|
1.INTRODUCTIONBreast cancer is the second most common form of cancer diagnosed worldwide, accounting for 11.9% of all cancers diagnosed in 2012.1 In spite of the high incidence and burden of disease the current histopathological analysis used for the diagnosis of breast cancers suffers from certain shortcomings. When an abnormality in the breast is discovered during a screening procedure such as mammography, a tissue biopsy is obtained by the pathologist and the section of tissue is stained using hematoxylin and eosin (H&E). This staining provides the necessary contrast needed for investigation of key morphological features by a trained, board certified pathologist using a conventional bright field microscope. Since this investigation is both qualitative and subjective it leads to both intra and inter observer discrepancy.2,3 Furthermore, due to the qualitative nature of the histopathological information, the imaging modality does not lend itself to automation resulting in low throughput and in some cases, late disease diagnosis – a critical shortcoming given that early diagnosis significantly improves chances of survival.4 Quantitative phase imaging (QPI) refers to a sub-set of label-free microscopy techniques where contrast is generated by the variation of optical path length transerve to the direction of light propagation. The resulting image is a phase map ϕ(x,y) that is a quantitative measure of the product of the refractive index difference between the tissue nt(x,y) and its surrounding medium nm and the thickness of the tissue t(x,y) given by the equation where λ is the wavelength of light.5-14 Since the information in the phase image ϕ(x,y) is a quantitative measure of the morphology of the tissue biopsy, diagnosis based on QPI provides the potential to eliminate inter and intra observer variation and lends itself to automation. Spatial light interference microscopy (SLIM) provides access to this phase information by measuring four interferograms formed by the scattered and unscattered fields and solving for ϕ(x,y). It has been shown in previous publications from our group that SLIM provides diffraction limited resolution as well as low spatial and temporal noise leading to optical path length sensitivity of less than a nanometer.15-17 The utility of refractive index of tissue (accessible through the phase images obtained by SLIM) as a marker for prostate tissue malignancy has also been shown in previous publications – motivating the investigation of SLIM as a histopathological analysis technique for detecting malignancy in breast tissue biopsies.18 In this work we present preliminary results that show the potential of a SLIM based technique for supporting diagnosis of breast cancers. Specifically, the resolution and contrast of SLIM phase images for diagnostic purposes were evaluated qualitatively by two board certified pathologists. As outlined in detail in the following sections, using the standard H&E staining based diagnosis protocol as a benchmark, the success of the pathologists in carrying out diagnosis on SLIM images was measured. Our results provide an indication of the signal to noise ratio available to us for subsequent quantitative analyses for carrying out diagnosis based on the relative phase values of various tissue components. 2.EXPERIMENTAL PROCEDURES2.1SampleThe samples comprised of an unstained tissue microarray (TMA) of 900 cores constructed from breast tissue biopsies of 400 different patients. The cores were mounted on three microscope slides and each core section was 4 µm thick. The sample was obtained from our collaborating pathologist at the University of Illinois at Chicago (UIC) Dr. Andre Balla. Corresponding to each unstained core, an H&E stained core from an adjacent parallel section of tissue was mounted onto separate slides for imaging under a bright field microscope, providing reference H&E images for comparison. 2.2Slide scanning and stitchingThe TMA was imaged using our SLIM imaging system, constructed as an add-on module to a commercial phase contrast microscope (Zeiss Axio Observer Z1). A 40x/0.75 NA phase contrast objective was used for the purpose. A slide scanning software, developed in-house in Visual C++, was used to obtain the raw images for the entire microscope slide (scanning area approx. 20 mm x 45 mm) at high throughput (approx. 2 hrs per slide). The phase maps were extracted using a MATLAB-based code. A Python-based stitching code was used for stitching the mosaic for the entire slide and segmenting out each individual core for subsequent processing and analysis. As shown in Figure 1, our processing allows the visualization of the entire TMA from the slide scale to the sub-cellular scale within each core. As illustrated in Figure 1 (c), our label-free SLIM images clearly delineate the epithelial stromal boundary allowing for assessment of tumor malignancy. 2.3Evaluation by pathologistsIn order to assess the diagnostic capabilities of our SLIM imaging modality, we asked two board certified pathologists to evaluate our SLIM images. For practical reasons, 109 cores out of the total of 900 were chosen for evaluation, and stacks of SLIM and corresponding H&E images for these cores were assembled in ImageJ. Since pathologists are generally trained to recognize morphological features in H&E stained tissue, a training step was performed before the actual experiment was conducted. In this training step, both SLIM and H&E images for 10 benign and 10 malignant cores (determined as such by a third board certified pathologist a priori) were shown side by side to each pathologist. By comparing the SLIM and H&E images for each core the pathologists were able to learn how to interpret the tissue morphological details from SLIM phase maps. The total training time for each pathologist ranged from 10 – 15 minutes approximately. Figure 2 compares and contrasts how different tissue components are resolved in SLIM and H&E stained tissue images. After the completion of the training step, each pathologist was first shown the stack of SLIM images for the 109 cores chosen. The pathologist classified each core as either benign or malignant. The process was repeated for the stack of H&E images for the same 109 cores. Using each pathologist’s diagnosis on the H&E stained cores as the gold standard, the success of diagnosis using SLIM images was measured by counting the number of agreements between SLIM and H&E based diagnoses. 3.RESULTSThe results of the classification of cores carried out by the two pathologists on both SLIM and H&E images are summarized in Figure 3. As shown in Figure 3 (c), the success rate of diagnosis on SLIM images (considering diagnosis on H&E as the gold standard) was 88%. As shown Figures 3 (a) and (b), the agreement between the two pathologists when rating SLIM images stood at 83% whereas the same for H&E images was much higher at 98%. The lower agreement between the two pathologists on SLIM images is not surprising when one takes into account the fact that, as part of their professional training, pathologists are trained to interpret images of H&E stained tissue for a number of years whereas, for this experiment, the training time for SLIM images was only a few minutes. We expect the agreement between the diagnoses of the two pathologists on SLIM images to increase significantly with longer training in interpreting SLIM images. 4.CONCLUSION AND FUTURE WORKOur preliminary results show the capability of our label-free imaging modality in resolving morphological features relevant for diagnosis of breast cancer. While this qualitative analysis shows the potential of quantitative phase imaging for diagnosing malignancy in breast cancer, the long term aim is to address the shortcomings in conventional histopathology with regards to inter and intra observer variability and low throughput by searching for quantitative parameters for classifying benign and malignant tumors. As has been reported in literature before, phase images of tissues can be used for extracting scattering parameters such as mean-scattering length and tissue anisotropy parameter that can be used to characterize the cellular scale organization of tissue.19 Since malignancy is associated with changes in tissue organization and morphological state of cells, comparing scattering parameters between benign and malignant tumors has the potential to provide a quantitative basis for diagnosis. Hence, our future work is focused on leveraging the quantitative information regarding tissue morphology available to us in phase maps to come up with parameters for diagnosis of breast cancer biopsy images and build classifiers that can achieve this in an automated fashion. In addition to discovering quantitative bases for diagnosis, we also hope to use our imaging modality for discovering new prognostic biomarkers. There is a search in the scientific community for new, readily accessible biomarkers to serve as prognostic indicators as the current set of biomarkers (histological grade, tumor size, hormonal receptor status etc.) are insufficient in predicting outcomes for some patients.20 We hope to address this need by discovering new biomarkers, using our label free images, to serve as predictors of patient outcome. ACKNOWLEDGEMENTSThis work was supported by National Science Foundation (grants CBET-1040462 MRI), Phi Optics and Agilent Laboratories. For more information visit http://light.ece.illinois.edu/. REFERENCES, “International Agency for Research on Cancer, Latest world cancer statistics Global cancer burden rises to 14.1 million new cases in 2012: Marked increase in breast cancers must be addressed,”
World Health Organization,
(2013) http://www.iarc.fr/en/media-centre/pr/2013/pdfs/pr223_E.pdf December 2015). Google Scholar
Benard, A., Desmedt, C., Smolina, M.,
“Infrared imaging in breast cancer: automated tissue component recognition and spectral characterization of breast cancer cells as well as the tumor microenvironment,”
Analyst, 139
(5), 1044
–1056
(2014). https://doi.org/10.1039/c3an01454a Google Scholar
Rakha, E., Reis-Filho, J., Baehner, F.,
“Breast cancer prognostic classification in the molecular era: the role of histological grade,”
Breast Cancer Research, 12
(4), 207
(2010). https://doi.org/10.1186/bcr2607 Google Scholar
Walsh, M. J., Holton, S. E., Kajdacsy-Balla, A.,
“Attenuated total reflectance Fourier-transform infrared spectroscopic imaging for breast histopathology,”
Vibrational Spectroscopy, 60 23
–28
(2012). https://doi.org/10.1016/j.vibspec.2012.01.010 Google Scholar
Popescu, G., Quantitative Phase Imaging of Cells and Tissues, McGraw Hill, New York
(2011). Google Scholar
Mann, C., Yu, L., Lo, C.-M.,
“High-resolution quantitative phase-contrast microscopy by digital holography,”
Optics Express, 13
(22), 8693
–8698
(2005). https://doi.org/10.1364/OPEX.13.008693 Google Scholar
Kim, M., Digital Holographic Microscopy: Principles, Techniques, and Applications, Springer, New York
(2011). https://doi.org/10.1007/978-1-4419-7793-9 Google Scholar
Ou, X., Horstmeyer, R., Yang, C.,
“Quantitative phase imaging via Fourier ptychographic microscopy,”
Optics Letters, 38
(22), 4845
–4848
(2013). https://doi.org/10.1364/OL.38.004845 Google Scholar
Wu, J., Yaqoob, Z., Heng, X.,
“Harmonically matched grating-based full-field quantitative high-resolution phase microscope for observing dynamics of transparent biological samples,”
Optics Express, 15
(26), 18141
–18155
(2007). https://doi.org/10.1364/OE.15.018141 Google Scholar
Memmolo, P., Miccio, L., Merola, F.,
“3D morphometry of red blood cells by digital holography,”
Cytometry Part A, 85
(12), 1030
–1036
(2014). https://doi.org/10.1002/cyto.a.v85.12 Google Scholar
Memmolo, P., Finizio, A., Paturzo, M.,
“Twin-beams digital holography for 3D tracking and quantitative phase-contrast microscopy in microfluidics,”
Optics Express, 19
(25), 25833
–25842
(2011). https://doi.org/10.1364/OE.19.025833 Google Scholar
Paturzo, M., Finizio, A., Memmolo, P.,
“Microscopy imaging and quantitative phase contrast mapping in turbid microfluidic channels by digital holography,”
Lab on a Chip, 12
(17), 3073
–3076
(2012). https://doi.org/10.1039/c2lc40114b Google Scholar
Ferraro, P., Miccio, L., Grilli, S.,
“Quantitative Phase Microscopy of microstructures with extended measurement range and correction of chromatic aberrations by multiwavelength digital holography,”
Optics Express, 15
(22), 14591
–14600
(2007). https://doi.org/10.1364/OE.15.014591 Google Scholar
Park, Y., Popescu, G., Badizadegan, K.,
“Diffraction phase and fluorescence microscopy,”
Optics Express, 14
(18), 8263
–8268
(2006). https://doi.org/10.1364/OE.14.008263 Google Scholar
Wang, Z., Millet, L., Mir, M.,
“Spatial light interference microscopy (SLIM),”
Optics Express, 19
(2), 1016
–1026
(2011). https://doi.org/10.1364/OE.19.001016 Google Scholar
Kim, T., Zhou, R., Mir, M.,
“White-light diffraction tomography of unlabelled live cells,”
Nat Photon, 8
(3), 256
–263
(2014). https://doi.org/10.1038/nphoton.2013.350 Google Scholar
Wang, Z., Millet, L., Chan, V.,
“Label-free intracellular transport measured by spatial light interference microscopy,”
Journal of Biomedical Optics, 16
(2), 026019
–026019-9
(2011). https://doi.org/10.1117/1.3549204 Google Scholar
Wang, Z., Tangella, K., Balla, A.,
“Tissue refractive index as marker of disease,”
Journal of Biomedical Optics, 16
(11), 116017
–1160177
(2011). https://doi.org/10.1117/1.3656732 Google Scholar
Wang, Z., Ding, H., and Popescu, G.,
“Scattering-phase theorem,”
Optics Letters, 36
(7), 1215
–1217
(2011). https://doi.org/10.1364/OL.36.001215 Google Scholar
Conklin, M. W., Eickhoff, J. C., Riching, K. M.,
“Aligned Collagen Is a Prognostic Signature for Survival in Human Breast Carcinoma,”
The American Journal of Pathology, 178
(3), 1221
–1232
(2011). https://doi.org/10.1016/j.ajpath.2010.11.076 Google Scholar
|