Detection of prostate cancer cells with FTIR and Raman micro-spectroscopy
10.3. RESULTS 132 statistical classifier
The first two PC coefficients from each analysis can be found in Figure 10.5(a) and (b) for CaF2 slides, (c) and (d) for glass slides, and a combination of CaF2 and glass are shown in (e) and (f). Similar features appear across the first two PC coefficients for all three cases, with common peaks observed at 782 (DNA), 1243 (amide III), 1304 (nucleic acids), 1340 (colla-gen), 1437 (lipids), 1485 (nucleic acids), 1573 (nucleic acids), and 1681 cm−1(amide I) for the first PC, and at 915 (RNA), 1336 (DNA), 1401 (collagen), and 1663 cm−1 (DNA) for the sec-ond PC. (163; 208) Importantly, there are no prominent features associated with the glass signal appearing in any of these coefficients. This result indicates that spectra recorded on glass slides could be directly compared to Raman spectra recorded on CaF2slides (assuming adequate sys-tem calibration has been achieved, see Section 4.2 for further information). This is an important result for research groups, or clinics, that would like to alter their current protocols to include glass slides, but would like to retain retrospective Raman datasets that have been recorded on CaF2 substrates in their lab for inclusion in ongoing studies.
10.3.2 FTIR results
Figure 10.6 shows the fingerprint region (1000–2000 cm−1) of the FTIR spectra recorded from PNT1A, PC3, and T24 cells on CaF2, with the raw spectra shown in Figure 10.6(a), and spectra following EMSC resonant Mie scatter correction shown in Figure 10.6(b). Similar to Raman data (as shown in Figures 10.2 and 10.3), the variance has been reduced across each FTIR dataset following EMSC, along with a reduction of any spectral shifting that may have been present in the raw spectra due to Mie scattering artifacts. (81)
Figure 10.6: FTIR spectra of PNT1A, PC3, and T24 cells, recorded on CaF2slides, with (a) rep-resenting the raw spectra, and (b) shows the corresponding spectra following EMSC resonant Mie scattering correction.
(a)
(b)
Figure 10.7: (a) PCA scores for FTIR spectra of PNT1A, PC3, and T24 cells, and (b) shows the first three corresponding PC coefficients.
10.4. SUMMARY 134 Table 10.1 shows the PC-LDA classification results obtained for the data shown in Fig-ure 10.6(b), based on a k−fold cross validation (k = 10). These results are similar to those obtained with Raman micro-spectroscopy, in terms of the ability to classify across all three cell lines, with sensitivities and specificities obtained of greater than 93%.
Based on the PCA scores shown in Figure 10.7(a), it can be seen that the FTIR spectra are separable based on PCA alone, whereas all three Raman spectral datasets are only separable following PC-LDA. Interestingly, PC3 and T24 cells are grouped closer together relative to PNT1A cells, allowing for complete separation between cancerous and healthy cells, similar to the separation observed between cancerous and healthy cells across the first LD for Raman spectra recorded on CaF2, as shown in Figure 10.4(b).
As shown in Figure 10.7(a), most of the separation across the FTIR spectral datasets oc-curs over the second PC. Figure 10.7(b) shows the first three PCs, with the main peak differ-ences seen in the second PC at 1060 (glycogen), 1220 (PO2asymmetric stretching, Amide III), 1400 (Amide III), 1470 (CH3 asymmetric stretching, proteins, lipids), and 1580 cm−1 (Amide III). (30; 191; 202)
10.4 Summary
The results presented in this chapter suggest that Raman and FTIR spectroscopy have the poten-tial to identify prostate cancer cells that are present in a urine cytology sample prepared using the ThinPrep UroCyte method, which is commonly used in clinical cytopathology laboratories.
The results presented here show that Raman micro-spectroscopy can be applied to both CaF2 and glass slides, and that these recordings can be carried out somewhat interchangeably, as well as showing that FTIR spectroscopy can be applied to CaF2slides that have been processed using the ThinPrep method. These results reinforce previous studies that have suggested that vibra-tional spectroscopy could be applied to slides obtained directly from the clinic. (78) However, it is important to highlight some of the incompatibility issues that have been observed with CaF2 and the T2 machine, as discussed in Section 10.2.1, as well as acknowledging the financial burden associated with operating a diagnostic technique using expensive CaF2substrates.
In order to overcome any issues with CaF2 and the T2 machine, it may be possible to in-corporate glass slides into FTIR based cytology studies in a similar manner as Raman micro-spectroscopy. Bassan et al (37) recently showed the ability to classify between paraffin embed-ded breast cancer tissue samples in the higher wavenumber region. However, the combination of glass slides and FTIR have not been considered here as this study focused only on the finger-print region of the FTIR spectrum. Furthermore, the complex resonant Mie scattering signals that are inherent in FTIR cytology spectra may prove problematic to remove using EMSC due to the varying levels of the glass signal present across each spectrum. Nonetheless, this may prove interesting for future studies, and it may be possible to expand the resonant Mie scattering EMSC algorithm to incorporate the glass signal, similar to that presented in Chapter 8.
Similar classification results have been achieved within this study for both vibrational spec-troscopy methods, as shown in Table 10.1. However, it is also important to consider other aspects of both these techniques in terms of the practical implementation in the clinic. For this study, Raman spectra are recorded from 50 cells on each slide, with a total acquisition time of 10 s per cell. With the inclusion of additional time required to locate cellular nuclei and the correct z-focal depth position, the overall time required to analyse one Raman cytology slide is on the order of 1-2 h, with longer times required for inexperienced users. FTIR on the other hand allows for the analysis of large regions of a slide, and therefore larger numbers of cells, within the order of 10-15 min, even when operated by an inexperienced user. Whilst the times required for both techniques could be improved by automating the systems involved, Raman micro-spectroscopy remains a much slower technique than FTIR. However, pre-processing of both groups of spectral datasets shows that Raman algorithms are significantly faster to run than the complex algorithms required to remove Mie scattering artifacts from FTIR spectra, resulting in similar time frames overall for the application of either technique.
Given the degree of accuracy obtained here, which is significantly higher than that achieved with PSA or DRE, there is a good argument for the possible introduction of vibrational spec-troscopic techniques into the clinic to assist cytopathologists in accurately identifying the pres-ence of prostate cancer cells. Prior to the widespread application of core biopsy, many prostate cancers were diagnosed with fine needle aspiration (FNA); however, core biopsy samples pro-duced higher accuracies in determining the presence of prostate carcinoma. With the ThinPrep method, it is also possible to prepare FNA samples to produce similar cytology slides as that achieved with urine cytology, with the added benefit of larger numbers of prostate cells. This study has shown that Raman and FTIR micro-spectroscopy may have the potential to improve upon the diagnostic capabilities of FNA samples. FNA carries the additional benefits of being quicker, less expensive to perform, and safer for the patient. (209) FNA samples are routinely obtained from patients suspected of harbouring breast cancer; Chapter 11 investigates the appli-cation of Raman micro-spectroscopy for the analysis of such simulated breast cytology samples.