- Research Article
- Open Access
- Open Peer Review
Pseudo-HE images derived from CARS/TPEF/SHG multimodal imaging in combination with Raman-spectroscopy as a pathological screening tool
© Bocklitz et al.; licensee BioMed Central. 2016
- Received: 18 December 2015
- Accepted: 5 July 2016
- Published: 26 July 2016
Due to the steadily increasing number of cancer patients worldwide the early diagnosis and treatment of cancer is a major field of research. The diagnosis of cancer is mostly performed by an experienced pathologist via the visual inspection of histo-pathological stained tissue sections. To save valuable time, low quality cryosections are frequently analyzed with diagnostic accuracies that are below those of high quality embedded tissue sections. Thus, alternative means have to be found that enable for fast and accurate diagnosis as the basis of following clinical decision making.
In this contribution we will show that the combination of the three label-free non-linear imaging modalities CARS (coherent anti-Stokes Raman-scattering), TPEF (two-photon excited autofluorescence) and SHG (second harmonic generation) yields information that can be translated into computational hematoxylin and eosin (HE) images by multivariate statistics. Thereby, a computational HE stain is generated resulting in pseudo-HE overview images that allow for identification of suspicious regions. The latter are analyzed further by Raman-spectroscopy retrieving the tissue’s molecular fingerprint.
The results suggest that the combination of non-linear multimodal imaging and Raman-spectroscopy possesses the potential as a precise and fast tool in routine histopathology.
As the key advantage, both optical methods are non-invasive enabling for further pathological investigations of the same tissue section, e.g. a direct comparison with the current pathological gold-standard.
- Cancer detection
- Multimodal imaging
- Pseudo HE-images
- Raman spectroscopy
The WHO expects the annual incidences of cancer to almost double to 21.6 million by 2030 . Evidently, an early cancer diagnosis is the key factor for the survival of patients. The diagnosis of cancer after initial suspicion is complex and involves a number of elaborated diagnostic approaches such as genomics and proteomics [2–4]. A histopathological examination of the excised tissue is the current gold-standard for deriving the final diagnosis . In the majority of cases, the pathologist works with fixated and embedded tissue samples. In order to save time, also native frozen sections are evaluated as part of a quick frozen section analysis. The analysis of frozen sections, however, is challenging and sometimes deviates from the results of an analysis of fixated and embedded sections . To overcome the current limitation of frozen section diagnostics, new pathological tools are required allowing for fast and accurate ex corpore in vivo diagnosis of malignant transformed tissue. Thereby, the term ex corpore in vivo refers to fresh biopsy tissue. Ideally, these techniques are also applicable for in vivo investigations adding the necessity to work non-invasive, i.e. preserving the tissue’s integrity.
Within the last years the development and application of optical methods for clinical pathology that potentially meet these requirements has rapidly increased . Among these methods, spectroscopic imaging approaches are of particular significance [6–9]. Ex vivo reflectance confocal microscopy (CRM) was used to detect residuals of non-melanoma skin cancer (NMSC) during Moh’s surgery . Optical coherence tomography (OCT) demonstrated its potential to differentiate between malignant and benign tissue areas in head and neck, skin, genital and bladder cancer . The potential of photo acoustic imaging (PAI) for cancer diagnostics was evaluated for melanoma  and breast cancer . One-photon excited autofluorescence (OPEF) was utilized to investigate fibrosarcoma, lung cancer and NMSC . Linear Raman-microscopy was applied for differentiation of healthy tissue from cancerous epithelium within human skin, colon, brain and breast . Further, various non-linear microscopy methods were shown to enable for detection of NMSC as well as head and neck, brain or lung cancer . However, most of these studies represent proof-of-concept studies and except of OCT these methods have not been transferred into routine clinical applications. The delay of technology transfer may be attributed – among other reasons – to the increasing complexity of state of the art cancer diagnostics reducing physician’s available time to familiarize with new imaging technologies as well as its image contrast and significance. Thus, it is the task of scientists to reduce the complexity of new technologies by translating the image information into a format that physicians are accustomed to such as hematoxylin and eosin (HE) stained images or by direct prediction of the tissue’s malignancy state. Ideally, this image translation is achieved entirely by computational image analysis requiring no assistance of scientists or physicians.
To automate such a translation of optical microscopy data, the image contrast is required to provide cancer specific information in the first place. Since the information transferred by a single modality is limited, various optical methods are frequently combined such as Raman-spectroscopy and OCT for skin cancer detection . These multimodal approaches can be grouped into those that gather techniques with similar image acquisition times and experimental equipment and combinations that merge sensitivity with specificity by coupling fast imaging tools with techniques of high information density per pixel. Here, we combine both multimodal concepts to maximize the image acquisition speed and information depth in order to improve the accuracy of image translation and diagnostic results.
First, the fast non-linear microscopy methods CARS = coherent anti-Stokes Raman-scattering, SHG = second harmonic generation and TPEF = two-photon excited autofluorescence were jointly applied to characterize the architecture and biochemical composition of frozen tissue sections. For the selection of regions of interest (ROI), we demonstrate for the first time the possibility to derive computationally pseudo-HE images from CARS/SHG/TPEF-images by applying multivariate statistics. The pseudo-HE image can be analyzed by a pathologist in the same manner as a normal HE image. Following the selection of ROI based on pseudo-HE images we applied Raman-spectroscopy for the prediction of the diagnosis. Though compromised by its poor sensitivity, Raman-spectroscopy is unprecedented for its high specificity yielding information based on inherent molecular vibrations that - like fingerprints - specifically characterize chemical structures and biochemical compositions of e.g. biological cells, tissues etc.
The non-invasiveness of non-linear multimodal imaging and Raman-spectroscopy enables for the direct comparison with the pathological gold standard requiring staining. Thus, further analyses can be performed on the samples pre-characterized by combining multimodal imaging and Raman-spectroscopy. This new approach, the combination of non-linear multimodal images and Raman-spectra of selected regions, was evaluated for ex vivo sections of colon tissue that arose from p53 knockout mice . The mouse model was selected to investigate a single type of tumor while minimizing the variance between individuals (mice). The results, therefore, allow for a reliable estimation of the generalizability of the proposed optical pathological tool with respect to the adenoma-carcinoma-sequence and its diagnostic value.
Mice with an intestinal epithelial cell (IEC)-specific deletion of the tumor suppressor gene p53 were received by crossing of floxed p53 mice with villin-Cre mice (Tp53 ΔIEC) . As controls Tp53 F/F mice were utilized. These mice were treated intraperitoneally once a week for 6 weeks with the carcinogen azoxymethane (AOM; 10 mg/kg). For histological and Raman-spectroscopic investigations mice were sacrificed at different time points after the AOM application. The specific loss of p53 in the intestine markedly enhances the carcinogen-induced tumor incidence and leads to the development of invasive colorectal tumors beginning about 12 weeks after the first AOM treatment.
Full preparation of colon and rectum was performed for 47 individuals. Biopsies were acquired from 8 out of these 47 individuals. Another cohort of 22 individuals was divided into two groups of which biopsies were taken in an alternating fashion every two weeks. In total, 69 individuals contributed to the complete training Raman-dataset. For the present study, 6 individuals (2 males, 4 females) were selected randomly covering the whole spectrum of diagnoses. In total 34 Raman-maps (46240 spectra) and the respective non-linear multimodal images were acquired to investigate this reduced cohort.
Cryosections for the histo-pathological evaluation were obtained following the standard procedure of washing the specimens with phosphate buffered saline (PBS Buffer), fixation using paraformaldehyde, embedding in paraffin and cutting the sample in 10 μm slices which were stained with hematoxylin and eosin (HE staining). For the Raman-spectroscopic analysis, the specimens were washed with PBS buffer and immediately frozen in liquid nitrogen. Distilled water was used as embedding medium for the cryosectioning step resulting in 20 μm thick slices. Thus, a native sample - preserving the lipid distribution - is obtained.
Brightfield images of the parallel HE stained as well as Raman-spectroscopically examined sections were recorded using a halogen illuminated Leica DM2500 microscope (Leica Microsystems, Wetzlar, Germany) equipped with a Nikon Digital Sight (DS) camera system using the DS-Fi1 CCD camera head (Nikon Instruments Europe B.V., Germany). To superimpose the measured Raman-maps with the their corresponding histopathological assessment, all HE stained scan areas were imaged using five different Leica objectives in the range of 1.25 × to 40 × (1.25 × HCX PL Fluotar (NA 0.04), 5 × N Plan (NA 0.12), 10 × N Plan (NA 0.25), 20 × N Plan (NA 0.4), 40 × N Plan (NA 0.65)).
Non-linear multimodal microscopy
Raman-spectra of 20 μm thick cryosections from colon and rectum of mice as well as from biopsies on CaF 2 slides were recorded with an upright micro-Raman-setup (CRM 300, WITec GmbH, Ulm, Germany) equipped with a 300 g/mm grating (7 cm −1 resolution) and a Deep Depletion CCD camera (DU401 BR-DD, ANDOR, 1024 × 127 pixels) cooled down to −65°C. The tissue Raman-spectra were recorded with a 785 nm diode laser which was focused through a Zeiss 50 × objective (NA 0.7) onto the sections. For the herein described study 34 Raman-scans were recorded in scanning mode with a step size of 5 μm and an integration time of 2 s per spectrum. The scan dimension was chosen to be around 34×40 pixels (170 μm × 200 μm), so every scan consists of approximately 1360 Raman-spectra. As a standard for the subsequent processing of the spectra a time series comprising 50 spectra of 4-acetamidophenol was collected with an integration time of 1 s per spectrum. Based on HE stained parallel sections a trained pathologist selected the areas to be measured. Here we measured the Raman-spectra before the multimodal images in order to check if burning effects occurred. In the presented data set burning effects were not observed.
All animal studies were approved by the governmental commission for animal protection (No. 02-007/13).
Generation of pseudo-HE images out of CARS/TPEF/SHG multimodal images
Spectral histopathology — statistical analysis of Raman-spectra
The data preprocessing in case of the Raman-spectra and statistical modeling were performed using the software package R . The packages used were ‘MASS’ , ‘pls’ , ‘KKNN’  and ‘Peaks’ . Several multivariate statistical tools were applied, such as principal component analysis (PCA), k-means and Weighted k-Nearest Neighbors (KKNN), which shall be described briefly. For dimension reduction and data compression PCA is the most popular and useful tool. The PCA transforms the variables of a dataset into a new set of variables that are linear combinations of the former variables. The new values are called principal components and are ranked according to the variance contributions so that the first principal component provides the highest data variance. Choosing only the first few principal components allows a dimension reduction with a marginal loss of information. K-means is an unsupervised clustering method that arranges the unlabeled dataset into a given number of groups (clusters). It starts with a random distribution of cluster centers and iteratively moves them in order to minimize the total within-cluster variance . KKNN is a non-parametric supervised classification method that is often used because of its simplicity and good performance. To assign a new observation, first the k observations in the training dataset have to be found, which are closest to the new observation. Then the new observation is classified through the majority vote among the k neighbors. KKNN - as an extension of k-Nearest Neighbors algorithm - also takes into account the individual distances of the nearest neighbors to the new observation in the form of weights . The performance of the classification was verified through individual-out-cross-validation (IO-CV).
Individual-Out-Cross-Validation (IO-CV) of a KKNN model; the model has a mean sensitivity of 100 % for the classification between tumor and normal regions, but the mean sensitivity drops to 80.16 % for a differential diagnosis, e.g. for the classification task normal-adenoma-carcinoma
Sensitivity / %
Specificity / %
In this contribution we combined two optical imaging approaches for a fast and precise pathological tissue assessment. The first approach is used to generate an unspecific overview. Therefore multimodal imaging quickly generates large tissue images, which were translated into a pseudo-HE image for identification of suspicious areas. These red flag areas were further analyzed in more detail using Raman-spectroscopy to receive a bio-spectroscopic fingerprint of these suspicious areas. With the help of these fingerprints we were able to construct a model for cancer diagnosis achieving a high sensitivity.
Multimodal imaging – an overview
The first step of the proposed diagnostic workflow (Fig. 3) consisted of recording a multimodal image, which is utilized to derive a pseudo-HE image by multivariate statistics. For exemplification, the first row (A) in Fig. 2 displays multimodal images of mouse colon sections (for sample details see “Methods” Section, sample preparation). False-colors code CARS in red, TPEF in green and SHG in blue. CARS was adjusted to map the CH 2-distribution outlining mostly lipids, while the CARS signal originating from proteins is weaker as their molecular mass-to-methylene group ratio is comparably decreased. The TPEF signal was collected in the spectral range between 426–490 nm, thereby, highlighting the distribution of strong auto-fluorophores such as elastin, NAD(P)H and keratin, while SHG collected at 415 nm visualizes the fibrous collagen network.
Proving the similarity of the information content of non-linear multimodal and classical HE stained images, we applied multivariate statistics in order to generate a computational HE stain image out of the multimodal dataset. The results are exemplified in row B of Fig. 2. For the prediction of pseudo-HE images, we utilized a model combining partial-least-square (PLS) regression with a linear-discriminant-analysis (LDA) that was trained using a single HE image. Subsequent to all multimodal measurements, the samples were HE stained and imaged. The corresponding HE images are depicted in row C of Fig. 2. Apparently, morphological information are readily retrieved from both set of images, i.e. in the pseudo-HE images derived from the multimodal images (Fig. 2 b) and the real HE images (Fig. 2 c). Some information, however, is missing in the pseudo-HE images (Fig. 2 b). In particular the cell nuclei are at certain areas not optimally resolved possibly due to signal intensity variations resulting in a non-uniform brightness of the images. It shall be noted that nuclei may be resolved in future pseudo-HE images if stimulated Raman scattering at DNA specific Raman resonances is applied featuring the required high signal-to-noise ratio . From Fig. 2 b and c differences in the color composition and brightness are visible. For example, the submucosa appears darker within the pseudo-HE images than in the corresponding HE images. Further deviations result from the ethanol washing step during the staining procedure removing soluble components. Consequently, some soluble components were imaged by non-linear multimodal microscopy but were removed in advance to the acquisition of the HE images. The advantage of the generated pseudo-HE images is that the measurement is non-invasive, therefore allowing for further analysis. Here, a HE stain was applied afterwards for comparison, but the employment of other stains or measurement modalities is also possible. Due to its speed and non-invasiveness, the pseudo-HE stain can be applied in a cryosection analysis setting or potentially even in-vivo. Based on the pseudo-HE images a pathologist can identify or define suspicious areas (red flags). These small red flag areas can be further investigated by various approaches like e.g. immunostains, or by other label-free spectroscopic techniques featuring a higher molecular sensitivity than non-linear multimodal imaging. In this contribution we applied Raman-spectroscopy as a second diagnostic technique, offering molecular fingerprint information. The results of the Raman-study are summarized in the following section.
Raman-microspectroscopic imaging – diagnosis
The mean Raman-spectra were subsequently tested for their diagnostic value. Therefore, the Raman-spectra were pre-treated and a Weighted k-Nearest Neighbors (KKNN) classifier was trained and evaluated. In order to estimate the generalization performance of the classifier, the biological variations between the different mice has to be accounted for. To do so, we used an individual-out-cross-validation scheme (IO-CV), where all Raman-spectra of one individual were excluded from training and then predicted. This procedure was iterated and the result is shown in Table. 1. The confusion table shows that the diagnosis of tumor regions, e.g. the classification between tumor regions and normal regions, is accomplished with 100 % mean sensitivity. The differential diagnosis, e.g. the discrimination between adenoma and carcinoma regions, is also possible, but with a lower sensitivity. Here, mis-classifications between adenoma and carcinoma regions occurred. Nevertheless, the overall mean sensitivity for the differential diagnosis is 80.16 % and may be increased in future experiments by improving the detection scheme using, e.g. shifted-excitation Raman-difference-spectroscopy (SERDS) . In a nutshell: the presented workflow allows for a robust and objective diagnosis at least for tumorous and normal epithelial tissue.
In the present study a combination of multimodal imaging and Raman-microspectroscopy is suggested as a fast and precise pathological screening tool. This combination of optical approaches bundles a fast overview technique (multimodal imaging) for identification of suspicious regions (red flags) that are diagnosed by a highly molecular sensitive but rather slow method (Raman-spectroscopy). By applying multivariate statistical methods the multimodal images could be converted into pseudo-HE stain images, which can be analyzed by a pathologist in the same manner as normal HE images. The comparison of pseudo-HE images derived from multimodal images with real HE images proofs that both HE staining and multimodal imaging (TPEF, SHG and CARS) feature similar information. This pseudo-HE stain image can be used by a pathologist to highlight suspicious areas and mark them with red-flags. These red-flag areas can be further analyzed with other techniques featuring a higher sensitivity than non-linear multimodal imaging. In this contribution, the slow but molecular selective technique Raman-spectroscopy was tested for a precise and robust diagnosis. Compared to a gold-standard diagnosis of an experienced pathologist, the Raman-based diagnostics featured 100 % mean sensitivity for the prediction of normal and tumor tissue. The differential diagnosis, the prediction of adenoma, carcinoma and normal epithelium, exhibits a mean sensitivity of around 80 %. Thus, further improvement is required if a differential diagnosis is desired. Nevertheless, the combination of multimodal imaging and Raman-microspectroscopy as a fast, reliable tool to screen large tissue areas and to diagnose normal epithelial tissue from malignant tissue (adenoma, carcinoma) could be demonstrated.
The presented combination of multimodal imaging and Raman spectroscopy can support a pathologist especially in a setting where the preparation of high quality fixated and embedded tissue sections is hardly possible, like for an intraoperative cryosection analysis. As both optical imaging methods are non-invasive, a subsequent staining with conventional HE stain remains feasible, allowing for a direct comparison with the current pathological gold-standard or other stains and methods. The non-invasive character of the methodology introduced within this article also allows for further in-vivo applications. The technical transformation of the combination of multimodal imaging and Raman-spectroscopy into an operation theater for in-vivo studies during an operation is possible and subject to current efforts of us. Such an in-vivo application would increase the possibilities of cancer diagnosis and treatment, since an online-monitoring of certain areas can be performed.
PCA, principal component analysis; SNIP, statistics-sensitive non-linear iterative peak-clipping; KKNN, weighted k-nearest neighbors; IO-CV, individual-out-cross-validation; NAD, Nicotinamide adenine dinucleotide; LDA, linear discriminant analysis model; OCT, optical coherence tomography; SHP, spectral-histo-pathology; CARS, coherent anti-Stokes Raman-scattering; TPEF, two-photon excited autofluorescence; SHG, second harmonic generation; HE, hematoxylin and eosin
The authors thank Tiantian Cui, Cornelia Hüttich and Renate Stöckigt for the excellent technical assistance.
Financial support of the German Research Foundation (DFG) for the research projects PO 563/13-1, PE 602/6-1 and STA 295/9-1 is gratefully acknowledged. Funding of the Bundesministerium für Bildung und Forschung for the project Fiber Health Probe (FKZ: 13N12525, 13N12526) and support of the Carl-Zeiss Foundation are highly acknowledged. The publication of this article was funded by the Open Access Fund of the Leibniz Association.
Availability of data and materials
The data can be requested from Prof. Juergen Popp (email@example.com) and Dr. Thomas Bocklitz (firstname.lastname@example.org) as no repositories are available for this kind of data.
TB, CS, AS, IP, RB, MW, MS, FRG, JP initiated the study. IP, TB, FSS and NV analyzed the HE images. TB and OC did the image analysis, while TB did the chemometrics. SH recorded the multimodal images and NV recorded the Raman spectra. RB coordinated the animal facility. All authors prepared the manuscript and reviewed it. All authors read and approved the final manuscript.
The authors declare that they have no competing interests.
Consent for publication
Ethics approval and consent to participate
All animal studies were approved by the governmental commission for animal protection (No. 02-007/13).
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License(http://creativecommons.org/licenses/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. The Creative Commons Public Domain Dedication waiver(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
- World Health Organization. Global Battle Against Cancer Won’t Be Won with Treatment Alone Effective Prevention Measures Urgently Needed to Prevent Cancer Crisis. PRESS RELEASE No 224. http://www.iarc.fr/en/publications/books/wcr/wcr-order.php. Accessed 13 Jul 2016.
- Bombonati A, Sgroi DC. The molecular pathology of breast cancer progression. J Pathol. 2011; 223(2):308–18.View ArticleGoogle Scholar
- Netto GJ, Saad R. Diagnostic molecular pathology, part 2: Proteomics and clinical applications of molecular diagnostics in hematopathology. Proc (Bayl Univ Med Cent). 2005; 18(1):7.Google Scholar
- Netto GJ, Saad RD, Dysert PA. Diagnostic molecular pathology: current techniques and clinical applications, part i. Proc (Bayl Univ Med Cent). 2003; 16(4):379.Google Scholar
- In: (Popp J, editor.)Ex-vivo and In-vivo Optical Molecular Pathology. Boschstr. 12, 69469 Weinheim, Germany: Wiley-VCH Verlag GmbH & Co. KGaA; 2014.Google Scholar
- Ntziachristos V. Going deeper than microscopy: the optical imaging frontier in biology. Nat Methods. 2010; 7(8):603–14.View ArticlePubMedGoogle Scholar
- Meyer T, Bergner N, Bielecki C, Krafft C, Akimov D, Romeike BFM, Reichart R, Kalff R, Dietzek B, Popp J. Nonlinear microscopy, infrared, and Raman microspectroscopy for brain tumor analysis. J Biomed Opt. 2011; 16:021113.Google Scholar
- Weissleder R. A clearer vision for in vivo imaging. Nat Biotechnol. 2001; 19(4):316–6.View ArticlePubMedGoogle Scholar
- Weissleder R. Molecular imaging in cancer. Science. 2006; 312(5777):1168–71.View ArticlePubMedGoogle Scholar
- Chung VQ, Dwyer PJ, Nehal KS, Rajadhyaksha M, Menaker GM, Charles C, Jiang SB. Use of ex vivo confocal scanning laser microscopy during mohs surgery for nonmelanoma skin cancers. Dermatol Surg. 2004; 30(12p1):1470–8.View ArticlePubMedGoogle Scholar
- Wessels R, De Bruin D, Faber D, Van Leeuwen T, Van Beurden M, Ruers T. Optical biopsy of epithelial cancers by optical coherence tomography (oct). Lasers Med Sci. 2014; 29(3):1297–305.PubMedGoogle Scholar
- Zhou Y, Xing W, Maslov KI, Cornelius LA, Wang LV. Handheld photoacoustic microscopy to detect melanoma depth in vivo. Optics Lett. 2014; 39(16):4731–4.View ArticleGoogle Scholar
- Kitai T, Torii M, Sugie T, Kanao S, Mikami Y, Shiina T, Toi M. Photoacoustic mammography: initial clinical results. Breast Cancer. 2014; 21(2):146–53.View ArticlePubMedGoogle Scholar
- Berezin MY, Achilefu S. Fluorescence lifetime measurements and biological imaging. Chem Rev. 2010; 110(5):2641–84.View ArticlePubMedPubMed CentralGoogle Scholar
- Vogler N, Heuke S, Bocklitz TW, Schmitt M, Popp J. Multimodal imaging spectroscopy of tissue. Annu Rev Anal Chem. 2015; 8. doi:http://dx.doi.org/10.1146/ANNUREV-ANCHEM-071114-040352.
- Meyer T, Schmitt M, Dietzek B, Popp J. Accumulating advantages, reducing limitations: Multimodal nonlinear imaging in biomedical sciences–the synergy of multiple contrast mechanisms. J Biophoton. 2013; 6(11–12):887–904.View ArticleGoogle Scholar
- Patil CA, Kirshnamoorthi H, Ellis DL, van Leeuwen TG, Mahadevan-Jansen A. A clinical instrument for combined raman spectroscopy-optical coherence tomography of skin cancers. Lasers Surg Med. 2011; 43(2):143–51.View ArticlePubMedPubMed CentralGoogle Scholar
- Schwitalla S, Ziegler PK, Horst D, Becker V, Kerle I, Begus-Nahrmann Y, Lechel A, Rudolph L, Langer R, Slotta-Huspenina J, Bader FG, Prazeres da Costa O, Neurath M, Meining A, Kirchner T, Greten FR. Loss of p53 in enterocytes generates an inflammatory microenvironment enabling invasion and lymph node metastasis of carcinogen-induced colorectal tumors. Cancer Cell. 2013; 23(1):93–106.View ArticlePubMedGoogle Scholar
- Schwitalla S, Fingerle AA, Cammareri P, Nebelsiek T, Göktuna SI, Ziegler PK, Canli O, Heijmans J, Huels DJ, Moreaux G, Rupec RA, Gerhard M, Schmid R, Barker N, Clevers H, Lang R, Neumann J, Kirchner T, Taketo MM, van den Brink GR, Sansom OJ, Arkan MC, Greten FR. Intestinal tumorigenesis initiated by dedifferentiation and acquisition of stem-cell-like properties. Cell. 2013; 152(1):25–38.View ArticlePubMedGoogle Scholar
- Heuke S, Vogler N, Meyer T, Akimov D, Kluschke F, Röwert-Huber HJ, Lademann J, Dietzek B, Popp J. Multimodal mapping of human skin. Br J Dermatol. 2013; 169(4):794–803.View ArticlePubMedGoogle Scholar
- Meyer T, Baumgartl M, Gottschall T, Pascher T, Wuttig A, Matthäus C, Romeike BF, Brehm BR, Limpert J, Tünnermann A, Guntinas-Lichius O, Dietzek B, Schmitt M, Popp J. A compact microscope setup for multimodal nonlinear imaging in clinics and its application to disease diagnostics. Analyst. 2013; 138(14):4048–57.View ArticlePubMedGoogle Scholar
- Gonzalez RC, Woods RE, Eddins SL, Vol. 2. Digital Image Processing Using MATLAB. Knoxville: Gatesmark Publishing; 2009.Google Scholar
- Wehrens R, Mevik BH. Pls: Partial Least Squares Regression (PLSR) and Principal Component Regression (PCR). 2007. R package version 2.1-0. http://mevik.net/work/software/pls.html. Accessed 13 Jul 2016.
- R Development Core Team. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing; 2007. R Foundation for Statistical Computing. http://www.R-project.org. Accessed 13 Jul 2016.Google Scholar
- Venables WN, Ripley BD. Modern Applied Statistics with S, 4th. New York: Springer; 2002. http://www.stats.ox.ac.uk/pub/MASS4. Accessed 13 Jul 2016.View ArticleGoogle Scholar
- Hechenbichler KSK.Kknn: Weighted k-Nearest Neighbors. 2013. R package version 1.2-2. http://CRAN.R-project.org/package=kknn. Accessed 13 Jul 2016.
- Morhac M. Peaks: Peaks. 2008. R package version 0.2, https://cran.r-project.org/web/packages/Peaks/index.html. Accessed 13 Jul 2016.
- Hastie T, Tibshirani R, Friedman J. The Elements of Statistical Learning – Data Mining, Inference, and Prediction. Heidelberg: Springer; 2008.Google Scholar
- Bocklitz T, Walter A, Hartmann K, Rösch P, Popp J. How to pre-process Raman spectra for reliable and stable models?Anal Chim Acta. 2011; 704:47–56. doi:http://dx.doi.org/10.1016/j.aca.2011.06.043.
- Ryan CG, Clayton E, Griffin WL, Sie SH, Cousens DR. SNIP, a statistics-sensitive background treatment for the quantitative analysis of pixe spectra in geosience aplications. Nucl Instrum Methods Phys Res Sect B. 1988; 34:396–402.Google Scholar
- Bielecki C, Bocklitz TW, Schmitt M, Krafft C, Marquardt C, Gharbi A, Knösel T, Stallmach A, Popp J. Classification of inflammatory bowel diseases by means of Raman spectroscopic imaging of epithelium cells. J Biomed Opt. 2012; 17(7):076030. doi:http://dx.doi.org/10.1117/1.JBO.17.7.076030.View ArticlePubMedGoogle Scholar
- Diem M, Mazur A, Lenau K, Schubert J, Bird B, Miljković M, Krafft C, Popp J. Molecular pathology via IR and Raman spectral imaging. J Biophoton. 2013; 6(11–12):855–86.View ArticleGoogle Scholar
- Zhang X, Roeffaers MB, Basu S, Daniele JR, Fu D, Freudiger CW, Holtom GR, Xie XS. Label-free live-cell imaging of nucleic acids using stimulated raman scattering microscopy. ChemPhysChem. 2012; 13(4):1054–9.View ArticlePubMedPubMed CentralGoogle Scholar
- Mavarani L, Petersen D, El-Mashtoly SF, Mosig A, Tannapfel A, Kötting C, Gerwert K. Spectral histopathology of colon cancer tissue sections by Raman imaging with 532 nm excitation provides label free annotation of lymphocytes, erythrocytes and proliferating nuclei of cancer cells. Analyst. 2013; 138:4035–9. doi:http://dx.doi.org/10.1039/c3an00370a.
- Notingher I, Hench LL. Raman microspectroscopy: a noninvasive tool for studies of individual living cells in vitro. Expert Rev Med Devices. 2006; 3(2):215–34.View ArticlePubMedGoogle Scholar
- Krafft C, Knetschke T, Siegner A, Funk RHW, Salzer R. Mapping of single cells by near infrared Raman microspectroscopy. Vib Spectrosc. 2003; 32(1):75–83.View ArticleGoogle Scholar
- Knipfer C, Motz J, Adler W, Brunner K, Tesfay GM, Hankel R, Agaimy A, Will S, Braeuer A, Wilhelm NF, Stelzle F. Raman difference spectroscopy: a non-invasive method for identification of oral squamous cell carcinoma. Biomed Optics Express. 2014; 5(9):3252–65.View ArticleGoogle Scholar