Raman spectroscopy for screening and diagnosis of cervical cancer
- 1.2k Downloads
Cervical cancer is the fourth most common cancer in women worldwide and mainly affects younger women. The mortality associated with cervical cancer can be reduced if the disease is detected at the pre-cancer stage. Current best-practice methods include cytopathology, HPV testing, and histopathology, but these methods are limited in terms of subjectivity, cost, and time. There is an unmet clinical need for new methods to aid clinicians in the early detection of cervical pre-cancer. These methods should be objective and rapid and require minimal sample preparation. Raman spectroscopy is a vibrational spectroscopic technique by which incident radiation is used to induce vibrations in the molecules of a sample and the scattered radiation may be used to characterise the sample in a rapid and non-destructive manner. Raman spectroscopy is sensitive to subtle biochemical changes occurring at the molecular level, enabling spectral variations corresponding to disease onset to be detected. Over the past 15 years, there have been numerous reports revealing the potential of Raman spectroscopy together with multivariate statistical analysis for the detection of a variety of cancers. This paper discusses the recent advances and challenges for cervical-cancer screening and diagnosis and offers some perspectives for the future.
KeywordsRaman spectroscopy Cervical cancer Cervical intraepithelial neoplasia (CIN) Low-grade squamous intraepithelial lesion (LSIL) High-grade squamous intraepithelial lesion (HSIL) Cytopathology Histopathology Human papilloma virus (HPV)
Cervical cancer is the fourth most common cancer in women worldwide, accounting for an estimated 528,000 new cases and 266,000 deaths in 2012 . The mortality associated with cervical cancer can be reduced if the disease is detected at the early stages of development or at the pre-malignant stage (cervical intraepithelial neoplasia, CIN). Unlike most other types of cancer, cervical cancer affects mainly younger women, with approximately 60 % of cases occurring in women under 50 years of age. Persistent infection with high-risk human papillomavirus (HPV) (e.g. HPV types 16 and 18) is accepted as the major cause of the development of cervical pre-cancer and cancer . Other risk factors include smoking, immunosuppression, long-term use of oral contraceptives, and low socioeconomic status .
Cervical cancer begins in the basal layer of cells lining the cervix when the normal cells slowly change into pre-cancer cells that have the potential to turn into cancer. The gradual progression of cervical cancer can allow the detection of dysplastic changes before invasive cancer develops, through cervical-cancer screening programmes. These screening programmes are common in developed countries, greatly reducing the mortality from cervical cancer, but are not yet implemented in developing countries because of lack of infrastructure and funding.
Cervical cancer screening and diagnosis
The Pap test
The Pap test, also called the Pap smear, cervical smear, or smear test, is a screening method invented independently by Georgios Papanicolau and Aurel Babeş, but named after Papanicolau. It was introduced in the mid-1940s and is currently the most common screening method for cervical neoplasia and its precursor lesions .
The smear is collected by scraping the internal wall of the cervix with a cervical brush to obtain representative material from the transformation zone where the stratified squamous epithelium of the ectocervix turns into the columnar mucus-secreting epithelium of the endocervix. The cells are then transferred onto a microscope slide, either by the conventional method, in which the cells are spread along the slide immediately after collection and fixed with a spray fixative, or by liquid-based cytology (LBC), in which the cells are transferred immediately after collection into a vial with a fixative solution and subsequently processed to remove debris and transferred to a slide (ThinPrep® (Hologic) or SurePath® (BD)).
Once on a slide, the cells are Pap stained and evaluated under the microscope by a highly trained cytotechnologist or a pathologist according to the Bethesda system .
Cervical cytology is normally graded as negative (negative for intraepithelial lesion or malignancy (NILM)), low-grade squamous intraepithelial lesion (LSIL), or high-grade squamous intraepithelial lesion (HSIL). LSIL may regress, but HSIL is unlikely to do so and may progress to invasive disease. Current guidelines are that LSIL cases are re-tested after six months and HSIL cases are referred to colposcopy .
The advantages of the Pap test are that it is non-invasive, inexpensive, and widely accepted. However, although it can have high specificity of up to 95–98 %, sensitivity has been revealed to vary from 74 to 96 % as a result of sampling, technical, and/or inter-observer errors mainly associated with the subjectivity of the cytological screening .
Semi-automated screening systems consist of an automated microscope coupled to a workstation running image-processing algorithms. Slides are scanned initially, and cells of interest are separated from the background of inflammatory cells, cellular debris, or overlapping cell clusters. Image-segmentation algorithms perform a separation of the nuclei from the cytoplasm of the cells, enabling the calculation of nuclear size, nucleus-to-cytoplasm ratio, or even definition of the texture of the observed object. However, neither of the currently available systems, the FocalPoint™ GS Imaging system (BD) or the ThinPrep™ Imaging system (Hologic), provides fully automated screening without human intervention at some stage. The final decision still lies with the cytologist, resulting in the same subjectivity problem as in manual screening. The MAVARIC trial  revealed no improvement in sensitivity or specificity of automated screening when compared with manual screening, nor in cost-effectiveness despite a 60–80 % increase in productivity for automated screening.
HPV-testing trials have recently been conducted to evaluate their effectiveness for primary and secondary screening and for proof-of-cure for safe and cost-effective lengthening of the cervical-screening interval as a result of high negative predictive values . Many studies have revealed that HPV testing has higher sensitivity (>95 %) but lower specificity (~84 %) than cytology [9, 10]. HPV DNA tests, including Hybrid Capture 2 (Qiagen), Cervista HPV HR and Cervista HPV 16/18 (Hologic), and cobas HPV (Roche) assays, identify high-risk HPV oncogene expression, whereas HPV E6/E7 mRNA-based assays, for example APTIMA HPV assay (Gen Probe), identify the messenger RNA of two HPV viral oncogenes, E6 and E7. However, these tests are expensive, time-consuming, and provide no information on cervical cytopathology. Over the last decade prophylactic HPV vaccines have been developed, but, despite the introduction of these vaccines, there is still a need for regular cervical screening, because the vaccines do not protect against all oncogenic HPV types. Additionally, some women may not benefit from the vaccines if there is a pre-existing HPV infection or they do not receive the complete number of doses. After vaccination, women must still have routine Pap tests which can detect abnormal cervical growth regardless of what HPV type causes it to develop .
An abnormal Pap smear is followed by colposcopy, biopsy, and histopathology. Microscopic evaluation by a pathologist of a stained tissue section obtained from a biopsy is currently regarded as the best practice in cancer diagnostics, and is called pathological histology or histopathology.
Current best-practice methods for detection of cervical cancer and pre-cancer are therefore limited, and there is an unmet clinical need for new objective screening or diagnostic tests.
Vibrational spectroscopy analyses vibrations within a molecule; the vibrations are characteristic of the molecular structure, resulting in a spectroscopic “fingerprint” . The exact energy required to excite a molecular vibration depends on the masses of the atoms involved in the vibration and the strength of the chemical bonds between these atoms, which may be affected by molecular structure, molecular interactions, and the chemical environment of the molecule.
Raman spectroscopy is based on inelastic light scattering, in which the sample is illuminated by monochromatic laser light and interactions between the incident photons and molecules in the sample result in scattering of the light. The coupling of the light generates vibrations within the sample which are characteristic of the chemical structure. The energy of the scattered light is reduced by an amount equal to the vibrational energy. As a result, the positions, relative intensities, and shapes of Raman bands carry in-depth information about the molecular composition of the sample.
Cells and tissues contain many biochemical components, including DNA, RNA, proteins, lipids, and carbohydrates, and the Raman spectra of these samples are a superposition of the contributions from each individual biochemical component. It follows that Raman spectroscopy can provide a “biochemical fingerprint” of the complete genome, proteome, and metabolome of the cell or tissue . Additional analyses can be performed subsequently on the cell or tissue samples, including staining, immunocytochemistry, etc., because Raman spectroscopy can be performed in a label-free, non-destructive manner. Over the past 15 years, Raman spectroscopy has been used for the diagnosis of a wide range of cancers, including breast, prostate, oesophageal, colon, lung, oral, and cervical cancer, with excellent sensitivity and specificity being reported [18, 19, 20].
Raman peak assignments
C–C twisting mode of phenylalanine (proteins)
C–C twisting mode of tyrosine and phenylalanine
Thymine, guanine (DNA and/or RNA)
C–N stretching in adenine and lipids
Symmetric breathing of tryptophan (protein)
Uracil, thymine, cytosine (ring-breathing modes in the DNA and/or RNA)
PO2 stretching in DNA, tyrosine
Ring breathing in tyrosine and proline (proteins)
C–C stretching mode of proline and valine
C–C aromatic-ring stretching in phenylalanine
C–H bending mode in phenylalanine, C–N stretching in proteins
PO2 symmetric stretching (DNA and/or RNA)
C–N stretching in proteins; C–O stretching in carbohydrates
C–C and C–N stretching of proteins and/or lipids
C–H in plane-bending mode of tyrosine and phenylalanine; cytosine, guanine
C–C6H5 stretching mode in tryptophan, phenylalanine;
Amide III (C–N stretching, N–H bending, proteins), PO2 asymmetric stretching (DNA and/or RNA)
CH3 and/or CH2 twisting mode of collagen and lipid
Guanine (DNA and/or RNA), CH def. in proteins and carbohydrates
CH (CH2) bending mode in proteins and lipids
Amide II (N–H bending, C–N stretching, proteins); adenine, guanine (DNA and/or RNA)
Adenine, guanine (DNA and/or RNA); C=C bending mode of phenylalanine
C=C phenylalanine, tyrosine, and tryptophan
Amide I (C=O stretching, C–N stretching, and N–H bending, proteins)
CH2 symmetric stretching (lipids)
CH2 and CH3 symmetric stretching (lipids)
CH3 symmetric stretching (lipids)
Raman spectroscopy for cytopathology
Interestingly, the Raman spectra in Fig. 4 were recorded directly from Thinprep slides prepared according to current cytopathology laboratory standard procedures, apart from the Pap stain. The x,y co-ordinates of each recorded cell were saved and, after Raman recording at 532 nm, the slides were Pap stained and each cell was re-visited to verify whether it was a superficial, intermediate, parabasal, or white blood cell. Spectroscopic substrates, for example calcium fluoride, are commonly used for research purposes and, although they reduce the presence of confounding contributions of the substrate [22, 23, 24], they are substantially more expensive and thus not really a viable choice for population-screening applications including cervical-cancer screening. Although 785 nm is commonly used for biological applications of Raman spectroscopy, glass has a strong background at this wavelength; so to use glass as a substrate for Raman spectroscopy, spectra must be recorded using shorter wavelengths, for example 532 nm.
In the early 1990s several infrared-spectroscopy studies on cervical cytopathology samples were reported [25, 26, 27]. However, spectra were recorded from cell pellets rather than from individual cells and several confounding factors, including the presence of metaplastic cells, endocervical columnar cells, polymorphs, blood, cervical mucus, and debris, were identified [28, 29, 30, 31, 32, 33]. Since then, most probably as a result of the problems with confounding factors, there have been relatively few studies using vibrational spectroscopy for cervical cytopathology.
Raman spectroscopy was used by Rubina et al.  to distinguish between normal and cervical-cancer cytology samples. Cytology samples were treated with red blood cell lysis buffer before Raman spectral acquisition because the spectra from the cervical-cancer samples were dominated by blood features. A relatively low classification accuracy of 80 % was reported, most probably caused by sample heterogeneity because of the use of cell pellets rather than recording Raman spectra from individual cells.
A recent study from our group  reported on discrimination between negative cytology and CIN cytology samples using Raman spectroscopy. Importantly, all data was recorded from cells on glass ThinPrep slides, commonly used in cytopathology laboratories. The study addressed many of the problems involved in recording Raman spectra from ThinPrep cervical cytology samples, and described a new method to clear blood-residue contamination before Raman recording based on pre-treatment of the slides with hydrogen peroxide. This was revealed to minimise variability and to result in the collection of highly reproducible data with excellent discrimination between negative cytology and CIN cytology.
Raman spectroscopy for detection of HPV infection
HPV infection is accepted as one of the major risk factors associated with cervical cancer, and detection of HPV infection is being introduced as a routine screening method. Incorporation of the virus into the cell induces significant changes in the biochemistry which should also be identifiable using Raman spectroscopy.
Raman microspectroscopy has been used to distinguish between primary human keratinocytes (PHK), PHK cells expressing the E7 gene of HPV16 (PHK E7), and cervical-cancer cells expressing HPV16 (CaSki). The mean Raman spectra revealed variations in DNA and protein consistent with HPV gene expression and neoplasia. Using principal component analysis (PCA), Raman spectroscopy was revealed to discriminate between PHK and CaSki cells with a sensitivity of 93 % and a specificity of 93 %, and between PHK and PHK E7 cells with a sensitivity of 93 % and a specificity of 80 % .
Ostrowska et al.  used both infrared absorption and Raman spectroscopy to study a range of cervical-cancer cell lines. HPV-negative (C33a) and low-HPV-copy-number (SiHa with one or two copies) cell lines were revealed to be biochemically similar, but significantly different from mid (HeLa) and high (CaSki)-HPV-copy-number cervical-cancer cell lines. The main variations were observed for protein, nucleic acid, and lipid, and were confirmed by both mean spectra and PCA analysis. Notably, the application of multivariate partial-least-squares regression analysis, with HPV copy number as target, revealed that the dataset can be used to evaluate the degree of HPV infection on the basis of the spectral profile of the cells.
A study by Vargis et al.  used both cell lines and cytology samples to investigate the potential of Raman microspectroscopy to detect the presence of HPV. Classification accuracies of 89–93 % were achieved for discrimination between a HPV-negative normal-human-keratinocyte cell line (NHEK), a HPV-negative cervical-cancer cell line (C33a), and HPV-positive cervical-cancer cell lines (HeLa and SiHa). A classification accuracy of 98.5 % was achieved for discrimination between HPV-positive and HPV-negative cytology samples.
Raman spectroscopy for histopathology
Krishna et al. studied formalin-fixed cervical tissues using both Raman and FTIR spectroscopy. Normal and malignant tissues could be distinguished by differences in protein, lipids, and nucleic-acid peaks and stronger amide III assignments, suggesting disordered, helical secondary-protein structure, in malignant conditions . Formalin-fixed paraffin-preserved (FFPP) cervical-tissue sections were also investigated by Lyng et al. . The underlying biochemical changes associated with cervical precancer and cancer were revealed to be caused by a reduction in glycogen and an increase in nucleic acids. The loss of differentiation, together with increased proliferation, in pre-cancer results in reduced levels of glycogen, because normal cervical cells accumulate glycogen as they mature. A Raman mapping study using frozen tissue sections revealed that Raman spectroscopy could distinguish normal cervical tissue from invasive cervical-cancer tissue, mainly on the basis of collagen bands and CH stretching bands . Tan et al.  used Raman spectral mapping and hierarchical cluster analysis (HCA) to differentiate between normal squamous epithelium and CIN2 in FFPP tissue sections, and it was revealed that the Raman spectra associated with the CIN2 lesion clustered predominantly with those of the basal epithelial cells of the normal squamous epithelium, suggesting that the cells of these regions share common biochemical profiles. Spectral features responsible for their differentiation were associated with the amide I and amide III bands.
More recently, FFPP tissue sections from cervical biopsies classified as NILM, LSIL, or HSIL were analysed by Raman spectral mapping . Together with K-means cluster analysis (KMCA), Raman mapping was able to differentiate the NILM cervical tissue into three layers including stroma, basal and/or parabasal, and superficial layers, characterised by spectral features of collagen, DNA bases, and glycogen, respectively. In the LSIL and HSIL samples, KMCA clustered regions of the superficial layer with the basal layer. Using PCA, biochemical changes associated with disease were also observed in normal areas of the abnormal samples, where morphological changes were not apparent, providing a clear indication of the potential of Raman spectroscopy to identify biochemical changes associated with the initial stages of the disease, rather than just the morphological changes associated with later-stage disease and current clinical diagnosis.
Raman spectroscopy for in-vivo applications
Mahadevan-Jansen et al. [44, 45] first revealed the potential of Raman spectroscopy for in-vivo detection of cervical cancer and pre-cancer and developed a fibre-optic probe for in-vivo measurements. Increases in phospholipids and DNA, ~1330, 1454, and 1650 cm−1, were associated with progression to high-grade dysplasia [44, 45]. Improvements in the overall classification accuracy of Raman spectroscopy from 88 % to 94 % were achieved by including information on menopausal status and menstrual cycle . Similarly, consideration of disease history and proximity to dysplastic lesions was found to result in disease classification accuracy of 97 % . Further studies investigated the effect of race, ethnicity, body mass index, parity, and socioeconomic status on Raman spectra from patients with a normal cervix, and concluded that only body mass index and parity resulted in significant variations of spectral profiles . Their effect on dysplasia and cancer has not been assessed, however. The potential of high-wavenumber (2800–3700 cm−1) Raman spectroscopy for in-vivo detection of cervical pre-cancer has been investigated by Mo et al. and Duraipandian et al. [49, 50]. Significant differences in Raman bands of lipids at 2850 and 2885 cm−1, proteins at 2940 cm−1, and the broad Raman band of water at 3400 cm−1 were observed in normal and dysplastic cervical tissue, with a sensitivity of 93.5 % and specificity of 97.8 % achieved for identification of dysplasia . Simultaneous fingerprint and high-wavenumber Raman spectroscopy has been revealed to outperform fingerprint or high-wavenumber Raman spectroscopy alone, resulting in a sensitivity of 85.0 %, specificity of 81.7 %, and overall diagnostic accuracy of 82.6 %. Raman spectral differences between normal and dysplastic cervical tissue were observed at 854, 937, 1001, 1095, 1253, 1313, 1445, 1654, 2946, and 3400 cm−1, mainly related to proteins, lipids, glycogen, nucleic acids, and water content in tissue .
The potential of Raman spectroscopy as a truly label-free, objective, automatable diagnostic technique has been well established through numerous research studies of numerous pathologies, both in vivo and ex vivo. Disease diagnostics have long relied on visual differences in tissue appearance, aided in modern histopathology and cytology by optical stains and microscope technology. However, such approaches, based on changes in tissue and cell morphology, often reveal the later stages of disease development, rather than the underlying biochemical changes associated with disease onset or aetiology. Raman spectroscopy provides a signature or fingerprint of the biological sample, based on the biochemical constituents, and subtle changes to the composition associated with disease or external insult can be identified with high sensitivity with the aid of multivariate statistical analysis. As an optical technology, it can be applied microscopically or endoscopically, for ex-vivo or in-vivo applications, the former including cytopathology or histopathology. Signatures of viral infection can also be clearly identified, indicating that the technique could ultimately compete with costly viral screening programmes, notably with the same instrumentation and an integrated screening procedure.
Advantages and disadvantages of current screening and/or diagnostic methods and Raman spectroscopy
Pap test and/or cytology
Well accepted screening method
High negative predictive values
Well-accepted best-practice method
Definitive diagnosis of tumour stage
Stromal invasion can be determined
Tumour margins can be determined
Can be used by non-specialists with suitable diagnostic algorithms
Low operating costs because no reagents required; based on a fingerprint of the biochemical composition
Can be used ex vivo or in vivo
Can be used for cells and tissues
Can provide information on HPV status
High spatial resolution enabling subcelluar imaging
Requires clinical expertise and experience
Reagents and instrumentation required
Must be confirmed with colposcopy and histopathology
Reagents and instrumentation required
Different tests for HPV DNA and mRNA
Does not provide information about cellular morphology and/or abnormality
Requires clinical expertise and experience
Reagents and instrumentation required
Pre-malignant lesions difficult to distinguish from benign conditions
Long spectral-acquisition times
Multivariate data analysis needed to extract information from the spectral data
However, despite the obvious potential, vibrational spectroscopic techniques have had limited if any translation into the clinical environment. A critical assessment of the challenges facing the translation of spectroscopic methods to the clinic has recently been presented . “Spectropathology for the Next Generation: Quo Vadis?” summarises the discussion sessions of the SPEC 2014 conference, the 8th in the series of biennial flagship conferences in the field, which were led by members of the International Advisory Board. Although the potential of the techniques has been well demonstrated in the research environment, and many of the technical challenges associated with sample presentation, data acquisition, processing, and analysis have been addressed, it is clear that the lack of studies of the scale of clinical relevance is an obstacle to their credibility and uptake. Widespread engagement of the medical community and commitment from instrument manufacturers, and the resources required for large-scale clinical trials, may depend on the availability of such large-scale studies. In this context, the identification of strategic, achievable target applications is recommended.
Because of the high throughput of established screening programmes, cervical cytopathology is potentially such a strategic target application. A crucial consideration is how the spectroscopic technique could fit into the current workflow. Notably, the Pap smear procedure introduces cytological stains which can interfere with the Raman acquisition, introducing substantial fluorescence background and/or photodegradation depending on the wavelength used. With automated spectral acquisition, however, the implementation could be similar to the Thinprep Imager (Hologic) or the Focal Point GS Imaging system (BD), which use image-processing algorithms to automatically review liquid-based cytology Thinprep and SurePath slides. Raman spectroscopy is reagent free, so costs should be favourable when compared with the imaging systems or with manual scoring where personnel costs could be reduced. A recent study by our group  has revealed that glass ThinPrep slides can be used for Raman spectral recording in place of spectroscopic substrates, for example calcium fluoride substrates. These substrates reduce the presence of confounding contributions of the substrate, but they are substantially more expensive than the glass slides used in the cytopathology laboratory.
Notably, the same data acquisition and screening procedure can be used for identification of HPV infection, because the technique is based on biochemical signatures, and thus, with appropriately trained spectral databases and data-mining procedures, the spectral profiling could provide an integrated analysis of health status and risk of developing HSIL or cervical cancer, combining the advantages of the currently used cytological and HPV screening standards with higher sensitivity, specificity, and throughput.
Similar potential advantages of Raman-spectroscopic approaches for histopathology can be identified, although it is recognised that current mapping and/or imaging times of large areas of tissue followed by current pre and post-data-processing procedures need to be improved . Although substantial progress has been made, there is much to be done in terms of standardising procedures and protocols. The demands on the ability to rapidly scan large areas of tissue probably currently favour the use of FTIR rather than Raman spectroscopy for such applications. In terms of tissue-processing procedures, there remains much debate, although consistency with current clinical practice probably favours the use of FFPP tissue samples. Notably, analyses of archived tissue libraries may add much to understanding of disease progression and patient prognosis. It has been revealed that it is not necessary to remove the paraffin to obtain usable spectral information . Leaving the paraffin in place reduces scattering artefacts and effects of further variable removal of aromatic solvent soluble components. However, it may be argued that greater consistency of spectral information is achieved when sections are deparaffinised. Deparaffinising also enables post-staining of the sections, although it has been revealed that the efficiency of the deparaffinisation process can depend on the tissue pathology .
In-vivo applications are reliant on the further development of spectroscopic probes, which have already been used for cervical applications [44, 45, 46, 47, 48, 49, 50]. Raman spectroscopy could be a potential candidate for an adjunct tool for “screen and treat” approaches for low-resource countries. Patients are screened by visual inspection during colposcopy and treated immediately by cryotherapy if required. Raman spectroscopy could be used to improve the poor sensitivity of colposcopy in identifying cervical cancer and pre-cancerous lesions.
However, much work still remains before these techniques could be translated into standard clinical practice. Many studies have used relatively small sample sizes, and as such may be biased. Validation in large multi-centre studies is needed, using real-world cytopathology and histopathology samples from screening programmes and colposcopy clinics. Large-scale clinical trials are needed to obtain the large volumes of data necessary for the development of robust classification algorithms.
If spectroscopy can be shown to be as good as, or better than, the methods of choice in current use, then there is great potential for these techniques to be used as an alternative or an adjunct to the current methods. The advantages would be higher accuracy, higher throughput, and reduced workload for the cytologist and/or pathologist, and higher accuracy and chance of earlier detection for the patient. It is important, however, to establish standard operating procedures, through such networks as the UK EPSRC Network CLIRSPEC (www.clirspec.org) and the EU COST Action Raman4Clinics (http://www.cost.eu/domains_actions/bmbs/Actions/BM1401), to engage spectroscopic-instrument and medical-device industries to optimise data-collection efficiencies, and to actively engage the medical and clinical communities to encourage uptake and translation of the technology.
The authors acknowledge funding from Enterprise Ireland co-funded by the European Regional Development Fund (ERDF) and Ireland’s EU Structural Funds Programme 2007–2013, CF2011 1045, the Health Research Board Collaborative Applied Research Grant, CARG2012/29, and Dublin Institute of Technology Fiosraigh Research Excellence Award.
- 1.Ferlay J, Soerjomataram I, Ervik M, Dikshit R, Eser S, Mathers C, Rebelo M, Parkin DM, Forman D, Bray F (2012) GLOBOCAN, v1.0, Cancer incidence and mortality worldwide: IARC Cancer Base No. 11, 2013. International Agency for Research on CancerGoogle Scholar
- 4.Koss LG, Melamed MR (2006) Koss’ diagnostic cytology and its histopathologic bases, 5th edn. Lippincott Williams & Wilkins, PhiladelphiaGoogle Scholar
- 12.Dalla Palma P, Giorgi Rossi P, Collina G, Buccoliero AM, Ghiringhello B, Lestani M et al (2008) The risk of false-positive histology according to the reason for colposcopy referral in cervical cancer screening: a blind revision of all histologic lesions found in the NTCC trial. Am J Clin Pathol 129:75–80CrossRefGoogle Scholar
- 16.Diem M (1993) Introduction to modern vibrational spectroscopy. Wiley, New YorkGoogle Scholar
- 22.Rashid N (2013) Raman microspectroscopy for the characterization of cervical cancer, PhD thesis, Dublin Institute of TechnologyGoogle Scholar
- 24.Kerr LT, Byrne HJ, Hennelly BM (2015) Optimal choice of sample substrate and laser wavelength for Raman spectroscopic analysis of biological specimens. Anal Methods (accepted)Google Scholar