Bladder cancer diagnosis during cystoscopy using Raman spectroscopy

(1)

PROCEEDINGS OF SPIE

SPIEDigitalLibrary.org/conference-proceedings-of-spie

Bladder cancer diagnosis during

cystoscopy using Raman

spectroscopy

Grimbergen, M. C. M., van Swol, C. F. P., Draga, R. O. P.,

van Diest, P., Verdaasdonk, R. M., et al.

M. C. M. Grimbergen, C. F. P. van Swol, R. O. P. Draga, P. van Diest, R. M.

Verdaasdonk, N. Stone, J. H. L. R. Bosch, "Bladder cancer diagnosis during

cystoscopy using Raman spectroscopy," Proc. SPIE 7161, Photonic

Therapeutics and Diagnostics V, 716114 (23 February 2009); doi:

10.1117/12.807811

Event: SPIE BiOS, 2009, San Jose, California, United States

(2)

Bladder cancer diagnosis during cystoscopy using Raman spectroscopy

MCM Grimbergen

a,d

, CFP van Swol

b

, ROP Draga

d

, P van Diest

c

, RM Verdaasdonk

a

,

N Stone

e

, JHLR Bosch

d

a

_{Department of Biomedical Engineering and Clinical Physics, UMC Utrecht, The Netherlands,}

b

_{Dept. Clinical Physics & Instrumentation, St Antonius Hospital Nieuwegein, The Netherlands,}

c

_{Dept. of Pathology, University Medical Centre Utrecht, The Netherlands,}

d

_{Dept of Urology, University Medical Centre Utrecht The Netherlands,}

e

_{Biophotonics Research Group, Gloucestershire Hospitals, Gloucester, The United Kingdom}

ABSTRACT

Raman spectroscopy is an optical technique that can be used to obtain specific molecular information of biological tissues. It has been used successfully to differentiate normal and pre-malignant tissue in many organs. The goal of this study is to determine the possibility to distinguish normal tissue from bladder cancer using this system.

The endoscopic Raman system consists of a 6 Fr endoscopic probe connected to a 785nm diode laser and a spectral recording system. A total of 107 tissue samples were obtained from 54 patients with known bladder cancer during transurethral tumor resection. Immediately after surgical removal the samples were placed under the Raman probe and spectra were collected and stored for further analysis. The collected spectra were analyzed using multivariate statistical methods.

In total 2949 Raman spectra were recorded ex vivo from cold cup biopsy samples with 2 seconds integration time. A multivariate algorithm allowed differentiation of normal and malignant tissue with a sensitivity and specificity of 78,5% and 78,9% respectively.

The results show the possibility of discerning normal from malignant bladder tissue by means of Raman spectroscopy using a small fiber based system. Despite the low number of samples the results indicate that it might be possible to use this technique to grade identified bladder wall lesions during endoscopy.

Keywords: Bladder cancer, Raman spectroscopy, biopsies

1. INTRODUCTION

Raman spectroscopy is an optical technique that can be used to obtain specific molecular information of biological tissues. As such it has been used successfully to differentiate epithelial neoplasma from normal tissue and inflammation in various organs[1-3]. Most of these studies were on ex-vivo material, but a small number of studies have reported on the possibility of using Raman spectroscopy in-vivo[1,3,4]; the reason for this being the lack of fiber optical Raman probes suited for use in-vivo. The standard fiber optical Raman probes have been developed for chemical sampling / monitoring applications and are commonly too bulky for a medical application, particularly when compared to the average biopsy volume for cancer diagnosis in minimal invasive procedures. Recently a new fiber optic Raman probe was developed in cooperation with EMVision LLC, that allows in vivo medical application. The bare fiber end of the distal tip of this probe enables collection of Raman signals from a relatively large (≈1cm3_{) volume.}

The mucosal lining of the normal bladder wall is the origin of bladder cancer. Known as the urothelium, it is a 7-cell layer think membrane covering the underlying basal membrane, muscularis and adipose tissue. Currently, two separate biological pathways are recognized in bladder cancer development, differing in tumor development as an exophytic papillary tumor that can be relatively easily discovered by endoscopic evaluation of the bladder wall, and intrinsic concealed dysplastic cell of the urothelium that consecutively develop from mild to severe dysplasia and carcinoma in situ and invasive tumors. In both situations the initial changes take place in the thin urothelial layer of the bladder wall. Previous Raman spectroscopic studies on bladder tissue were focused on determining the Raman characteristics of bladder pathologies of small tissue samples. The spectral data acquired by these Raman micro spectrometers allowed for accurate correlation with tissue pathology by means of multivariate statistical techniques[5]. These techniques allow

Photonic Therapeutics and Diagnostics V, edited by Nikiforos Kollias, Bernard Choi, Haishan Zeng, Reza S. Malek, Brian Jet-Fei Wong, Justus F. R. Ilgner, Kenton W. Gregory, Guillermo J. Tearney, Laura Marcu, Henry Hirschberg, Steen J.

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 05 Feb 2020 Terms of Use: https://www.spiedigitallibrary.org/terms-of-use

(3)

prediction of pathology groups as used in urological practice to assess the patients risk and subsequently the treatment options. The present study aims to provide the pathology prediction by means of Raman spectroscopy during the cystoscopy procedure.

1.1 Aim

The aim of this study is to determine whether the newly available fiber optic Raman probes, sampling large tissue volumes, provide ‘good quality’ Raman spectra of the bladder wall allowing differentiation of pathology.

2. METHODOLOGY

2.1 Raman system

The endoscopic Raman system consisted of a custom build 6 Fr endoscopic probe holding 1 illumination fiber and 7 collection fibers (Emvision LLC, Loxahatchee, FL). The bare fiber ends at the distal tip of the probe allow for collection of Raman signals from a large tissue volume.

Fig. 1.a) Schematic of the Raman Setup, b) Tip of the EMVision Raman probe, with central excitation fiber and 7 surrounding collection fibers

2.2 Tissue samples

A total of 107 tissue samples were obtained from 54 patients with known bladder cancer during transurethral tumor resection. Photodynamic diagnosis (5-ALA) was used to enhance the contrast during the bladder wall evaluation. The samples were obtained from clearly suspicious red fluorescing areas as well as normal non fluorescent looking bladder wall. Immediately after surgical removal the samples were placed, orientated with the urothelium facing up, on aluminum foil to minimize background interference during the Raman measurement. The Raman probe was placed in contact with the urothelium and 10 spectra were collected of each location with 2s integration time and stored for further analysis. One pathologist reviewed all biopsy samples. Urothelial carcinomas were graded and staged according to the WHO 1997 classification and the UICC/AJC 1992 system. Moreover, an estimation of the percentage of tumor tissue and the presence of bladder wall tissue layers (urothelium, connective tissue, muscularis, and adipose tissue) was recorded. Biopsy samples also were classified as single or multiple pathology (i.e. concurrent neoplasia and inflammation).

2.3 Data analysis

The spectral quality was assessed by the signal-to-noise ratio in the fingerprint region (400-1800wvn) of the Raman spectrum. The average signal-to-noise ratio (S/N) over the spectral region was determined by

S/N=10*log10(PRaman/Rnoise) (1)

where:

PRaman = ∫Raman2(x) dx

Pnoise = ∫(Raman-AverageRaman)2(x) dx

The collected spectra were analyzed using leave one out cross validated multivariate statistical methods to determine the prediction performance of the developed algorithm. The algorithm uses principal component analysis for data

spectrometer

laser

CCD

probe

PC

a b

Proc. of SPIE Vol. 7161 716114-2

(4)

T 40 1 60 samples [ -] 80 100 120

compression. Only principal components that show an S/N that contributes to the Raman signal were allowed in the discriminant analysis.

Data preprocessing for quantitative analysis, using multivariate statistical methods, consisted of instrument-correction of the raw spectra. Algorithms were developed using Principal-Component fed, Linear-Discriminant Analysis (PCA/LDA) in the MATLAB® environment (The MathWorks, Inc., Natick, MA)

Despite data compression by PCA, the number of remaining variables in the form of principal components (PC’s) indicates possible overfitting of the data when used as input for the discriminant analysis. The mean PC weights per pathology group retained 25 components and were evaluated to assess their relative contribution to the original spectral signal. The importance of these components for the algorithm’s prediction performance was evaluated by the coefficients of the linear discriminant functions. The combined factor of PC weights and LD coefficients determined the relative importance for each component in the prediction of the pathology groups by the developed algorithm. Only the PC’s that evidently showed spectral information above noise background were included in the final model.

3. RESULTS

In total 2,949 Raman spectra were recorded ex-vivo from cold cup biopsy samples with 2 seconds integration time. The spectral quality was assessed by the average signal-to-noise ratio over the fingerprint region. The noise spectrum was determined from the standard deviation of ten spectra obtained from one single location. The Raman power spectrum is the squared raw spectrum minus the average noise. The average signal-to-noise ratio over the spectral range was found to be over 10dB for most biopsy samples.

Fig. 2. Average signal to noise ratio over the fingerprint region of the Raman spectra of bladder wall biopsies, b) The Clinical Raman System in the OR during transurethral resection of a bladder tumor.

The analysis contained 1,287 spectra from 57 samples obtained in 30 patients. The pathology in these biopsies are given in table 1. Samples with multiple pathologies consisted of normal samples with (benign) deviations or (pre)malignant lesions with concurrent inflammation of different severity. CIS also concurred with high grade carcinoma’s, in various volume percentages.

Table 1. Pathology classification and single / multiple pathology group classification.

Pathology Sample of single pathology Sample multiple pathology Total Samples

Normal 18 24 42 Cystitis 8 8 16 CIS 9 5 14 G1 2 1 3 G2 13 2 15 G3 7 3 10 Atypia 0 7 7

Samples per group 57 50 107

a b

(5)

PC 1 PC 2 PC 3 0.2 0.5 0.5 0.2 0 0 0 0 -0.2 -0.5 -0.5 -0.2 0 1000 2000 0 1000 2000 0 1000 2000 0 PC 4 ) \It"MtVrn`}YV r': PC 5 PC 6 PC 7 0.2 0.5 0.2 0.5 0 -0.2 -0.5 -0.2 -0.5 0 1000 2000 0 1000 2000 0 1000 2000 0 0.5 0

ú

-0.5 V o U N CT) U

a

1000 2000 PC 8 PC 9 PC 10 PC 11 0.2 0.5 0.5 0

ili*I

0 0,,,,4004,,,,A-1 . 0 -0.2 -0.5 -0.5 1000 2000 0 1000 2000 0 1000 2000 0 1000 2000 PC 12 PC 13 PC 14 PC 15 0.2 0.5 0.5 0

0 W

0 w 0

-0.2 -0.5 1000 2000 0 1000 2000 PC 17 PC 18 0.5 0.2 0.2 0 0

,+yl

0 -0.5 -0.2 -0.2 0 1000 2000 0 1000 2000 PC 21 PC 22 0.2 0.5 0.2 -0.2 -0.5 -0.2 0 1000 2000 0 1000 2000 0 1000 PC 19 2000 0 1000 PC 23 2000 0 1000 2000 Raman Shift [cm -1] -0.5 0.5 0 -0.5 0.2 o -0.2 1000 2000 PC 16 0016,.*,,Wn 0 1000 2000 PC 20 0 1000 2000 PC 24 0 1000 2000

Fig. 3. Loadings of the first 24 Principal Components in Raman spectra obtained from bladder wall biopsies.

Figure 3 shows the component loadings of the first 24 components. Though they describe the variance of the dataset and no direct relation can be established with the signal to noise ratio of the Raman spectra. The signal quality can be determined from the number of components that clearly show Raman features. Beyond component 15 Raman features

(6)

0.5 LL -0.5 5 10 15 Principal Component I -1 20 _I_ 0.1 0.05 o cJi -0.05 a 5 10 15

Principal Corn pone 5 [ -]

20 25 30 Score Combined Score Combined Score Combined Score o 6 6 6 6 6 6 Combined Score Combined Score Combined Score Combined

can no longer be distinguished from the background noise. Therefore the algorithm used in the analysis only incorporates the first 15 components to predict the pathology class of the biopsies.

Fig. 4. Mean Principal Component weights per pathology group, b) Linear Discriminant Function Coefficients for retained Prinicipal Components. c) Combined PC*LD weight factor for group discrimination in prediction model.

Figure 4a shows the principal component scores combined with the weights of the linear discriminant functions in 4b this shows the relative importance of each PC for the discrimination of the pathology groups.

a

b

c

(7)

!I11I

o

g

!I11I

o

g

A multivariate algorithm allowed differentiation of normal and malignant tissues with an overall sensitivity and specificity of 78,5% and 78,9% respectively. The prediction of multiple pathologies shows some misclassifications in figure 5b possibly related to the concurrent other pathologies.

Fig. 5. Prediction results of the 7group algorithm of bladder wall biopsies in cases of one pathology and b) samples with multiple pathologies.

4. CONCLUSIONS

The quality of Raman spectra are determined here by assessing the average signal to noise ratio over the spectral range. The average noise spectrum is determined by the mean of the spectra, which then still contains (1/√n)*noise of the noise in the spectrum. As the signal to noise improves, smaller Raman features will likely show in higher order principal components which may contribute to the discrimination of different pathologies.

Multiple pathologies within a given sample pose a difficulty for Raman probes with large collection volumes. The region of interest in bladder cancer detection is limited to the upper urothelium where transitional cell carcinoma of the bladder it is the layer of origin. No suitable probes are available which focus the collection optics at the layer of interest, which makes accurate detection of superficial lesions complicated particularly in biopsies with confounding pathologic deviations like inflammations.

These results show the possibility of discerning normal from malignant bladder tissue by means of Raman spectroscopy using a small fiber based endoscopic Raman probe in ideal cases and the necessity to focus the Raman collection optics to the superficial layer of the bladder for cancer diagnosis.

REFERENCES

[1]_{A. Molckovsky, L. M. Song, M. G. Shim, N. E. Marcon, and B. C. Wilson, "Diagnostic potential of near-infrared}

Raman spectroscopy in the colon: differentiating adenomatous from hyperplastic polyps," Gastrointest.Endosc. 57(3), 396-402 (2003)).

[2]_{T. C. B. Schut, M. J. H. Witjes, H. J. C. M. Sterenborg, O. C. Speelman, J. L. N. Roodenburg, E. T. Marple, H. A.}

Bruining, and G. J. Puppels, "In vivo detection of dysplastic tissue by Raman spectroscopy," Analytical Chemistry 72(24), 6010-6018 (2000)

[3]_{L. M. W. K. Song, M. G. Shim, B. C. Wilson, S. Hassaram, M. Cirrocco, G. P. Kandel, P. P. Kortan, G. B. Haber,}

and N. E. Marcon, "Identifying dysplasia within Barrett's esophagus using Raman spectroscopy," Gastrointestinal Endoscopy 51(4), AB226-AB226 (2000)

[4]_{A. Robichaux, H. Shappell, B. Huff, H. Jones, and A. Mahadevan-Jansen, In vivo detection of cervical dysplasia}

using near infrared Raman spectroscopy, Lasers in Surgery and Medicine,14, 3-3 (2002)

[5]_{P. Crow, J. S. Uff, J. A. Farmer, M. P. Wright, and N. Stone, The use of Raman spectroscopy to identify and}

characterize transitional cell carcinoma in vitro, BJU. Int. 93, 1232-1236 (2004)

a b