Open Access
Add to BibSonomy
Abstract
Ovarian cancer has the highest mortality rate among all gynecological cancers, containing complicated heterogeneous histotypes, each with different treatment plans and prognoses. The lack of screening test makes new perspectives for the biomarker of ovarian cancer of great significance. As the main component of extracellular matrix, collagen fibers undergo dynamic remodeling caused by neoplastic activity. Second harmonic generation (SHG) enables label-free, non-destructive imaging of collagen fibers with submicron resolution and deep sectioning. In this study, we developed a new metric named local coverage to quantify morphologically localized distribution of collagen fibers and combined it with overall density to characterize 3D SHG images of collagen fibers from normal, benign and malignant human ovarian biopsies. An overall diagnosis accuracy of 96.3% in distinguishing these tissue types made local and overall density signatures a sensitive biomarker of tumor progression. Quantitative, multi-parametric SHG imaging might serve as a potential screening test tool for ovarian cancer.
© 2022 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Ovarian cancer is the most deadly gynecologic cancer which accounts for 2.5% of all female malignancies but 5% cancer deaths [1]. Epithelial ovarian cancer (the most common type of ovarian cancer, making up 90% of the cases) can be broadly classified into two types based on morphologic and clinical features. Type I is low-grade benign tumor with slow growth rate and better prognosis which is diagnosed at early stage I/II, including low-grade serous, low-grade endometrioid, clear cell, mucinous and transitional (Brenner) carcinoma. Type II at later stage III/IV includes high-grade serous, high-grade endometrioid, undifferentiated tumors and carcinosarcomas, showing aggressive growth and high morality [2]. Different types of ovarian cancer are accompanied with different clinicopathological features, treatment responses, and prognosis, therefore associated with different treatment options. Thus, automated, noninvasive, highly-accurate auxiliary classification and diagnosis technique for ovarian cancer is urgently needed for the determination of the prognosis and consecutive treatment schemes.
At present, the screening test of ovarian cancer mainly depends on the combination of ultrasound imaging and blood test of CA125, yet both with limitations [3–5]. Nonneoplastic entities such as functional cysts, tubal and inflammatory diseases may have similar ultrasound appearances to the tumor and limited by the resolution of ultrasound imaging, small lesions on the ovaries could hardly be captured [3,4]. Sensitivity and specificity of CA125 to distinguish benign pelvic masses from ovarian cancer are insufficient and the understanding of the biochemical structure of CA125 is incomplete [5]. Therefore, it is necessary to provide new perspectives for a better understanding of tumor progression and screening techniques of ovarian cancer.
Extracellular matrix (ECM) is a highly-organized network of macromolecules with distinctive physical, biochemical, and biomechanical properties, playing key physiological and pathological roles in many life processes [6]. Collagen fiber is the main component of ECM and continuous undergoes remodeling caused by cell-matrix interactions such as cell migration, differentiation and proliferation [7,8]. Moreover, collagen fibers are the most abundant proteins of the ovarian interstitial matrix and several studies have demonstrated that the alterations of the mechanical and the biochemical states of collagen fibers are related to the degree of malignancy and neoplastic activity [9–12], which makes structural characteristics of collagen fibers potential biomarkers for the ovarian cancer progression. Second harmonic generation (SHG) is a second order coherent process in which two low-energy photons are converted into one photon with exactly twice the frequency (half the wavelength), and has emerged as a powerful nonlinear optical contrast mechanism for the label-free, non-destructive and deep-sectioning biological and biophysical visualization applications [13,14]. The SHG imaging technology enables submicron resolution and 3D imaging of tissues several hundred microns thick with the image contrast produced purely from endogenous species. Restricted by the second order nature of SHG, environment for harmonophores that can be imaged must be non-centrosymmetric on the size scale of emission wavelength. The fibrillar structure of collagen can produce high SHG contrast, making SHG imaging a strong tool for collagen visualization [15–17]. However, even as high-resolution 3D images of collagen fibers become more readily accessible, there is a lack of quantitative analysis algorithms for their morphology and organization to map the subtle and dynamic remodeling of collagen fibers during disease progression, especially in a 3D context, with only a few notable exceptions [18–21].
In this work, we developed a new optical metric, named local coverage (LC), to quantify the morphologically localized distribution of collagen fibers in both 2D and 3D contexts in an automated manner and applied it to the 3D image stacks of normal, benign and malignant ovarian tissues acquired from SHG imaging. We found differences in the LC value and distinct properties regarding the depth-dependent heterogeneity in collagen distribution of ovarian tissues belonging to different groups. Finally, we performed canonical linear discriminant analysis to achieve ovarian cancer diagnosis with a combination of multiple parameters including mean of local coverage, depth-dependent variation of local coverage and overall density. A high classification accuracy was acquired which proved the multi-parametric, collagen signatures-based analysis as a potential tool for auxiliary diagnosis of ovarian cancer.
2. Materials and methods
2.1 Sample preparation
With the approval of the institutional review board, we collected human ovarian tissues of normal, benign and malignant (stage IV high-grade serous tumor) types surgically removed from 23 patients (12 malignant and 11 benign) from the Affiliated Cancer Hospital of Fujian Medical University. Ovarian tissues were washed with phosphate buffered saline first to remove residual blood and then placed on glass slides for SHG imaging. Three fields from each tissue were imaged by SHG, with these fields marked after the imaging procedure. Ovarian tissues were then prepared for hematoxylin-eosin (H&E) staining (with typical images from normal, benign and malignant tissues shown in Fig. S1) and then the marked fields subject to SHG imaging were confirmed by experienced pathologists according to H&E readouts. If the imaged fields were not representative of the corresponding tissue type, then these fields would not be used for the subsequent characterization. If all the three fields from a tissue sample were not well confirmed as representative ones, then this tissue piece would not be considered. Finally, a total of 27 tissue samples were included in our study, including 9 malignant (from 9 malignant patients separately), 9 benign (from 9 benign patients separately), and 9 normal (obtained from non-cancerous areas from 5 malignant and 4 benign patients) ones. These fresh tissues were from live patients, sectioned without fixation, and used for 3D SHG imaging. It should be noted that frozen tissues can be used as well for assessment of collagen fiber organization [22]. All the tissues were sliced laterally with a thickness of ∼50 µm for imaging.
2.2 SHG imaging
SHG images of ovarian tissues were acquired by a commercial laser scanning microscope (Zeiss, LSM 880) equipped with mode-locked Ti:Sapphire femtosecond laser (140 fs, 80 MHz) that was tunable in the range of 680–1080 nm (Coherent, Chameleon Ultra II). The circularly polarized light (810 nm, a typical wavelength for SHG excitation [22]) was focused onto the tissue with a plan-apochromat objective (20×, NA = 0.8, Zeiss, Jena Germany). The SHG signals at 405 ± 15 nm were then collected in the backward direction using the same objective and were finally detected using a photomultiplier tube with a band-pass filter. Compared with forward SHG, backward SHG was widely used for collagen analysis and could visualize more small-diameter, segmental fibrils [23] resulting from tumor-related collagen decomposition and synthesis during cancer progression, enabling more accurate characterization of collagen proportion and distribution in this study. Laser power at the sample was measured to be ∼50 mW for all the tissues, and these samples might start to burn when the laser power approached ∼100 mW. To account for the increased tissue scattering as the laser went deeper into the sample, an automated power correction mechanism provided by the commercial software was activated during the 3D imaging. By acquiring the two-photon excited fluorescence intensity profile of a 0.1 µm diameter fluorescent bead along the lateral and axial direction, respectively, the lateral and axial resolutions of this system, provided by the full-width-at-half maximum of the Gaussian fit to the respective fluorescence intensity profile, were about 0.57 × 1.83 µm (lateral × axial). Therefore, 3D SHG image stacks were obtained with an interslice distance of 0.634 µm (according to Nyquist criteria for 2-fold over sampling) and a field-of-view of 425.1×425.1 µm, which contained 512×512 pixels with pixel dwell time as 2.05 µs. These settings guaranteed suitable field of view and comparable resolution levels as reported in previous studies to resolve collagen fibers [22,24]. Meanwhile, we collected the two-photon excited fluorescence image of fluorescein solution at the same settings (such as the excitation wavelength at 810 nm and the objective), and used it to correct the acquired SHG images to guarantee uniform field. Imaging of these tissues was performed right after the tissue excision.
2.3 Overview of the local coverage metric
The local coverage (LC) metric developed in this study quantified the pixel-wise localized distribution of collagen fibers with normalized value ranging from 0 to 1, with a higher value corresponding to morphologically higher collagen coverage in a local area. The calculation process of LC is shown in Fig. 1. The SHG image of collagen fibers [Fig. 1(a)] firstly underwent image segmentation through intensity thresholding to identify collagen pixels from background [Fig. 1(b)]. Threshold determination was very important for local coverage and overall density quantifications. To account for slight differences in experimental conditions during imaging of different tissues, we first applied intensity normalization for all the stacks. Owing to the second order nature and purely endogenous contrast of SHG, background signals were very weak. Therefore, we set the threshold to be 0.1 for these normalized image stacks and generated binary mask accordingly for the calculation of local coverage and overall density. Then $m \times m$ pixel windows [e.g., 11×11 for the case in Fig. 1(c)] centered at all collagen pixels were generated containing information from a local area of the binary matrix. For each window, LC was calculated through dividing the number of collagen fiber pixels (with a binary value of 1) by the total number of pixels ${m^2}$ in the window, and the LC value was then assigned to the center pixel of the window [marked in red in Fig. 1(c)]. To deal with the edge issue, the image was expanded according to the size of the assessing window and those extended pixels were padded by the intensity of the boundary pixels (Fig. S2). In this manner, LC matrix was acquired characterizing all the collagen pixels with the localized distribution information. In order to demonstrate the LC more intuitively, color-coded map was obtained based on the pixel-wise LC value, with redder hues representing higher coverage and denser distribution of collagen in the local area [Fig. 1(d), left]. LC value distribution histograms from three distinct regions marked by dotted white box were consistent with observations [Fig. 1(d), right]. LC values were close to 1 for the nearly full coverage of collagen fibers in region 1, yet lower in regions 2 and 3 due to sparser distribution of collagen fibers.
Fig. 1. Flowchart of local coverage quantification. (a) SHG image of collagen fibers. Scale bar: 50 µm. (b) Binary matrix acquired from SHG signal thresholding. (c) Brief schematic for pixel-wise local coverage calculation. (d) Color-coded maps of local coverage (left, see color bar for corresponding local coverage value) and histograms of local coverage values from three representative regions (right).
During the calculation process of the LC metric, window size ($m$) was a critical parameter which would affect the LC result. To assess the influence of window size on LC determination, we carried out LC characterization for two SHG images with relatively dense and sparse fiber distribution [ Fig. 2(a) and 2(b)]. Mean of LC (MLC) for all the fiber pixels in the image was calculated to reflect the global collagen distribution in the image. Generally, larger window size would lead to lower MLC because it might contain more pixels from background [Fig. 2(c)]. However, for two images with dense and sparse collagen distribution [Fig. 2(a) and 2(b)], the difference of MLC remained highly consistent as the window size varied [Fig. 2(c)], which promised that as long as the same localized window was applied to different groups in a certain application, LC metric could accurately resolve the distribution of collagen fibers among different groups. Generally, for a typical SHG image of 512×512 pixels, we chose a window size of 50×50 pixels to represent a local region for LC characterization.
Fig. 2. Assessment of the robustness of the LC metric regarding its dependence on the window size. (a), (b) Two groups of SHG collagen images (left) and corresponding LC maps (right). Different window sizes are marked with dotted rectangle using different colors to show the sizes relative to the image. Scale bar: 50 µm. (c) The mean local coverage (MLC) of (a), (b) along with their difference using different window sizes for LC quantification.
In recent years, confocal and multiphoton microscopies, especially SHG, were able to capture the fine filamentous structure of biological tissues in a 3D context [25,26], which made it essential to extend the LC algorithm to the 3D format. The difference between 3D and 2D technique was to generate 3D windows which contained local regions centered at all the collagen voxels. In this way, the axial information of the 3D stack, in addition to the lateral one, was considered in LC calculation. Besides direct extension from 2D to 3D format, LC calculation was highly time-efficient, which resolved a typical biomedical 3D image stack with a size of 512×512×100 voxels with less than 30 s (Fig. S3).
2.4 Overall density calculation
The overall density (OD) metric quantified the total proportion of collagen fiber pixels over the total pixel number in a 2D image or a 3D stack based on the image segmentation result, which was an important characteristic for the analysis of collagen fibers [27–29].
2.5 Statistical analysis
For ovarian tissue samples of three different groups (normal, benign and malignant), a one-way ANOVA post-hoc Tukey HSD test was performed using JMP 12 to assess differences of optical metrics. Results are considered significant at p < 0.05. To verify the ability of the three collagen characteristic metrics (mean local coverage, depth-dependent variation of local coverage, overall density) in distinguishing the three ovarian tissue groups, we performed canonical linear discriminant analysis using SPSS, and original classification accuracy (OCA) and cross-validated classification accuracy (CVCA) were acquired respectively based on the entire data set and the leave-one-out cross validation data set.
3. Results
3.1 Characterization of collagen fibers with local coverage and overall density
The LC metric quantified the spatial distribution of collagen fibers in the imaging field, while OD assessed the overall collagen proportion in a 2D image or 3D stack. Therefore, LC and OD characterized collagen fibers from two different perspectives and provided complementary insights into morphological features of collagen fibers.
Figure 3 shows the detailed quantifications of LC and OD. Collagen fiber images with different distribution properties were used for demonstration [Fig. 3(a)–3(c)]. SHG images, binary masks and LC maps were shown. Corresponding histograms of LC values for these examples were acquired accordingly [Fig. 3(d)–3(f)], with mean of LC (MLC) calculated to reflect the collagen distribution in each case. OD was almost the same for images in Fig. 3(a) and Fig. 3(b), which revealed the same collagen proportion in the two images. However, the local distribution of collagen fibers in these two images was completely different. Collagen fibers were concentrated in a long and narrow region in Fig. 3(a), corresponding to a high LC level [Fig. 3(d)], while in Fig. 3(b) they were sparsely distributed in the whole image, in which LC was low for fiber pixels [Fig. 3(e)]. Generally, higher OD indicated more collagen proportion in the image, yet did not necessarily lead to morphologically denser distribution in the field of view. OD in Fig. 3(c) was higher than that in Fig. 3(a), while MLC was lower, which revealed a sparser distribution of collagen, consistent with observations from both images. In Fig. 3(c), the collagen distribution was complex with collagen concentrated in some small areas but sparse in the whole field of view, leading to relatively even distribution of LC in a range from 0.2 to 0.7 [Fig. 3(f)].
Fig. 3. Characterization of collagen morphological features with local coverage and overall density. (a)-(c) Three groups of representative SHG images (left), binary masks (middle) and LC maps (right) of collagen fibers. Scale bar: 50 µm. (d)-(f) Histograms of LC values corresponding to images shown in (a)-(c), respectively.
3.2 Evaluation of collagen fiber morphological distribution in ovarian tissues with tumor progression
As the main component of extracellular matrix, collagen fibers underwent spatially dynamic remodeling with tumor progression, which made LC and OD potential biomarkers for the diagnosis of ovarian cancer. 3D image stacks of fresh-excised ovarian tissue samples of three groups: normal, benign and malignant, were acquired using SHG imaging and quantitative analysis of LC and OD was then applied to these image stacks. Representative 3D stacks and 2D images of these tissue types are shown in Fig. 4(a) and 4(b), and collagen-background binary matrices are shown in Fig. S4. As can be seen from these images, morphologically denser collagen fibers were distributed in normal tissues and almost occupied the whole imaging area. While in benign tissues, collagen fibers were concentrated into large lumps with gaps between each other. Collagen fibers in malignant tissues were sparsely distributed with pores widely dispersed (more examples from different patients can be found in Fig. S5). The distribution of collagen fibers was clearly visualized from the LC maps [Fig. 4(c)], and information of collagen coverage in a localized region could be directly identified from hues. Histograms of the LC values for collagen voxels validated the observation [Fig. 4(d)]. LC values concentrated near to 1 in normal tissues, indicating that collagen fibers almost fully occupied the field of view. Most LC values in benign tissues were over 0.5, yet with a high variation, in contrast to malignant tissues which had the lowest LC level.
Fig. 4. Overview of collagen images and LC properties for normal, benign and malignant human ovarian tissues. Representative 3D rendering (a), 2D optical sections (b) and LC maps (c) of these tissue types. (d) Corresponding histograms of 3D LC values of these representative fields. Scale bar: 50 µm.
3D imaging of collagen fibers and the 3D extension of the LC metric offered the chance to analyze the axial distribution of collagen fibers in ovarian tissues. The 3D stack of LC maps and representative LC maps at different depths are shown in Fig. 5(a). With the increase of the imaging depth, the MLC of all three representative groups (normal, benign and malignant) increased first and then decreased [Fig. 5(b)], with the morphologically densest collagen distribution at the middle of the tissue [Fig. 5(a), Fig. S6 andVisualization 1, Visualization 2, Visualization 3]. However, the depth-dependent variation in MLC of malignant and benign tissues was more dramatic than that of normal tissues [Fig. 5(b)], indicating sharper changes of collagen fiber distributions [Fig. 5(a), Fig. S6 and Visualization 1, Visualization 2, Visualization 3]. We also calculated the OD at different depths [Fig. 5(c)], yet no significant differences were found among different groups (Fig. S7). These results indicated that both mean and variation of LC were sensitive to tumor progression, corresponding to entirety and heterogeneity of collagen distribution, respectively.
Fig. 5. Depth-dependent heterogeneity of collagen distribution analysis. (a) 3D rendering of LC maps and representative maps at different depths (see Visualizations 1-3). Scale bar: 50 µm. (b), (c) Depth-dependent profiles of mean local coverage (MLC) and overall density (OD).
3.3 Ovarian cancer diagnosis with multi-parametric analysis
A total of 27 ovarian tissue samples, including 9 normal, 9 benign and 9 malignant from 23 patients, were imaged for multi-parametric analysis. For each tissue sample, typically two or three fields were imaged and confirmed by pathologists, with one 3D stack acquired for each field. During the quantitative analysis, the readouts of each metric were first averaged through the 3D stack, and these values from different 3D stacks were then averaged as the final output for each tissue sample. The brief flowchart of the assessment process is shown in Fig. 6(a). Overall density (OD) was calculated to quantify the collagen proportion. Local coverage (LC) was evaluated at voxel level, and mean of LC (MLC) and depth-dependent variation of LC (DVLC), quantified as the standard deviation of MLC at different depth, were calculated to reflect the entirety and heterogeneity of morphological collagen distribution, respectively. Canonical linear discriminant analysis was performed using collagen metrics including OD, MLC and DVLC. Both original classification accuracy (OCA) and cross-validated classification accuracy (CVCA) were obtained to verify the ability of multi-parametric analysis in distinguishing cancerous samples from normal ones.
Fig. 6. Ovarian cancer diagnosis with multi-parametric analysis. (a) Brief flowchart for ovarian tissue diagnosis. Scale bar: 100 µm. (b) Box plots of each optical metric, including MLC (left), OD (middle), and DVLC (right). *, p < 0.05; **, p < 0.01; ***, p < 0.001. (c) 3D scatter plot showing good discrimination of multi-parameters in classifying different groups. (d) Classification accuracy results.
The discrimination of MLC was highly significant for three groups of ovarian tissues [Fig. 6(b), left]. Obvious lower MLC level for malignant tissues was observed compared with normal and benign tissues, indicating morphologically sparser collagen distribution. MLC of normal tissues was higher than that of benign tissues on the whole, although a few samples could not be distinguished. The collagen proportion decreased in malignant tissues, as shown by OD results [Fig. 6(b), middle], while there was no significant difference in OD between normal and benign tissues [Fig. 6(b), middle]. These results revealed similar collagen proportion in normal and benign tissues, yet with different morphological distribution. Similar as results in Fig. 5(b), the DVLC of benign and malignant tissues was higher than that of normal tissues, resulting from more heterogeneity in depth-dependent collagen distribution [Fig. 6(b), right]. To fully utilize the complementary information provided by the multi-parametric quantification, linear discriminant analysis was generated using a combination of MLC, DVLC and OD. Three groups of ovarian tissues occupied separate areas in the 3D scatter plot [Fig. 6(c)], showing good classification of normal, benign and malignant tissues. Both OCA and CVCA achieved 96.3%, which exhibited the ability of the multi-parametric analysis as a sensitive biomarker for human ovarian cancer diagnosis [Fig. 6(d)]. Interestingly, in our study we found that OD did not increase the discriminative power for the identification of normal, benign and malignant ovarian tissues, and both MLC and DVLC were more sensitive than OD in tissue classification [Fig. 6(b)]. Moreover, the good classification result between malignant and benign tissues exhibited the potential application of multi-parametric analysis in treatment method selection for patients. Optical sectioning ability of SHG imaging and 3D analysis improved the discrimination accuracy between normal and benign tissues by offering the axial information compared with 2D analysis (Fig. 7). For 2D analysis, SHG images were assessed frame by frame [Fig. 7(a)], and only MLC and OD were obtained (without DVLC, which was quantified using the depth-dependent information). OCA and CVCA values indicated more samples were misclassified for normal and benign tissues compared with 3D analysis [Fig. 7(b)].
Fig. 7. 2D analysis and classification results for ovarian tissues. (a) For 2D analysis, SHG images were assessed frame by frame, with only 2D local coverage and overall density quantified and applied to the discrimination of ovarian tissues. (b) Classification results of normal, benign and malignant tissues using 2D local coverage and overall density.
4. Discussion
Ovarian cancer is the most common cause of gynecologic cancer death [30], yet unfortunately, there was no recommended screening test for ovarian cancer [1], and the combination of CA125 and transvaginal ultrasound used for screening test did not lead to a reduction in ovarian cancer morality after up to 19 years of follow-up [31]. ECM underwent dynamic deregulation and disorganization in cancer by directly promoting cellular transformation and metastasis, and generating tumorigenic microenvironment [6,32]. Structural remodeling of collagen fibers (the main component of ECM) was proved as a sensitive biomarker of tumor progression for different cancer types, including breast cancer [33,34], pancreatic cancer [22], renal cell carcinoma [35] and ovarian cancer [36], which highlighted collagen fiber imaging and characterization as a potential tool for auxiliary cancer diagnosis. Benefit from the sub-micron resolution and label-free advantage of SHG imaging, 3D fine structures of collagen fibers can be visualized and assessed without any destruction to the tissue, offering the opportunity to detect small lesions at early stages. With advances in the all-fiber-optic SHG endomicroscopy [37–39], real-time in vivo assessment of epithelium and collagen fibers became possible, providing further optimism for clinical potential of the ovarian cancer screening and diagnosis with collagen characteristics.
Multi-parametric analysis of MLC, OD and DVLC provided complementary insights into collagen fiber characteristics and shed new light on the screening test of ovarian cancer and a better understanding of collagen remodeling in tumor progression. Provenzano et al. used collagen density (OD in this study) as one of the characteristics to define three tumor-associated collagen signatures (TACS) to assess the collagen remodeling during breast cancer progression [33,34]. However, quantification method regarding the spatial distribution of collagen fibers has not yet been reported. Local coverage (LC) metric developed in this study provided a novel perspective to characterize the spatial distribution of collagen fibers in a localized region. The LC metric exhibited good discrimination of normal, benign and malignant ovarian tissues, while the difference between normal and benign tissues could hardly be told by OD [Fig. 6(b), middle]. Enabled by the great ability of optical sectioning offered by SHG, DVLC could be assessed which provided information in heterogeneity of collagen distribution along the depth dimension. Higher DVLC level in benign and malignant tissues was possibly caused by the abnormal ECM buildup and deregulated expression of ECM remodeling enzymes, as a result of tumor progression [40,41]. Both OD and LC, acquired based on the image segmentation, were also applicable to images from other tools of cancer screening, especially for hematoxylin and eosin (H&E) stain histological examination [42,43], which was the gold standard of definite cancer diagnosis. Moreover, deep-learning based cross-modality image synthesis method enabled generation of SHG-like collagen fiber images from H&E slides without additional specialized imaging stains, systems or equipment [44], which made diagnosis based on collagen signatures more accessible.
We observed a decrease in LC from malignant tissues in this study. However, it is worth mentioning that ovarian cancer encompassed a group of heterogenous malignancies, which were differentiated by cell/site origin, pathologic grade, risk factors, prognosis and treatments [1,45]. Different types and stages of ovarian cancer might be accompanied by variable tumor behavior and ECM reaction. A series of works showed that benign ovarian tumors contained decreased collagen I compared with normal tissues [46] and malignant ovarian tumors with a lower level of collagen I and III compared with benign tissues [47,48]. Nevertheless, Campbell et al. found a higher collagen density in stage III ovarian cancer by measuring SHG scattering coefficient [49]. Moreover, they found collagen density of ovarian cancerous tissues at stage IV was lower than that of benign ones [50], which further supported that collagen remodeling in ovarian cancer might be stage dependent. Zhu et al. found the collagen fiber proliferation in benign and borderline tumors (tumor type between benign and malignant) and decreased collagen fibers in malignant tissues caused by collagenolytic enzymes produced either by the tumor cells or stromal cells [48], which was a possible reason for a higher collagen density at stage III while a decrease at stage IV of ovarian cancer observed by Campbell et al.
In this work, we focused on the proportion and spatial distribution of collagen fibers within ovarian tissues, corresponding to the OD and LC metric respectively. It should be pointed out that density-associated measures solely might not be able to provide a comprehensive understanding of different ovarian tissue types [36,50,51], and a possible solution would be a combination of different fiber organizational features. Directional variance, ranging between 0 and 1, was a metric developed to quantify the spatial alignment of fiber-like structures, with 0 corresponding to perfectly parallel alignment while 1 corresponding to complete randomness [22]. By combining LC with directional variance [ Fig. 8(a)], we found that for some fields with similar LC level between normal and benign ovarian tissues [Fig. 8(b)], obvious differences were observed in distribution histograms of directional variance [Fig. 8(c)], indicating that in some cases the alignment measure might be a good supplement to density assessments, and a combination of these features might contribute to challenging situations with extremely subtle alterations in collagen fiber organization taking place, such as the early stage of cancer progression, although future work should be needed to validate this possibility. Other structural features were also reported to distinguish different ovarian tissue types. For example, Wen et al. assessed collagen remodeling during ovarian cancer progression using texture analysis and achieved classification of normal, benign and different types of malignant tumors in both 2D and 3D formats [36,52]. Therefore, a combination of density-associated measures and other fiber-specific features, such as alignment, thickness [53,54] and waviness [55,56], would be our future work for a better understanding of different ovarian tissue types and cell-matrix interactions.
Fig. 8. Assessment of collagen fibers with both LC and directional variance. (a) SHG intensity images (left), LC maps (middle) and directional variance maps (right) of normal (top) and benign (bottom) ovarian tissues. Scale bar: 100 µm. (b) Corresponding distribution histograms of LC values for these two tissue types. (c) Corresponding distribution histograms of directional variance values.
5. Conclusion
In this study, we develop local coverage (LC) metric to quantify the pixel/voxel-wise localized distribution of collagen fibers, and combine LC with overall density (OD) to achieve multi-parametric analysis of ovarian tissue images acquired from SHG imaging. Quantification results exhibit distinct discrimination of LC in different tissue groups, and a high level of heterogeneity in collagen distribution along the depth dimension is observed in benign and malignant tissues. An overall classification accuracy of 96.3% in diagnosis of normal, benign and malignant ovarian tissues proves the multi-parametric analysis as a potential tool for the screening test of ovarian lesions. With recent advances in SHG endoscopy, it becomes more readily accessible to acquire high-resolution images of ovarian collagen fibers in situ and in vivo, which facilitates the potential clinical translation of the technique presented.
Funding
National Key Research and Development Program of China (2019YFE0113700, 2017YFA0700501); National Natural Science Foundation of China (61905214, 62035011, 11974310, 31927801); Natural Science Foundation of Zhejiang Province (LR20F050001).
Disclosures
The authors declare no conflict of interest.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
Supplemental document
See Supplement 1 for supporting content.
References
1. L. A. Torre, B. Trabert, C. E. DeSantis, K. D. Miller, G. Samimi, C. D. Runowicz, M. M. Gaudet, A. Jemal, and R. L. Siegel, “Ovarian cancer statistics, 2018,” CA: A Cancer J. Clin. 68(4), 284–296 (2018). [CrossRef]
2. U. Menon, M. Griffin, and A. Gentry-Maharaj, “Ovarian cancer screening–current status, future directions,” Gynecol. Oncol. 132(2), 490–495 (2014). [CrossRef]
3. D. M. Twickler and E. Moschos, “Ultrasound and assessment of ovarian cancer risk,” AJR, Am. J. Roentgenol. 194(2), 322–329 (2010). [CrossRef]
4. D. Timmerman, L. Ameye, D. Fischerova, E. Epstein, G. B. Melis, S. Guerriero, C. Van Holsbeke, L. Savelli, R. Fruscio, A. A. Lissoni, A. C. Testa, J. Veldman, I. Vergote, S. Van Huffel, T. Bourne, and L. Valentin, “Simple ultrasound rules to distinguish between benign and malignant adnexal masses before surgery: prospective validation by IOTA group,” BMJ 341(dec14 1), c6839 (2010). [CrossRef]
5. M. Felder, A. Kapur, J. Gonzalez-Bosquet, S. Horibata, and M. S. Patankar, “MUC16 (CA125): Tumor biomarker to cancer therapy, a work in progress,” Mol. Cancer 13(1), 129 (2014). [CrossRef]
6. P. Lu, V. M. Weaver, and Z. Werb, “The extracellular matrix: a dynamic niche in cancer progression,” J Cell Biol 196(4), 395–406 (2012). [CrossRef]
7. D. E. Discher, D. J. Mooney, and P. W. Zandstra, “Growth Factors, Matrices, and Forces Combine and Control Stem Cells,” Science 324(5935), 1673–1677 (2009). [CrossRef]
8. E. Zamir, M. Katz, Y. Posen, N. Erez, K. M. Yamada, B. Z. Katz, S. Lin, D. C. Lin, A. Bershadsky, Z. Kam, and B. Geiger, “Dynamics and segregation of cell-matrix adhesions in cultured fibroblasts,” Nat. Cell Biol. 2(4), 191–196 (2000). [CrossRef]
9. C. Ricciardelli and R. J. Rodgers, “Extracellular matrix of ovarian tumors,” Semin Reprod Med 24(4), 270–282 (2006). [CrossRef]
10. D. T. Butcher, T. Alliston, and V. M. Weaver, “A tense situation: forcing tumour progression,” Nat. Rev. Cancer 9(2), 108–122 (2009). [CrossRef]
11. T. R. Cox and J. T. Erler, “Remodeling and homeostasis of the extracellular matrix: implications for fibrotic diseases and cancer,” Dis Model Mech 4(2), 165–178 (2011). [CrossRef]
12. K. R. Levental, H. Yu, L. Kass, J. N. Lakins, M. Egeblad, J. T. Erler, S. F. T. Fong, K. Csiszar, A. Giaccia, W. Weninger, M. Yamauchi, D. L. Gasser, and V. M. Weaver, “Matrix Crosslinking Forces Tumor Progression by Enhancing Integrin Signaling,” Cell 139(5), 891–906 (2009). [CrossRef]
13. P. Campagnola, “Second Harmonic Generation Imaging Microscopy: Applications to Diseases Diagnostics,” Anal. Chem. 83(9), 3224–3231 (2011). [CrossRef]
14. P. J. Campagnola and C. Y. Dong, “Second harmonic generation microscopy: principles and applications to disease diagnosis,” Laser Photonics Rev. 5(1), 13–26 (2011). [CrossRef]
15. P. J. Campagnola and L. M. Loew, “Second-harmonic imaging microscopy for visualizing biomolecular arrays in cells, tissues and organisms,” Nat. Biotechnol. 21(11), 1356–1360 (2003). [CrossRef]
16. R. Cicchi, N. Vogler, D. Kapsokalyvas, B. Dietzek, J. Popp, and F. S. Pavone, “From molecular structure to tissue architecture: collagen organization probed by SHG microscopy,” J. Biophotonics 6(2), 129–142 (2013). [CrossRef]
17. L. Mostaço-Guidolin, N. Rosin, and T.-L. Hackett, “Imaging Collagen in Scar Tissue: Developments in Second Harmonic Generation Microscopy for Biomedical Applications,” Int. J. Mol. Sci. 18(8), 1772 (2017). [CrossRef]
18. H. Altendorf, E. Decenciere, D. Jeulin, P. De sa Peixoto, A. Deniset-Besseau, E. Angelini, G. Mosser, and M. C. Schanne-Klein, “Imaging and 3D morphological analysis of collagen fibrils,” J Microsc 247(2), 161–175 (2012). [CrossRef]
19. T. Y. Lau, R. Ambekar, and K. C. Toussaint, “Quantification of collagen fiber organization using three-dimensional Fourier transform-second-harmonic generation imaging,” Opt. Express 20(19), 21821–21832 (2012). [CrossRef]
20. V. J. Napadow, Q. Chen, V. Mai, P. T. So, and R. J. Gilbert, “Quantitative analysis of three-dimensional-resolved fiber architecture in heterogeneous skeletal muscle tissue using nmr and optical imaging methods,” Biophys. J. 80(6), 2968–2975 (2001). [CrossRef]
21. Z. Liu, K. P. Quinn, L. Speroni, L. Arendt, C. Kuperwasser, C. Sonnenschein, A. M. Soto, and I. Georgakoudi, “Rapid three-dimensional quantification of voxel-wise collagen fiber orientation,” Biomed. Opt. Express 6(7), 2294–2310 (2015). [CrossRef]
22. Z. Liu, D. Pouli, D. Sood, A. Sundarakrishnan, C. K. Hui Mingalone, L. M. Arendt, C. Alonzo, K. P. Quinn, C. Kuperwasser, L. Zeng, T. Schnelldorfer, D. L. Kaplan, and I. Georgakoudi, “Automated quantification of three-dimensional organization of fiber-like structures in biological tissues,” Biomaterials 116, 34–47 (2017). [CrossRef]
23. R. M. Williams, W. R. Zipfel, and W. W. Webb, “Interpreting second-harmonic generation images of collagen I fibrils,” Biophys. J. 88(2), 1377–1386 (2005). [CrossRef]
24. Z. Liu, L. Speroni, K. P. Quinn, C. Alonzo, D. Pouli, Y. Zhang, E. Stuntz, C. Sonnenschein, A. M. Soto, and I. Georgakoudi, “3D organizational mapping of collagen fibers elucidates matrix remodeling in a hormone-sensitive 3D breast tissue model,” Biomaterials 179, 96–108 (2018). [CrossRef]
25. W. R. Zipfel, R. M. Williams, and W. W. Webb, “Nonlinear magic: multiphoton microscopy in the biosciences,” Nat. Biotechnol. 21(11), 1369–1377 (2003). [CrossRef]
26. K. Schenke-Layland, I. Riemann, O. Damour, U. A. Stock, and K. König, “Two-photon microscopes and in vivo multiphoton tomographs — Powerful diagnostic tools for tissue engineering and drug delivery,” Adv. Drug Delivery Rev. 58(7), 878–896 (2006). [CrossRef]
27. S. M. Kakkad, M. Solaiyappan, P. Argani, S. Sukumar, L. K. Jacobs, D. Leibfritz, Z. M. Bhujwalla, and K. Glunde, “Collagen I fiber density increases in lymph node positive breast cancers: pilot study,” J. Biomed. Opt. 17(11), 116017 (2012). [CrossRef]
28. T. Hompland, A. Erikson, M. Lindgren, T. Lindmo, and C. De Lange Davies, “Second-harmonic generation in collagen as a potential cancer diagnostic parameter,” J. Biomed. Opt. 13(5), 054050 (2008). [CrossRef]
29. S. Wanxin, C. Shi, C. S. T. Dean, T. Nancy, X. Guangfa, T. Huihuan, and Y. Hanry, “Nonlinear optical microscopy: use of second harmonic generation and two-photon microscopy for automated quantitative liver fibrosis studies,” J. Biomed. Opt. 13, 1–7 (2008). [CrossRef]
30. N. A. Howlader, M Krapcho, D Miller, A Brest, M Yu, J Ruhl, Z Tatalovich, A Mariotto, D. R. Lewis, H. S. Chen, E. J. Feuer, and K. A. Cronin, eds., “SEER Cancer Statistics Review, 1975-2017,” (National Cancer Institute, Bethesda, MD, 2017) https://seer.cancer.gov/csr/1975_2017/, based on November 2019 SEER data submission, posted to the SEER web site, April 2020.
31. P. F. Pinsky, K. Yu, B. S. Kramer, A. Black, S. S. Buys, E. Partridge, J. Gohagan, C. D. Berg, and P. C. Prorok, “Extended mortality results for ovarian cancer screening in the PLCO trial with median 15 years follow-up,” Gynecol. Oncol. 143(2), 270–275 (2016). [CrossRef]
32. C. Frantz, K. M. Stewart, and V. M. Weaver, “The extracellular matrix at a glance,” J. Cell Sci. 123(24), 4195–4200 (2010). [CrossRef]
33. P. P. Provenzano, K. W. Eliceiri, J. M. Campbell, D. R. Inman, J. G. White, and P. J. Keely, “Collagen reorganization at the tumor-stromal interface facilitates local invasion,” BMC Med. 4(1), 38 (2006). [CrossRef]
34. P. P. Provenzano, D. R. Inman, K. W. Eliceiri, J. G. Knittel, L. Yan, C. T. Rueden, J. G. White, and P. J. Keely, “Collagen density promotes mammary tumor initiation and progression,” BMC Med. 6(1), 11 (2008). [CrossRef]
35. S. L. Best, Y. Liu, A. Keikhosravi, C. R. Drifka, K. M. Woo, G. S. Mehta, M. Altwegg, T. N. Thimm, M. Houlihan, J. S. Bredfeldt, E. J. Abel, W. Huang, and K. W. Eliceiri, “Collagen organization of renal cell carcinoma differs between low and high grade tumors,” BMC Cancer 19(1), 490 (2019). [CrossRef]
36. B. Wen, K. R. Campbell, K. Tilbury, O. Nadiarnykh, M. A. Brewer, M. Patankar, V. Singh, K. W. Eliceiri, and P. J. Campagnola, “3D texture analysis for classification of second harmonic generation images of human ovarian cancer,” Sci. Rep. 6(1), 35734 (2016). [CrossRef]
37. Y. Wu, J. Xi, M. J. Cobb, and X. Li, “Scanning fiber-optic nonlinear endomicroscopy with miniature aspherical compound lens and multimode fiber collector,” Opt. Lett. 34(7), 953–955 (2009). [CrossRef]
38. Y. Wu, Y. Leng, J. Xi, and X. Li, “Scanning all-fiber-optic endomicroscopy system for 3D nonlinear optical imaging of biological tissues,” Opt. Express 17(10), 7907–7915 (2009). [CrossRef]
39. Y. Zhang, M. L. Akins, K. Murari, J. Xi, M. J. Li, K. Luby-Phelps, M. Mahendroo, and X. Li, “A compact fiber-optic SHG scanning endomicroscope and its application to visualize cervical remodeling during pregnancy,” Proc. Natl. Acad. Sci. U. S. A. 109(32), 12878–12883 (2012). [CrossRef]
40. J. T. Erler, K. L. Bennewith, T. R. Cox, G. Lang, D. Bird, A. Koong, Q.-T. Le, and A. J. Giaccia, “Hypoxia-Induced Lysyl Oxidase Is a Critical Mediator of Bone Marrow Cell Recruitment to Form the Premetastatic Niche,” Cancer Cell 15(1), 35–44 (2009). [CrossRef]
41. A. Page-Mccaw, A. J. Ewald, and Z. Werb, “Matrix metalloproteinases and the regulation of tissue remodelling,” Nat. Rev. Mol. Cell Biol. 8(3), 221–233 (2007). [CrossRef]
42. K. Bera, K. A. Schalper, D. L. Rimm, V. Velcheti, and A. Madabhushi, “Artificial intelligence in digital pathology — new tools for diagnosis and precision oncology,” Nat. Rev. Clin. Oncol. 16(11), 703–715 (2019). [CrossRef]
43. O. G. F. Geessink, A. Baidoshvili, J. M. Klaase, B. Ehteshami Bejnordi, G. J. S. Litjens, G. W. Van Pelt, W. E. Mesker, I. D. Nagtegaal, F. Ciompi, and J. A. W. M. Van Der Laak, “Computer aided quantification of intratumoral stroma yields an independent prognosticator in rectal cancer,” Cell. Oncol. 42(3), 331–341 (2019). [CrossRef]
44. A. Keikhosravi, B. Li, Y. Liu, M. W. Conklin, A. G. Loeffler, and K. W. Eliceiri, “Non-disruptive collagen characterization in clinical histopathology using cross-modality image synthesis,” Commun. Biol. 3(1), 414 (2020). [CrossRef]
45. N. Wentzensen, E. M. Poole, B. Trabert, E. White, A. A. Arslan, A. V. Patel, V. W. Setiawan, K. Visvanathan, E. Weiderpass, H.-O. Adami, A. Black, L. Bernstein, L. A. Brinton, J. Buring, L. M. Butler, S. Chamosa, T. V. Clendenen, L. Dossus, R. Fortner, S. M. Gapstur, M. M. Gaudet, I. T. Gram, P. Hartge, J. Hoffman-Bolton, A. Idahl, M. Jones, R. Kaaks, V. Kirsh, W.-P. Koh, J. V. Lacey, I. M. Lee, E. Lundin, M. A. Merritt, N. C. Onland-Moret, U. Peters, J. N. Poynter, S. Rinaldi, K. Robien, T. Rohan, D. P. Sandler, C. Schairer, L. J. Schouten, L. K. Sjöholm, S. Sieri, A. Swerdlow, A. Tjonneland, R. Travis, A. Trichopoulou, P. A. Van Den Brandt, L. Wilkens, A. Wolk, H. P. Yang, A. Zeleniuch-Jacquotte, and S. S. Tworoger, “Ovarian Cancer Risk Factors by Histologic Subtype: An Analysis From the Ovarian Cancer Cohort Consortium,” J. Clin. Oncol. 34(24), 2888–2898 (2016). [CrossRef]
46. M. Wolańska, K. Sobolewski, E. Bańkowski, and M. Drozdzewicz, “Accumulation of collagen in ovarian benign tumours,” Acta Biochim Pol 46(4), 941–947 (1999). [CrossRef]
47. S. Kauppila, M. K. Bode, F. Stenbäck, L. Risteli, and J. Risteli, “Cross-linked telopeptides of type I and III collagens in malignant ovarian tumours in vivo,” Br. J. Cancer 81(4), 654–661 (1999). [CrossRef]
48. G.-G. Zhu, A. Kauppila, L. Risteli, M. Mäkinen, F. Stenbäck, and J. Risteli, “Immunohistochemical study of type I collagen and type I pN-collagen in benign and malignant ovarian neoplasms,” Cancer 75(4), 1010–1017 (1995). [CrossRef]
49. K. R. Campbell, R. Chaudhary, J. M. Handel, M. S. Patankar, and P. J. Campagnola, “Polarization-resolved second harmonic generation imaging of human ovarian cancer,” J Biomed Opt 23, 066501 (2018). [CrossRef]
50. K. B. Tilbury, K. R. Campbell, K. W. Eliceiri, S. M. Salih, M. Patankar, and P. J. Campagnola, “Stromal alterations in ovarian cancers via wavelength dependent Second Harmonic Generation microscopy and optical scattering,” BMC Cancer 17(1), 102 (2017). [CrossRef]
51. S. Alkmin, R. Brodziski, H. Simon, D. Hinton, R. H. Goldsmith, M. Patankar, and P. J. Campagnola, “Role of Collagen Fiber Morphology on Ovarian Cancer Cell Migration Using Image-Based Models of the Extracellular Matrix,” Cancers 12(6), 1390 (2020). [CrossRef]
52. B. L. Wen, M. A. Brewer, O. Nadiarnykh, J. Hocker, V. Singh, T. R. Mackie, and P. J. Campagnola, “Texture analysis applied to second harmonic generation image data for ovarian cancer classification,” J. Biomed. Opt. 19(9), 096007 (2014). [CrossRef]
53. J. S. Bredfeldt, Y. Liu, C. A. Pehlke, M. W. Conklin, and K. W. Eliceiri, “Computational segmentation of collagen fibers from second-harmonic generation images of breast cancer,” J. Biomed. Opt. 19(1), 016007 (2014). [CrossRef]
54. J. Meng, Z. Feng, S. Qian, C. Wang, X. Li, L. Gao, Z. Ding, J. Qian, and Z. Liu, “Mapping physiological and pathological functions of cortical vasculature through aggregation-induced emission nanoprobes assisted quantitative, in vivo NIR-II imaging,” Biomaterials Advances 136, 212760 (2022). [CrossRef]
55. F. Benboujja and C. Hartnick, “Quantitative evaluation of the human vocal fold extracellular matrix using multiphoton microscopy and optical coherence tomography,” Sci. Rep. 11(1), 2440 (2021). [CrossRef]
56. S. Qian, J. Meng, W. Liu, C. Wang, R. Jiang, L. Yang, X. Liu, C. Kuang, Z. Ding, and Z. Liu, “Identification of endoplasmic reticulum formation mechanism by multi-parametric, quantitative super-resolution imaging,” Opt. Lett. 47(2), 357–360 (2022). [CrossRef]
