analysis of the sentinel lymph nodes included in this study yielded a sensitivity of 75% and a specificity of 100%. These values are in line with typical values found in the literature. ^{34 –38} The total population was however smaller (49 samples), since the FS analysis was performed on sentinel nodes while the FFOCT imaging could be performed on a bigger node population comprising SLN and nodes from ALND. Furthermore, different sentinel node halves were analyzed with each technique. The FFOCT thus appears to be a comparable technique, if not superior, to FS. It further holds the advantage of simplicity, as only one instrument is required, instead of a cryostat microtome with coloration material and a microscope. No expertise is necessary for the image acquisition, contrary to FS preparation, where technical expertise is required to obtain good-quality slides. Furthermore, the learning curve is fairly short as highlighted by the results of this study. Currently, the FS preparation time is very similar to the FFOCT image acquisition time of a sentinel node section: about 10 minutes for a 1 cm diameter node. Camera technology developments for FFOCT that are currently in progress are expected to dramatically reduce the acquisition time by a factor of 10 to 40, hence reducing the acquisition time for a node section to under a minute. As with FS analysis, FFOCT analysis of single node sections may miss micrometastasis. In fact, the American Society of Clinical Oncology recommends cutting the SLNB in sections no thicker than 2 mm to analyze all sections. ⁷ Such methodology would imply an increased acquisition time for the FFOCT analysis: for example, 5 sections for a 10-mm SLNB. With faster technology such as that mentioned earlier, the procedure would appear to be reasonable for intraoperative use. Such a fast system would need to be tested, in particular in real clinical conditions, to ensure that the whole imaging and analysis process, including sample preparation, matches clinical requirements, not only in terms of information retrieved but also with regard to time constraints.

When compared to cytological examination and molecular pathology, FFOCT holds the advantage of preserving sample architecture, hence it is still usable for standard histology processing. The FFOCT cannot achieve the very high sensitivity offered by molecular pathology. However, the prognostic value of removing lymph nodes containing micrometastasis or isolated tumor clusters is uncertain. ^2,8,9 Finally, the technique is faster and less expensive than molecular pathology.

Looking further into the possibilities offered by FFOCT, it could appear relevant to use the in-depth imaging mode that has been demonstrated on other organs. ²⁴ One problem that remains with the gold standard histology process on lymph nodes is sampling density in the tissue. As each slice implies increased cost, slices are selected at a density that may miss areas of metastasis or mistake the size of the metastasis if the slice does not pass through the largest dimension of the metastatic region. In this way, FFOCT could offer an advantage in comparison to gold standard histology as images could be obtained over several hundreds of micrometers with no additional cost. This diagnosis strategy could be investigated once faster camera technology becomes available. Finally, in vivo use of FFOCT for sentinel lymph node imaging could also be explored. Endoscopic FFOCT has been demonstrated in human skin in vivo with a rigid probe. ³⁹ Further miniaturization development is underway to open its application to other tissues. The possibility to image lymph nodes without bisection, that is, through the capsule, would have to be validated.

Conclusion

This study evaluated the performance of FFOCT analysis of axillary lymph nodes in the field of breast carcinoma. It showed very good sensitivity and specificity values (between 83% and 94%) on the blind benign/malignant diagnosis by a pathologist and a nonpathologist. Furthermore, the technique is about as fast as FS analysis but does not consume tissue. Therefore, we anticipate that analysis of ex vivo sentinel node biopsies with FFOCT may have potential to replace FS analysis. In future, endoscopic FFOCT could also be investigated to provide surgeons with the opportunity to perform real-time in situ in vivo sentinel node analysis without the need for biopsy.