Method for identification of low grade cervical cytology cases likely to progress to high grade/cancer

Abstract

The present invention provides a method of using Raman spectroscopy for identification of low grade cervical cytology cases likely to progress to high grade/cancer. The Applicant has found that high quality Raman spectra can be successfully acquired from morphologically normal appearing cells from negative, LSIL and HSIL Thinprep® specimens and different grades of cervical pre-cancer can be separated with good sensitivities and specificities. Raman spectroscopy can further identify different categories of the LSIL cases i.e., whether they are likely to regress to negative or progress to HSIL cytology.

Claims

1. A method for distinguishing between low-grade squamous intraepithelial lesions (LSIL) that are likely to progress to high-grade squamous intraepithelial lesions (HSIL) and LSIL that are likely to regress to negative, the method comprising: providing a biological sample comprising cervical cells; obtaining a Raman spectrum for the biological sample; analysing the Raman spectrum to determine whether the Raman spectrum falls within one or more predefined classes of cells, wherein the one or more predefined classes of cells comprise cells comprising LSIL that are likely to progress to HSIL and cells comprising LSIL that are likely to regress to negative wherein analysing the Raman spectrum to determine whether the Raman spectrum falls within one or more predefined classes of cells comprises using a classification model built using a database of reference Raman spectra.

2. The method as claimed in claim 1 wherein the classification model comprises multivariate statistical analysis.

3. The method as claimed in claim 2 wherein the multivariate statistical analysis is selected from the group consisting of Partial Least Squares Discriminant Analysis (PLS-DA), principal component analysis, linear discriminant analysis, support vector machines, and random forest.

4. The method as claimed in claim 1 wherein the biological sample is obtained during a Pap smear.

5. The method as claimed in claim 1 wherein the biological sample is processed using a liquid-based cytology test or conventional cytology.

6. The method as claimed in claim 1 wherein the biological sample comprises morphologically normal looking cells and the Raman spectrum is obtained for the morphologically normal looking cells.

7. The method as claimed in claim 1 wherein the biological sample comprises superficial and intermediate epithelial cells.

8. The method as claimed in claim 1 wherein the Raman spectrum is obtained from cell nuclei of the cells from the biological sample.

9. The method as claimed in claim 1 wherein the Raman spectrum is obtained using a low resolution Raman spectroscopy device.

10. The method as claimed in claim 1 wherein the step of analysing the Raman spectrum comprises analysing Raman peaks selected from one or more of the following: 482, 621, 728, 828, 855, 936, 957, 1092, 1176, 1210, 1338, 1422, 1450, 1578, 1610, 1619, 1669 cm.sup.−1.

11. The method as claimed in claim 1 wherein the step of analysing the Raman spectrum comprises analysing Raman peaks selected from one or more of the following: 785, 936, 1000, 1046, 1097, 1124, 1238, 1340, 1575 and 1652 cm.sup.−1.

12. The method as claimed in claim 1 wherein the method does not include a separate step of screening for the presence of HPV DNA and HPV mRNA.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) The present application will now be described with reference to the accompanying drawings in which:

(2) FIG. 1 shows (A) Raman spectra and (B) Principal Components Analysis scores plot showing good discrimination between normal (black), LSIL (light grey) and HSIL (dark grey) cells;

(3) FIG. 2(a) is an unstained image ×10 and (b) Pap stained image ×10 of negative cytology specimen; in the Pap stained image (b) superficial cells stained orange to pink in color (indicated by arrows), and intermediate cells stained turquoise green to blue in color (indicated by arrowheads);

(4) FIG. 3(a) shows the Mean Raman spectra±1 standard deviation (SD) acquired from the intermediate cells of negative cytology ThinPrep® specimen (n=17) and morphologically normal looking intermediate cells of high-grade squamous intraepithelial lesion (HSIL) ThinPrep® specimens (n=15) collected during routine Pap smear, (b) Mean Raman spectra±1 SD acquired from the superficial cells of negative cytology specimen (n=17) and morphologically normal looking superficial cells of HSIL cytology ThinPrep® specimens (n=12) collected during routine Pap smear;

(5) FIG. 4 shows the different Raman spectra±1 standard deviation (SD) obtained from (a) intermediate cells of negative and high-grade squamous intraepithelial lesion (HSIL) specimen (b) superficial cells of negative and HSIL specimen. It has to be noted that in the HSIL specimen Raman spectra were acquired from morphologically normal looking intermediate and superficial cells;

(6) FIG. 5 shows PLS loadings (LVs) of the developed PLS-DA model for the dataset obtained from (a) intermediate cells, (b) superficial cells and (c) mixed intermediate and superficial cells. (Latent variables (LVs), partial least squares discriminant analysis (PLS-DA));

(7) FIG. 6 Scatter plot of the significant latent variables (LVs) obtained from the Raman spectral dataset of mixed intermediate and superficial cells of negative and high-grade squamous intraepithelial lesion (HSIL) cytology specimen;

(8) FIG. 7 is a scatter plot of the posterior probability values calculated from the Raman dataset obtained from (a) intermediate cells, (b) superficial cells and (c) mixed intermediate and superficial cells of negative and HSIL cytology specimens;

(9) FIG. 8(a) Mean Raman spectra±1 standard deviation (SD) acquired from the negative, low-grade squamous intraepithelial lesion (LSIL) and high-grade squamous intraepithelial lesion (HSIL) ThinPrep® specimens (n=39) collected during routine Pap smear, (b) Difference Raman spectra±1 standard deviation (SD) of (a) negative−low-grade squamous intraepithelial lesion (LSIL) (b) negative−high-grade squamous intraepithelial lesion (HSIL);

(10) FIG. 9(a) Mean Raman spectra±1 standard deviation (SD) acquired from the low-grade squamous intraepithelial lesion (LSIL) ThinPrep® specimens (n=12) with different HPV test results (a) Cobas-negative and Aptima-negative, (b) Cobas-positive and Aptima-negative, (c) Cobas-positive and Aptima-positive; (b) Difference Raman spectra±1 standard deviation (SD) calculated from LSIL ThinPrep® specimens with different HPV test results (a) Cobas-negative Aptima-negative−Cobas-positive Aptima-negative (b) Cobas-negative Aptima-negative−Cobas-positive Aptima-positive (c) Cobas-positive Aptima-negative−Cobas-positive Aptima-positive;

(11) FIG. 10 PLS loadings (LVs) of the multi-class PLS-DA model developed from the dataset of true negative, low-grade squamous intraepithelial lesion (LSIL) and high-grade squamous intraepithelial lesion (HSIL) cytology specimens. (Latent variables (LVs), partial least squares discriminant analysis (PLS-DA));

(12) FIG. 11 Scatter plot of the posterior probability values obtained from the dataset of true negative, low-grade squamous intraepithelial lesion (LSIL) and high-grade squamous intraepithelial lesion (HSIL) cytology specimens; and

(13) FIG. 12 is a schematic of a flow chart showing the steps involved in the method of the present invention.

DETAILED DESCRIPTION

(14) The words comprises/comprising when used in this specification are to specify the presence of stated features, integers, steps or components, but do not preclude the presence or addition of one or more other features, integers, steps, components or groups thereof. In certain embodiments, the term comprises or comprising may be understood to mean includes or including, i.e. other components are also present. In alternative embodiments, the term comprises or comprising may be understood to mean consists of or consisting of, i.e. no other components are present.

(15) Typically the terms “subject” and “patient” are used interchangeably herein. The subject is typically a mammal, more typically a human.

(16) The present invention provides a system and method using Raman spectroscopy for accurately discriminating between normal (NAD), LSIL and HSIL Thinprep® cytology samples. Specifically, a system and method using Raman spectroscopy to successfully discriminate between the spectra of normal cells (blue), and abnormal cells, LSIL (green) and HSIL (red) (FIG. 1) are provided. The Inventors surprisingly found that Raman spectra from morphologically normal appearing cells from HSIL cases could be reliably discriminated from cells from normal cases. This was an important finding which meant that it was not necessary to find the rare, abnormal-appearing HSIL cells on each abnormal slide in order to distinguish the case as abnormal. This proved to be a major benefit as finding the rare abnormal cells on the unstained slide preparations was very difficult and time consuming.

(17) The following Examples describe the invention.

Example 1

Use of Superficial or Intermediate Epithelial Cells for Discrimination of Negative and HSIL Cytology Cases

(18) Materials and Methods

(19) Sample Collection and Processing

(20) True negative cervical liquid based cytology samples were obtained from the cytology laboratory, Coombe Women and Infants University Hospital (CWIUH), Dublin, Ireland. HSIL cytology specimens were collected during the routine Pap smear from the Colposcopy clinic, CWIUH, Dublin, Ireland. The collected smears were processed via the ThinPrep® method. This study was approved by the Research Ethics Committee CWIUH. The cells were collected from the cervix using a cyto brush and then rinsed in the specimen vial containing PreservCyt transport medium (ThinPrep® Pap Test; Cytyc Corporation, Boxborough, Mass.). The labelled ThinPrep® sample vial was sent to the cytology laboratory equipped with a ThinPrep® processor. All samples were prepared using a ThinPrep® 2000 processor (Hologic Inc., Marlborough, Mass. 01752). The ThinPrep® processor homogenizes the sample by spinning either the filter (T2000) or the vial (T3000), creating shear forces in the fluid that are strong enough to disaggregate randomly joined material, break up blood, mucus and non-diagnostic debris while keeping true cell clusters intact. The cells were then collected onto the membrane of the TransCyt filter and further transferred onto a glass slide to create a monolayer deposit of cells (˜20 mm in diameter). The slide was transferred into a fixative bath of 95% ethanol automatically. In total, 32 unstained cytology samples on ThinPrep® slides (17 negative and 15 HSIL) were obtained and subjected to Raman spectroscopic analysis. Before Raman measurement, each slide was pre-treated with hydrogen peroxidase (H.sub.2O.sub.2) to remove any contaminating blood and debris.

(21) Raman Instrumentation

(22) Raman spectra were recorded using a HORIBA Jobin Yvon XploRA® system (Villeneuve d'Ascq, France), incorporating an Olympus microscope BX41 equipped with a ×100 objective (MPlanN, Olympus, NA=0.9). The system consists of a 532 nm diode laser, 1200 lines/mm grating and an air-cooled CCD detector (Andor, 1024×256 pixels). The system was wavelength calibrated to the 520.7 cm−.sup.1 spectral line of silicon and also intensity-calibrated using a relative intensity correction standard (NIST 2242). A total of ˜770 Raman signals were measured from the ThinPrep® specimens of 32 patients (17 negative and 15 HSIL). From each slide, 15 to 20 intermediate and superficial epithelial cells were randomly selected and good quality Raman spectra were obtained with an integration time of 30 sec and 2 accumulations to improve the signal to noise ratio. The laser power on the sample was ˜1 mW. The image of the Raman measured cells were recorded and x- and y-coordinates of the measured cells were also stored. After the Raman spectral acquisition, the samples were Pap stained and each recorded cells was re-visited using the stored x- and y-co-ordinates to verify whether the cells were from the intermediate layer or superficial layer.

(23) Data Analysis

(24) All the recorded Raman spectra were corrected for the glass background using a linear least-squares method with non-negative constraints (NNLS). The least-squares model was developed using the basis spectra obtained from the pure glass slides and selected pure biochemicals (e.g., actin, collagen, RNA, DNA, etc.) that approximate the biochemical composition of cervical cells. The Raman dataset has also been corrected for the baseline and then vector normalized. The Raman data was mean-centered and then subjected to partial least squares discriminant analysis (PLS-DA) diagnostic algorithm together with leave-one-out, cross-validation for discriminating negative cytology and HSIL cytology. PLS-DA establishes a regression model between the Raman spectral dataset and the class membership. The class membership is a dummy dichotomous variable, coded with 0s and 1s to represent each observation. PLS-DA rotates the latent variables to obtain maximum separation among the classes. The analysis was performed using the PLS toolbox (Eigenvector Research, Wenatchee, Wash.) in the Matlab® (Mathworks Inc., Natick, Mass.) scripting environment.

(25) Results

(26) In this study, Raman spectra were acquired from the nuclei of 15 to 20 randomly selected cells from each ThinPrep® cervical cytology specimen. Here, Raman spectra were measured from the intermediate and superficial cells of negative cytology specimens and from morphologically normal appearing intermediate and superficial cells of HSIL cytology specimens. FIG. 2 shows the intermediate and the superficial cells collected from the negative cytology specimen. The intermediate and the superficial cells are indistinguishable in unstained slides ((FIG. 2(a)). After the Pap staining, the superficial cells are orange to pink in color, and the intermediate cells are turquoise green to blue in color (FIG. 2(b)). Raman spectra were recorded from the unstained slides. The mean normalized Raman spectra obtained from the intermediate (negative (n=177), HSIL (n=258), FIG. 3a) and the superficial (negative (n=176), HSIL (n=157) FIG. 3b) cells are shown in FIG. 3. The Raman spectra obtained from the intermediate (maximum standard deviation (SD)=±0.0049) and superficial (maximum SD=±0.0056) cells are quite consistent. However larger spectral variability have been reported for the superficial layer due to the variation in the glycogen content associated with women's age, hormonal levels and menstrual cycle. The consistency observed in our Raman spectra of superficial cells is because the Raman signals were measured from the cell nucleus; whereas high spectral variability due to the glycogen content is seen if the Raman spectra are taken from the cytoplasm. The measured Raman spectrum contains complex, overlapped spectral signatures in relation to the tissue biochemistry. Weak Raman peaks can be observed at 482 cm.sup.−1 (glycogen), 621 and 644 cm.sup.−1 (proteins), 728 and 784 cm.sup.−1 (DNA), and 828 cm.sup.−1 (DNA/RNA). More prominent peaks are visible in the vicinity of 855 and 936 cm.sup.−1 (glycogen and proteins), 957 cm.sup.−1 (DNA), 1004 cm.sup.−1 (phenylalanine), 1035 cm.sup.−1 (proteins), 1092 cm.sup.−1 (DNA phosphate backbone), 1127 cm.sup.−1 (proteins), 1176 cm.sup.−1 (cytosine/guanine), 1210 cm.sup.−1 (tryptophan and phenylalanine), 1245 cm.sup.−1 (amide III), 1320 cm.sup.−1 (DNA/RNA, proteins, amide III), 1338 cm.sup.−1 (proteins and nucleic acids), 1422 cm.sup.−1 (related to the DNA/RNA content), 1450 cm.sup.−1 (proteins, lipids), 1578 cm.sup.−1 (nucleic acids), 1610 cm.sup.−1 (phenylalanine, tyrosine), 1619 cm.sup.−1 (heme), and 1669 cm.sup.−1 (amide I). The main Raman peaks and their tentative assignments are summarized in Table 1. Visually, the Raman spectra obtained from the negative cytology specimens are dominated by protein bands (e.g., 1669 cm.sup.−1). DNA and nucleic acids (1422 and 1578 cm.sup.−1), heme and other proteins (1610 and 1619 cm.sup.−1) are higher in abnormal samples. The main differences between the Raman spectra acquired from the intermediate (FIG. 4a) or superficial (FIG. 4b) cells from the negative and HSIL cytology specimens were observed at 482, 621, 728, 828, 855, 936, 957, 1092, 1176, 1422, 1450, 1578, 1610, 1619, and 1669 cm.sup.−1 (unpaired two-sided Student's t-test, p<0.001). These bands are mainly related to DNA, proteins and glycogen. This can be explained from the fact that the intermediate layer is the transitional layer between the immature cells from the basal/parabasal layer and the mature cells of the superficial layer. Hence, both of the cell layers (i.e., intermediate and superficial layers) may have similar molecular composition such as glycogen, DNA, proteins and lipids. The intermediate cells also showed significant changes at 1210 cm.sup.−1 corresponding to proteins (FIG. 4(a)). Similarly, the superficial cells also indicated prominent changes for proteins and nucleic acids (1338 cm.sup.−1, FIG. 4(b)). When the spectra of both cell types were mixed, the spectral changes were observed around the Raman peaks 482, 621, 828, 855, 936, 957, 1092, 1210, 1338, 1422, 1450, 1578, and 1669 cm.sup.−1 (FIG. 4). In short, both of the cell types yielded similar diagnostic information associated with HSIL.

(27) As the visual difference between the Raman spectra from the negative and HSIL cytology specimens are however very subtle, multivariate analysis, PLS-DA, was utilized to enhance the spectral differences. Leave-one-out, cross-validated PLS-DA models were developed from the dataset collected from the intermediate cells and the superficial cells from negative and HSIL cytology specimens. The PLS-DA model was also developed for the mixed intermediate/superficial dataset. The number of components (4 LVs, 4 LVs and 6 LVs, FIG. 5) corresponding to the minimum cross-validation error were utilized to develop the models for the intermediate, superficial and mixed intermediate/superficial datasets, respectively. The LVs for the superficial dataset accounted for 64.78% and 81.48% of the total Raman spectral variations in the X and Y directions and predominantly provided information about DNA and proteins (620, 640, 1004, 1338, 1450, and 1578 cm.sup.−1), glycogen and proteins (482, 855, 936, and 1655 cm.sup.−1). Similarly, the LVs (51.72% and 71.19% variations in X and Y direction) corresponding to the intermediate dataset mainly showed the changes in glycogen, DNA and protein features (482, 620, 640, 780, 855, 1004, 1450, 1578, and 1669 cm.sup.−1). Combining the intermediate and superficial dataset, the LVs (68.23% and 75.57% variations in X and Y direction) extracted the information around the Raman peaks such as 780, and 1578 cm.sup.−1 (DNA), and 482, 620, 640, 855, 936, 1004, 1338, 1450, and 1669 cm.sup.−1 (glycogen and proteins). From the above results, it is clear that both of the cells types either utilized separately or in combination provide almost the same diagnostic information for HSIL identification. The LV scores scatter plot (FIG. 6) obtained from the mixed intermediate/superficial dataset visually shows that the scores for the negative cytology specimens of intermediate and superficial cells are overlapped due to minimal variability. Similarly, the scores for the HSIL cytology specimens of intermediate and superficial cells are overlapped. However, the scores of both intermediate and superficial cells showed clear separation between the negative and HSIL cytology specimens. This reinforces the observation that the changes between the negative and HSIL cytology specimens is highly significant compared to the differences between the two cell types. However, some of the LV scores of the HSIL intermediate dataset are skewed towards the scores of the negative cytology specimens (FIG. 6). This can be attributed to the fact that some of the HSIL cytology specimens can regress back to normal or some of the cells in the HSIL specimen are still normal. The posterior probability plots from the developed PLS-DA models provided sensitivities of 91.5%, 94.9% and 93.0% and specificities of 95.5%, 96.6%, and 95.8% (Table 2) for identifying the HSIL cases based on the spectral dataset obtained from the intermediate (FIG. 7a), superficial (FIG. 7b), and mixed intermediate/superficial cells (FIG. 7c). The posterior probability plot of mixed intermediate/superficial cells (FIG. 7c) further shows that the diagnostic efficacy of identifying HSIL is not affected by mixing the intermediate and superficial cell types. This shows that either intermediate or superficial epithelial cells or both can be used to discriminate between negative and HSIL cytology cases. Superficial and intermediate cells are the most abundant cells on the Pap smear and they are very difficult to distinguish in the unstained slides so being able to use either cell type to classify normal vs. abnormal smears is of major benefit.

(28) TABLE-US-00001 TABLE 1 Main Raman peaks Wave- number (cm.sup.−1) Raman Peak Assignments 482 Glycogen 621 C—C twisting mode of Phenylalanine (proteins) 644 C—C twisting mode of Tyrosine and Phenylalanine 728 C—N stretching in Adenine and lipids 784 Uracil, Thymine, Cytosine (ring breathing modes in the DNA/RNA) 828 PO.sub.2 stretching in DNA, Tyrosine 855 Ring breathing in Tyrosine and Proline (proteins) 936 C—C stretching mode of Proline and Valine 957 C—C and C—N stretch PO.sub.3.sup.2− stretch (DNA) 1004 C—C aromatic ring stretching in Phenylalanine 1035 C—H bending mode in Phenylalanine, C—N stretching in proteins 1092 Symmetric PO.sub.2.sup.− stretching vibration of the DNA 1127 C—N stretching in proteins 1176 C—H in plane bending mode of Tryptophan & Phenylalanine; Cytosine, Guanine 1210 C—C.sub.6H.sub.5 stretching mode in Tryptophan & Phenylalanine 1245 Amide III (of collagen) 1320 Guanine (ring breathing modes of the DNA/RNA bases) - C—H deformation (protein); Amide III (α-helix) 1338 CH.sub.2/CH.sub.3 wagging & twisting mode in collagen, nucleic acid & tryptophan 1422 CH.sub.3 asymmetric stretch (lipids, aromatics) 1450 CH (CH.sub.2) bending mode in proteins and lipids 1578 Adenine, Guanine (DNA/RNA); C═C bending mode of Phenylalanine 1610 C═C Phenylalanine, Tyrosine and Tryptophan 1619 C═C Phenylalanine, Tyrosine and Tryptophan 1669 Amide I (C═O stretching, C—N stretching and N—H bending, proteins)

(29) TABLE-US-00002 TABLE 2 Calculated accuracy, sensitivity and specificity for differentiating negative and high-grade squamous intraepithelial lesion (HSIL) cytology using the Raman spectral dataset obtained from (i) intermediate cells, (ii) superficial cells and (iii) intermediate + superficial cells Type of cells Sensitivity (%) Specificity (%) Accuracy (%) Intermediate 91.5 (236/258) 95.5 (169/177) 93.1 (405/435) Superficial 94.9 (149/157) 96.6 (170/176) 95.8 (319/333) Intermediate + 93.0 (386/415) 95.8 (338/353) 94.3 (724/768) Superficial

(30) This method was also used to classify negative, LSIL and HSIL cases. Table 3 shows sensitivity of 89.2%, 63.2% and 81.4% and specificity of 85.3%, 89.0% and 91.4% for identifying negative, LSIL and HSIL (Table 3). Negative and HSIL samples could be classified very well, but LSIL samples were more difficult to classify correctly. Some LSIL cases classified as normal while some classified as HSIL.

(31) TABLE-US-00003 TABLE 3 Sensitivity and specificity for identifying negative, LSIL and HSIL samples TN (True Negative) LSIL HSIL TN 149 11 7 LSIL 50 129 25 HSIL 6 27 144 Sensitivity 89.2 63.2 81.4 Specificity 85.3 89.0 91.4
The following example explains this mis-classification by also considering HPV DNA and mRNA status.

Example 2

LSIL-HPV

(32) Materials and Methods

(33) Sample Collection

(34) This current study was approved by the Research Ethics Committee at the Coombe Women and Infants University Hospital (CWIUH), Dublin. A total of 39 cervical liquid based cytology samples (15 true negative (TN) specimens, 12 LSIL specimens and 12 HSIL specimens) were collected for this Raman study. True negative cytology samples were obtained from the cytology laboratory, CWIUH, Dublin, Ireland. The LSIL and HSIL cytology specimens were collected from the Colposcopy clinic, CWIUH, Dublin, Ireland. The collected smears were processed via ThinPrep® method. For ThinPrep®, an adequate sampling of cells was collected from the ectocervix of true negative, LSIL and HSIL patients using the cytobrush. The cytobrush was rinsed in the vial containing PreservCyt transport medium (ThinPrep® Pap Test; Cytyc Corporation, Boxborough, Mass.). The vial was named with the patient name and ID and then sent to the cytology laboratory equipped with a ThinPrep® 2000 processor (Hologic Inc., Marlborough, Mass. 01752). The ThinPrep® Pap test filter rotates within the sample vial and produces mild current that separates the debris and mucus without affecting the appearance of the cells. A gentle vacuum collects the cells on the exterior surface of the Pap test filter membrane. The filter is then inverted and gently pressed against the glass ThinPrep® slide. A gentle air pressure and surface tension helps the cells to adhere to the slide and creates an evenly distributed monolayer deposit of cells with a diameter of ˜20 mm. The slide was then transferred into a fixative bath of 95% ethanol automatically. The slide was then air dried and then the Raman spectral measurements were performed.

(35) All samples were tested for HPV DNA status using the Cobas HPV DNA test (Roche) and LSIL and HSIL samples were further tested for HPV mRNA status using the APTIMA HPV mRNA test (Hologic).

(36) Instrumentation

(37) Cell Raman spectra were acquired using a HORIBA Jobin Yvon XploRA® system (Villeneuve d'Ascq, France). The Raman microscopy system combines an Olympus microscope BX41 equipped with a ×100 objective (MPlanN, Olympus, NA=0.9). The spectroscopy system incorporates a 532-nm diode laser, 1200 lines/mm grating and an air-cooled CCD detector (Andor, 1024×256 pixels). Silicon (spectral peak at 520.7 cm.sup.−1) was used as the reference standard for the wavelength calibration. Intensity correction was performed using a relative intensity correction standard (NIST 2242). Each Raman spectrum from the cells was recorded with the laser power of ˜1 mW on the sample, with an integration time of 30 sec and 2 accumulations. Following the Raman measurement on each cell, the image of the cell along with its x- and y-coordinates was also obtained.

(38) A total of 548 Raman spectra (true negative=167, LSIL=204 and HSIL=177) were collected from the recruited 39 patients. Out of 204 spectra from LSIL cases, 66 spectra were HPV Cobas-negative (HPV DNA-negative), HPV Aptima-negative (HPV mRNA-negative) (CNAN); 69 were HPV Cobas-positive (HPV DNA-positive) and Aptima-negative (HPV mRNA-negative) (CPAN); 69 spectra were HPV Cobas-positive (HPV DNA-positive) and Aptima-positive (HPV mRNA-positive) (CPAP).

(39) Data Analysis

(40) All the recorded Raman spectra from true negative, LSIL and HSIL categories were subjected to data pretreatment including glass background correction, baseline correction and normalization. Following the data pretreatment, the Raman spectra were then mean centered to remove any magnitude dependency. The multi-class partial least squares discriminant analysis (PLS-DA) together with leave-one patient-out cross-validation model was developed for the mean-centered spectral dataset. The multivariate PLS-DA analysis was performed using the PLS toolbox (Eigenvector Research, Wenatchee, Wash.) in the Matlab® (Mathworks Inc., Natick, Mass.) scripting environment.

(41) Results

(42) A total number of 548 Raman spectra were acquired from the true negative specimens (n=167) and from the morphologically normal appearing cells from LSIL (n=204) and HSIL (n=177) specimens (FIG. 8a). The acquired Raman spectrum shows peaks at 482 cm.sup.−1 (glycogen), 621 and 644 cm.sup.−1 (proteins), 728 and 784 cm.sup.−1 (DNA), and 828 cm.sup.−1 (DNA/RNA), 855 and 936 cm.sup.−1 (glycogen and proteins), 957 cm.sup.−1 (DNA), 1004 cm.sup.−1 (phenylalanine), 1035 cm.sup.−1 (proteins), 1092 cm.sup.−1 (DNA phosphate backbone), 1127 cm.sup.−1 (proteins), 1176 cm.sup.−1 (cytosine/guanine), 1210 cm.sup.−1 (tryptophan and phenylalanine), 1245 cm.sup.−1 (amide III), 1320 cm.sup.−1 (DNA/RNA, proteins, amide III), 1338 cm.sup.−1 (proteins and nucleic acids), 1422 cm.sup.−1 (related to the DNA/RNA content), 1450 cm.sup.−1 (proteins, lipids), 1578 cm.sup.−1 (nucleic acids), 1610 cm.sup.−1 (phenylalanine, tyrosine), 1619 cm.sup.−1 (heme), and 1669 cm.sup.−1 (amide I). The difference spectra (FIG. 8b) show changes around 482, 786, 831, 1096, 1129, 1237, 1400, 1453 and 1656 cm.sup.−1. In general, the protein band is stronger (1669 cm.sup.−1) in negative samples and DNA and nucleic acids (1422 and 1578 cm.sup.−1) are stronger in the abnormal samples.

(43) The Cobas HPV DNA and Aptima HPV mRNA assays were used to categorize the LSIL samples based on HPV result. There are three categories according to HPV result (i.e., CNAN, CPAN, and CPAP) and the corresponding mean spectrum for each category is shown in FIG. 9a. The difference spectra among different categories of LSIL are also shown in FIG. 9b, depicting the important differences around the Raman peaks 785, 936, 1000, 1046, 1097, 1124, 1238, 1340, 1575 and 1652 cm.sup.−1. For instance, 1652, 1340, 1046 cm.sup.−1 are higher in LSIL-CNAN.

(44) As these differences among the spectra measured from true negative, LSIL and HSIL are very subtle, chemometrics including PLS-DA model was utilized to discern the minor differences among true negative, LSIL and HSIL. The multi-class PLS-DA model together with leave-one patient-out, cross-validation was developed using 5 PLS components corresponding to the minimum cross-validation error. The model was built and categorized into true negative, LSIL and HSIL based on the gold standard histology result, negative, CIN1 or CIN2/3. The total variance explained by the PLS components (latent variables (LVs)) are 59.76% in X-direction (LV1-23.31%, LV2-14.95%, LV3-6.45%, LV4-8.03%, LV5-7.01%) and 61.56% in Y-direction (LV1-21.02%, LV2-14.52%, LV3-15.43%, LV4-5.80%, LV5-4.80%). The PLS components extracted the information around the major Raman peaks (482, 625, 646, 728, 786, 831, 853, 939, 1005, 1097, 1171, 1440, 1610, 1620, 1655 and 1670 cm.sup.−1), related to the changes in cell biochemical constituents such as glycogen, nucleic acids, heme, proteins, and lipids associated with different grades of cervical precancer (FIG. 10). The posterior probability for each sample was calculated from the PLS scores. The posterior probability plot provided sensitivity of 89.2%, 63.2% and 81.4% and specificity of 85.3%, 89.0% and 91.4% for identifying true negative, LSIL and HSIL (Table 3). The posterior probability plot (FIG. 11) shows that the true negative and HSIL samples can be separated very well, but different categories in the LSIL samples are clustered differently. Out of 204 LSIL spectra, 66 spectra belong to LSIL-CNAN, 69 spectra to LSIL-CPAN and 69 spectra to LSIL-CPAP. Among 66 spectra from LSIL-CNAN, 43 are classified as true negative, 14 as LSIL and 9 as HSIL. Most of the LSIL-CNAN samples are misclassified as true negative because of their similarity to the true negative samples. This is most likely because there is no HPV presence/integration in the host cell. Out of 69 LSIL-CPAN samples, 68 of them classified as LSIL indicating that they are distinct from true negative or HSIL cases. Out of 69 LSIL-CPAP samples, 47 are classed as LSIL, 7 are classed as true negative and 15 are classified as HSIL. Many LSIL-CPAP samples are dispersed from CIN 1 towards CIN2/3, but not towards negative. This demonstrates that the patients with LSIL-CPAP may have a high risk of progression due to overexpression of the E6/E7 oncoproteins in the host cells i.e. the HPV genome is integrated with the host genome.

(45) Thus, the Applicant has surprisingly found that high quality Raman spectra can be successfully acquired from morphologically normal appearing cells from true negative, LSIL and HSIL Thinprep® specimens and different grades of cervical pre-cancer can be separated with good sensitivities and specificities. Raman spectroscopy can further identify different categories of the LSIL cases i.e., whether they are likely to regress to negative or progress to HSIL cytology. This is an important finding and explains the mixed discrimination of LSIL cases. The presence of HPV E6/E7 mRNA indicates an active transforming HPV infection suggesting that these LSIL cases are more likely to progress to HSIL. Thus, the present invention provides a system and method of using Raman spectroscopy for discriminating between LSIL cases likely to progress to HSIL or cancer from those likely to regress. This represents an important unmet clinical need as a high proportion of women with LSIL who are at a relatively low risk of developing cancer undergo unnecessary colposcopy follow up and in many instances also undergo unnecessary treatment. A reliable test to identify LSIL cases likely to progress or regress would greatly improve management of women presenting with low grade cytological abnormalities.

(46) Referring now to FIG. 12, the schematic flow chart shows the steps involved in carrying out the method of the present invention and the method will now be summarised with reference to FIG. 12.

(47) The method of the present invention comprises the following steps:

(48) 1. Carrying out Pap test—cervical cells collected from the cervix using a cyto brush and rinsed in the specimen vial containing liquid preservative (ThinPrep® Pap Test or SurePath® Pap Test);

(49) 2. Preparation of liquid based cytology slide—Samples are prepared using Thin Prep® or SurePath® liquid based cytology method—cells are transferred onto a glass slide to create a monolayer of cells;

(50) 3. Pre-treatment with hydrogen peroxide—before Raman measurement, each slide is pre-treated with hydrogen peroxidase (H.sub.2O.sub.2) to remove any contaminating blood and debris;

(51) 4. Selection of epithelial cells—slide placed on the Raman microscope, and using low power objective lens (eg. ×10 or ×20), unstained cervical epithelial cells (intermediate and superficial cells) visualized as polygonal cells with small nuclei and large cytoplasm;

(52) 5. Raman acquisition from cell nuclei—using high power objective lens (×100), laser (532 nm) directed at cell nuclei and Raman spectra acquired (eg. integration time of 30 sec and 2 accumulations); and

(53) 6. Carry out statistical learning algorithm and comparison of spectra to reference database/Classification of unknown spectrum—Raman spectra pre-processed by correcting for glass background using least-squares method with non-negative constraints (NNLS) method and for baseline. Vector normalisation and mean centring carried out. Unknown spectra tested against the classification model.

(54) The above referenced pre-treatment step (step 3 above) with hydrogen peroxide is described in detail in the Applicant's patent specification No. EP2984488.

REFERENCES

(55) 1. Kitchener H. C.; Blanks R.; Cubie H.; Desai M.; Dunn G.; Legood R.; Gray A.; Sadique Z.; Moss S. (2011) MAVARIC Trial Study Group. MAVARIC—A comparison of automation-assisted and manual cervical screening: A randomised controlled trial. Health Technol. Assess., 15:1-170. 2. Ellis, D. I.; Cowcher, D. P.; Ashton, L.; O'Hagan, S.; Goodacre, R. Illuminating disease and enlightening biomedicine: Raman spectroscopy as a diagnostic tool. Analyst 2013, 138, 3871-3884. 3. Kendall, C.; Isabelle, M.; Bazant-Hegemark, F.; Hutchings, J.; Orr, L.; Babrah, J.; Baker, R.; Stone, N. (2009) Vibrational spectroscopy: A clinical tool for cancer diagnostics. Analyst, 134:1029-1045. 4. Nijssen, A.; Koljenovic, S.; Bakker Schut, T. C.; Caspers, P. J.; Puppels, G. J. (2009) Towards oncological application of Raman spectroscopy. J. Biophotonics, 2:29-36. 5. Jess P. R. T., Simth D. D. W., Mazilu M., Dholakia K., Riches A. C., Herrington C. S. (2007) Early detection of cervical neoplasia by Raman spectroscopy. International Journal of Cancer 121:2723-2728. 6. Ostroswka K. M., Malkin A., Meade A., O'Leary J., Martin C., Spillane C., Byrne H. J., Lyng F. M. (2010) Investigation of the influence of high-risk human papillomavirus on the biochemical composition of cervical cancer cells using vibrational spectroscopy. Analyst 135:3087-3093. 7. Vargis, E.; Tang, Y.-W.; Khabele, D.; Mahadevan-Jansen, A. (2012) Near-infrared Raman Microspectroscopy Detects High-risk Human Papillomaviruses. Transl. Oncol. 5: 172-179. 8. Kelly J G, Cheung K T, Martin C, O'Leary J J, Prendiville W, Martin-Hirsch P L, Martin F L. A spectral phenotype of oncogenic human papillomavirus-infected exfoliative cervical cytology distinguishes women based on age. Clin Chim Acta. 2010 Aug. 5; 411(15-16):1027-33. 9. Schubert J M, Bird B, Papamarkakis K, Miljković M, Bedrossian K, Laver N, Diem M. Spectral cytopathology of cervical samples: detecting cellular abnormalities in cytologically normal cells. Lab Invest. 2010 July; 90(7):1068-77.

REFERENCES I-V

(56) .sup.i Rana D N, Marshall J, Desai M, Kitchener H C, Perera D M, El Teraifi H, Persad R V. Five-year follow-up of women with borderline and mildly dyskaryotic cervical smears. Cytopathology. 2004 15(5):263-70. .sup.ii Melnikow J, Nuovo J, Willan A R, Chan B K, Howell L P. Natural history of cervical squamous intraepithelial lesions: a meta-analysis. Obstet Gynecol. 1998 92(4 Pt 2):727-35 .sup.iii Bentley E, Cotton S C, Cruickshank M E, Duncan I, Gray N M, Jenkins D, Little J, Neal K, Philips Z, Russell I, Seth R, Sharp L, Waugh N; Trial of Management of Borderline and Other Low-Grade Abnormal Smears (TOMBOLA) Group. Refining the management of low-grade cervical abnormalities in the UK National Health Service and defining the potential for human papillomavirus testing: a commentary on emerging evidence. J Low Genit Tract Dis. 2006 10(1):26-38. .sup.iv Cox J T; American Society for Colposcopy and Cervical Pathology. The clinician's view: role of human papillomavirus testing in the American Society for Colposcopy and Cervical Pathology Guidelines for the management of abnormal cervical cytology and cervical cancer precursors. Arch Pathol Lab Med. 2003 127(8):950-8. .sup.1v Bosch F X, Manos M M, Muñoz N, et al. Prevalence of human papillomavirus in cervical cancer: a worldwide perspective. International biological study on cervical cancer (IBSCC) study group. Journal of the National Cancer Institute 87(11):796-802 (1995)

(57) Clauses

(58) 1. A cytology system for analyzing a biological sample on a glass slide, the system comprising: a stage for receiving the sample holder, a low resolution Raman spectroscopy device having a spectral resolution worse than 3 wavenumbers, the Raman spectroscopy device having an analysis module for determining whether the spectrum falls within one or more predefined classes of cell.

(59) 2. A system according to clause 1, wherein the biological sample is a Pap smear on a glass slide.

(60) 3. A system according to clause 1 or clause 2, wherein the one or more predefined classes comprise the following: a) normal b) invasive carcinoma and c) cervical intraepithelial neoplasia (CIN).

(61) 4. A system according to clause 3, wherein the one or more predefined classes are further delineated into one of the classifications of low-grade squamous intraepithelial lesions (LSIL) and high-grade squamous intraepithelial lesions (HSIL) and optionally, may be delineated into the classes of: a) LSIL (CIN I); and HSIL which comprises b) CIN II and C) CIN III.

(62) 5. A system according to any preceding clause, wherein the image analysis identifies cells as areas of interest.

(63) 6. A system according to any preceding clause, further comprising: a controller, and a microscope for viewing the sample holder, the microscope having a central optical axis, wherein the Raman spectroscopy device shares the central optical axis of the microscope, and wherein the controller is adapted to cause the stage to move an identified area of interest on the slide to be in-line with the central optical axis and to cause a spectrum to be obtained by the Raman spectroscopy device for the area of interest.

(64) 7. A system according to clause 6, further comprising a graphical user interface comprising a window display the view from the microscope, wherein the interface is configured to allow a user to use a pointer to identify the area of interest.

(65) 8. A system according to clause 7 wherein the result of the determination of whether the spectrum falls within one or more predefined classes of cell is displayed within the graphical user interface.

(66) 9. A system according to any one of clauses 6 to 8, wherein the analysis module is configured to perform image analysis on an image acquired by the microscope to identify areas of interest.

(67) 10. A system according to any one of clauses 6 to 9, wherein the system further comprises a light source for illuminating the slide, wherein the controller is adapted to switch off the light source when operating the Raman spectroscopy device.

(68) 11. A system according to any one of clauses 6 to 10 further comprising a moveable mirror for switching the optical path between the microscope and the Raman spectroscopy device, wherein the moveable mirror is responsive to the controller.

(69) 12. A method of analyzing biological samples where the biological sample is a Pap smear, the method comprising the steps of:

(70) performing low resolution Raman sprectroscopy with a spectral resolution worse than 3 wavenumbers to obtain a spectrum for the biological sample, and analysing the spectrum to determine whether the spectrum falls within one or more predefined classes of cells.

(71) 13. A method according to clause 12, wherein the one or more predefined classes comprise the following: low-grade squamous intraepithelial lesions (LSIL) and high-grade squamous intraepithelial lesions (HSIL) and optionally, may be delineated into the classes of: a) LSIL (CIN I); and HSIL which comprises b) CIN II and C) CIN III.

(72) 14. A method according to clause 13, wherein the CIN class is further delineated into the classes of: a) CIN I b) CIN II and C) CIN III.

(73) 15. A method according to any one of clauses 12 to 14 wherein the method comprises using Raman Spectroscopy for identification of low-grade squamous intraepithelial lesions (LSIL) that are likely to progress to high-grade squamous intraepithelial lesions (HSIL).

(74) 16. A method according to any one of clauses 12 to 15 wherein superficial or classification method multivariate analysisintermediate epithelial cells are used in the method for discrimination of negative and low-grade squamous intraepithelial lesions (LSIL) and high-grade squamous intraepithelial lesions (HSIL) cytology cases.

(75) 17. A method according to any one of clauses 12 to 16 wherein the method comprises the step of sub classification of low-grade squamous intraepithelial lesions (LSIL) based on HPV status.

(76) 18. The method of the present invention comprises the following steps: Carrying out a Pap smear;

(77) Preparation of liquid based cytology slide whereby cells are transferred onto a glass slide to create a monolayer of cells;

(78) Pre-treatment using hydrogen peroxide to remove blood contamination; Selection of epithelial cells;

(79) Acquiring Raman spectra from cell nuclei; and

(80) Carrying out Statistical learning algorithm and comparison of spectra to reference database for Classification of unknown spectrum.

(81) 19. A method according to clause 18 wherein the step of selection of epithelial cells comprises placing the glass slide on the Raman microscope, and using low power objective lens unstained cervical epithelial cells (intermediate and superficial cells) visualized as cells with small nuclei and large cytoplasm.

(82) 20. A method according to clause 19 wherein the low power objective lens is of the order of ×10 or ×20 magnification.

(83) 21. A method according to clause 18 wherein the step of acquiring Raman spectra from cell nuclei comprises using a high power objective lens and laser directed at cell nuclei for acquiring the Raman spectra.

(84) 22. A method according to clause 21 wherein the laser is operated at approx. 532 nm and optionally, wherein the high power objective lens is of the order of ×100 magnification.

(85) 23. A method as per clause 18 wherein the step of carrying out statistical learning algorithm and comparison of spectra to reference database comprises Raman spectra pre-processing by correcting for glass background and for baseline and vector normalisation.

(86) 24. A method as per clause 23 wherein the step of correcting for glass background comprises using least-squares method with non-negative constraints (NNLS) method.