Concentration of FAD as a marker for cervical precancer detection

Abstract.

We report the ex vivo results of an in-house fabricated portable device based on polarized fluorescence measurements in the clinical environment. This device measures the polarized fluorescence and elastic scattering spectra with 405-nm laser and white light sources, respectively. The dominating fluorophore with 405-nm excitation is flavin adenine dinucleotide (FAD) with a fluorescence peak around 510 nm. The measured spectra are highly modulated by the interplay of scattering and absorption effects. Due to this, valuable information gets masked. To reduce these effects, intrinsic fluorescence was extracted by normalizing polarized fluorescence spectra with polarized elastic scattering spectra obtained. A number of fluorophores contribute to the fluorescence spectra and need to be decoupled to understand their roles in the progression of cancer. Nelder–Mead method has been utilized to fit the spectral profile with Gaussian to decouple the different bands of contributing fluorophores (FAD and porphyrin). The change in concentration of FAD during disease progression manifests in the change in ratio of total area to FWHM of its Gaussian profile. Receiver operating characteristic (ROC) curve analysis has been used to discriminate different grades of cervical precancer by using the ratio as input parameter. The sensitivity and specificity for discrimination of normal samples from CIN I (cervical intraepithelial neoplasia) are 75% and 54%, respectively. Further, the normal samples can be discriminated from CIN II samples with 100% and 82% sensitivity and specificity, respectively, and the CIN I from CIN II samples can also be discriminated with 100% sensitivity and 90% specificity, respectively. The results show that the change in the concentration of (FAD) can be used as a marker to discriminate the different grades of the cancer and biochemical changes at an early stage of the cancer can also be monitored with this technique.

1. Introduction

Fluorescence spectroscopy is a sensitive technique, applied extensively for detection of different types of cancers. It has the potential to monitor the biochemical changes that take place during the disease progression.^1–4 The spectroscopic study of breast tissue was pioneered by Alfano et al.,⁵ and Feld et al.⁶ applied it to human aorta tissue. Subsequently different research groups worked on its application for detection of breast cancer.^7–10 Autofluorescence from buccal mucosa has been used by Schwarz et al.,¹¹ Madhuri et al.,¹² and Kumar et al.¹³ in oral cancer detection. Several groups have worked on diagnosis of cervical precancer using laser-induced fluorescence spectroscopy^14,15 and in combination with reflectance spectroscopy^16–18 for comparison and to improve the detection efficacy. Intrinsic fluorescence^19–21 has also been extracted to understand the biochemical changes with disease progression through different experimental²² and simulation-based techniques.^23–26 The use of different algorithms, such as machine learning,²⁷ wavelet analysis,⁹ and other multivariate algorithms^10,20,28,29 have made detection more effective and robust.

As mentioned above, the natural fluorescences are taken as markers for detection of disease. It is well known that biological tissue has several natural fluorophores with absorption and fluorescence spectra in UV–VIS range.^12,30–33 The development of the cancer is associated with the changes taking place in tissue optical properties, which includes changes in the concentration of contributing fluorophores, such as flavin adenine dinucleotide (FAD), reduced form of nicotinamide adenine dinucleotide (NADH), and change in collagen cross-links in stroma.^31,32 Further, the morphological and microvascularization changes results in variation in scattering and absorption properties of the biological tissue sample.^8,24,29 The fluorescence features in biological tissue can be distorted significantly due to scattering and absorption effects. Hence extraction of intrinsic fluorescence is needed to monitor the relative changes in the concentration of contributing fluorophores.^{20,22–26,34} Earlier, Biswal et al.²² developed an experimental technique to extract the intrinsic fluorescence based on polarized fluorescence and polarized scattering measurements, which is free from absorption and scattering effects. Further ex vivo studies on biopsy samples of the cervix have illustrated the efficacy of this technique.^19,20 This has led to the fabrication of a portable device for in vivo detection of cervical precancer.²⁰

In this study, we report the ex vivo results from the portable device, by using the change in the concentration of FAD as a marker for early cervical cancer detection. A number of fluorophores contribute to the broad fluorescence spectrum when tissue is excited at 405 nm. These fluorophores need to be decoupled to understand their roles in the progression of cancer.³⁵ Nelder–Mead method has been utilized to fit the spectral profile with Gaussian to decouple the different bands of contributing fluorophores (FAD and porphyrin).^20,31 The change in concentration of FAD during disease progression manifests in the change in ratio of total area and FWHM of its Gaussian profile. Receiver operating characteristic (ROC) curve analysis has been used to discriminate different grades of cervical precancer by using the ratio as input parameter.

2. Materials and Methods

2.1. Instrumentation

Figure 1 shows the schematic of the experimental setup used for the measurements of polarized fluorescence and polarized elastic scattering spectra from total hysterectomy samples. The details of the system have been discussed in other references.^20,21 In brief, two light sources, a 405-nm diode laser (Pegasus, Shanghai, Optical System Co. Ltd.) and a white light source (Xe-lamp, Newport Oriel Instruments) were used for polarized fluorescence and polarized elastic scattering measurements, respectively. A miniature spectrometer (HR2000+, Ocean Optics, Inc., Dunedin, Florida) was used for collection of spectra. The sample was illuminated with focused vertically polarized light through different optical components. Fluorescence and scattering signals were collected sequentially from the same site of the sample with different integration times in co- and crosspolarization states in backscattering mode and background corrections were done separately for each.

Fig. 1.

(a) Schematic block diagram and (b) cartoon representation of the experimental setup for polarized fluorescence measurements from cervical hysterectomy tissue samples.

2.2. Sample Handling and Analysis Method

The study protocol was approved by institutional ethics committee at IIT Kanpur, India, and GSVM Medical College Kanpur, India, under the protocol number IITK/IEC/2012-13/1/3. The study was performed in the hospital (at GSVM Medical College) on the total hysterectomy samples. Figure 2 shows the schematic representation of the study protocol followed during the study on the hysterectomy cervical samples in the clinical environment. The details of sample handling and data acquisition have been discussed elsewhere.²⁰ The doctors involved in this study ensured that a written consent was obtained from each patient before the experiment was performed. A total of 156 sites of 28 patients include, 68 normal sites of 13 patients, 72 CIN I sites of 13 patients, and 16 sites of 2 patients from different social backgrounds and age groups were tested, and they were divided into different groups according to postbiopsy findings. A long time delay after the operation degrades the contribution from different fluorophores, mainly the contribution of the porphyrin.²¹ Therefore, the measurements on the total hysterectomy samples has been perform immediately after the operation in the hospital within time delay of 5 to 10 min, which minimized the loss of the signal. Intrinsic fluorescence was extracted from the measured data using^19–22

$$\mathrm{IF}=\frac{{[I_{\mathrm{vv}}(\lambda )-G(\lambda )*I_{\mathrm{vh}}(\lambda )]}_{\mathrm{fl}}}{{[I_{\mathrm{vv}}(\lambda )-G(\lambda )*I_{\mathrm{vh}}(\lambda )]}_{\mathrm{scat}}},$$ (1)

where $I_{vv}(\lambda )$ and $I_{\mathrm{vh}}(\lambda )$ are the co- and crosspolarized signals, respectively. Subscript “fl” and “scat” represent the fluorescence and scattering, respectively. $G(\lambda )=I_{\mathrm{hh}}(\lambda )/I_{\mathrm{hv}}(\lambda )$ is the ratio of the sensitivity of the instrument to the vertically and horizontally polarized light, keeping the source light in horizontal polarized state.

Fig. 2.

Schematic representation of the ex vivo study protocol in the clinical environment with the cervical hysterectomy tissue samples.

After extraction of intrinsic fluorescence, Nelder–Mead method has been utilized to fit the spectral profile with Gaussian to decouple the different bands of contributing fluorophores (FAD and porphyrin). The change in concentration of FAD during disease progression manifests in the change in ratio of total area and FWHM of its Gaussian profile. ROC curve analysis has been used to discriminate different grades of cervical precancer samples from normal cervical samples by using the ratio as input parameter.

3. Results and Discussion

Figures 3(a) and 3(b) show the co- and crosspolarized fluorescence spectra, respectively, from normal and CIN I counterpart of the same sample. The major fluorescence bands of FAD (peak at 510 nm) and porphyrin (peaks at 590 and 650 nm) have been observed.^20,31 The elimination of absorption and scattering effects in extracted intrinsic fluorescence spectra of normal and CIN I can be seen in Fig. 3(c). The fluorescence intensity from the normal site of the sample is found to be higher than that with CIN I site of the same sample, as expected, due to the conversion of FAD to its reduced form with disease progression.^12,33 Figure 3(d) shows the area normalized polarized fluorescence, polarized elastic scattering, and extracted intrinsic fluorescence spectra of CIN I sample. The absorption dips of hemoglobin/oxyhemoglobin are evident in polarized fluorescence and polarized elastic scattering spectra at 476, 527, and 556 nm.²⁰

Fig. 3.

Typical fluorescence spectra of normal and CIN I counterparts of the same sample: (a) copolarized fluorescence, (b) crosspolarized fluorescence, (c) intrinsic fluorescence, and (d) area normalized polarized fluorescence, polarized elastic scattering, and intrinsic fluorescence spectra of CIN I counter part of the sample.

Figures 4(a) and 4(b) show the typical fitted curves with Nelder–Mead method and the bottom plots show the residual spectra obtained from the corresponding Gaussian fit of the area normalized signals of normal, and CIN I counterparts of the same sample, respectively. The fitting was done by fixing the position of the different peaks and the total area under the different contributing fluorescence peaks and FWHM were calculated. A similar process was followed for co-polarized and crosspolarized spectra (not shown here). The ratio of the area under the FAD peak to the FWHM was used as input in ROC analysis. Figures 5(a) and 5(b) show the scatter plots of the ratio, using intrinsic fluorescence for the discrimination of CIN I and CIN II samples from normal, respectively, and the discrimination of CIN I from CIN II samples is shown in Fig. 5(c). Figure 5(d) shows the ROC analysis with the ratio from extracted intrinsic fluorescence signal of different grades of the cervical samples. The sensitivity and specificity for the discrimination of normal from CIN I samples is low but the discrimination of higher grade from normal and from CIN I is high (shown in 1). Figures 6(a)–6(c) show the scatter plot of the ratio for copolarized fluorescence spectra and the ROC analysis for normal versus CIN I, normal versus CIN II, and CIN I versus CIN II are shown in Fig. 6(d). Similarly, the analysis was done with cross-polarized fluorescence signal also and shown in Figs. 7(a)–7(d).

Fig. 4.

Typical fitted intrinsic fluorescence spectra of (a) normal and (b) CIN I counterparts of the same sample (bottom plot shows the residual of the corresponding plot).

Fig. 5.

Scatter plot of the ratio of the area under the curve and the FWHM of the fitted curve for FAD peak to discriminate different class pairs using intrinsic fluorescence: (a) normal versus CIN I, (b) normal versus CIN II, (c) CIN I versus CIN II, and (d) ROC curves sowing the discrimination efficiency of the ratio for different class pairs.

Fig. 6.

Scatter plot of the ratio of the area under the curve and the FWHM of the fitted curve for FAD peak to discriminate different class pairs using copolarized fluorescence: (a) normal versus CIN I, (b) normal versus CIN II, (c) CIN I versus CIN II, and (d) ROC curves sowing the discrimination efficiency of the ratio for different class pairs.

Fig. 7.

Scatter plot of the ratio of the area under the curve and the FWHM of the fitted curve for FAD peak to discriminate different class pairs using cross-polarized fluorescence: (a) normal versus CIN I, (b) normal versus CIN II, (c) CIN I versus CIN II, and (d) ROC curves sowing the discrimination efficiency of the ratio for different class pairs.

The statistical findings of the analysis are shown in Table 1. As is evident from the table, normal and CIN I are well discriminated from CIN II with high sensitivity and specificity. The results show that although the sensitivity of discrimination of normal and CIN I from CIN II with co- and crosspolarized spectra is similar to intrinsic fluorescence, the specificity in both cases is low as compare to intrinsic fluorescence. The advantage of reducing absorption and scattering effects in the extracted intrinsic fluorescence is thus evident. The results show that the change in the concentration of FAD can be used as a marker to discriminate the different grades of the cancer, and biochemical changes at early stage of the cancer can also be monitored using intrinsic fluorescence.

Table 1.

Statistical findings of the classification of different tissue classes of the cervical hysterectomy samples with intrinsic fluorescence, copolarized fluorescence, and crosspolarized fluorescence spectra.

Tissue classes	Sensitivity (%)	Specificity (%)
Intrinsic fluorescence
Normal versus CIN I	75	54
Normal versus CIN II	100	82
CIN I versus CIN II	100	90
Copolarized fluorescence
Normal versus CIN I	55	64
Normal versus CIN II	100	64
CIN I versus CIN II	100	65
Crosspolarized fluorescence
Normal versus CIN I	65	52
Normal versus CIN II	100	55
CIN I versus CIN II	100	69

Biographies

Bharat L. Meena is a research scholar at Indian Institute of Technology Kanpur and an assistant professor at the University of Rajasthan Jaipur. He received his MS degree in physics from the University of Delhi in 2010. His research interests have been in the area of biophotonics. The recent focus lies in the development of a portable device for detection of cervical cancer at an early stage, using polarized light.

Kiran Pandey is one of the senior most gynecologists of India, who heads the Department of OBG at GSVM Medical College, Kanpur. Her areas of expertise include gyne oncology, infertility, urogynecology, and high risk pregnancy. She is the secretary of the U.P chapter of Association of Gyne Oncologists of India. She has published more than 100 research papers in national and international journals. She has been bestowed with 11 national level, 8 state level, and numerous district level awards.

Asima Pradhan is a professor in the Department of Physics at the Indian Institute of Technology (IIT), Kanpur, India, and is also associated with the Center for Lasers and Photonics. She received her PhD from City University of New York in 1991. Her areas of interest include laser spectroscopy and biophotonics. Her expertise is in fluorescence spectroscopy, light scattering spectroscopy, time-resolved fluorescence, polarization-based fluorescence, and Mueller matrix imaging of biosamples.

Biographies of the other authors are not available.

Disclosures

The authors have no relevant financial interests in this article and no potential conflicts of interest to disclose.