Quadratic of Half Ellipse Smoothing Technique for Cervical ... - Core

Available online at www.sciencedirect.com

ScienceDirect Procedia Computer Science 59 (2015) 133 – 141

International Conference on Computer Science and Computational Intelligence (ICCSCI 2015)

Quadratic of Half Ellipse smoothing technique for cervical cells FTIR spectra in a screening system Yessi Jusmana, Nor Ashidi Mat Isab, Siew Cheok Nga,Siti Nurul Aqmariah Mohd Kanafiahc, Noor Azuan Abu Osmana a

Department of Biomedical Engineering, Faculty of Engineering Building, University of Malaya, 50603 Kuala Lumpur, Malaysia bSchool of Electrical & Electronic Engineering Universiti Sains Malaysia, Engineering Campus, 14300 Nibong Tebal, Penang, Malaysia cSchool of Microelectronic Engineering, Universiti Malaysia Perlis, Pauh Putra Campus, 02600 Arau, Perlis, Malaysia

Abstract Fourier Transform Infrared (FTIR) spectroscopy technology has been currently used to study the structural changes of cells at the molecular level in a cervical cancer. However, in the FTIR data acquisition process, it has undergone twice noise degradations. In this study, to apply the FTIR spectra for input data into classification tools, the FTIR spectra need to be smoothed before feature extraction process. A smoothing technique namely Quadratic of Half Ellipse (QHE) filter was proposed for smoothing the cervical cells FTIR spectra. The QHE filter was tested using 850 cervical cells FTIR spectra. The cervical cells FTIR spectra were corrupted by 20, 25, 30, 25, 40, 45 dB of S/N. The proposed filter achieved good performance for smoothing the corrupted cervical cells FTIR spectra. © 2015The TheAuthors. Authors.Published Published Elsevier © 2015 by by Elsevier B.V.B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of organizing committee of the International Conference on Computer Science and Peer-review under responsibility of organizing Computational Intelligence (ICCSCI 2015).committee of the International Conference on Computer Science and Computational Intelligence (ICCSCI 2015)

Keywords:Cervical precancerous; smoothing filter; feature extraction; and intelligent classification.

1. Introduction Fourier Transform Infrared (FTIR) spectroscopy technology has been used to measure content and detect chemical compounds in many materials. For medical applications, FTIR has been currently used to detect the abnormality of tissue and/or cells. Applications of the FTIR for cervical cancer screening have been done by several researchers in the world. Based on the FTIR detection, the structural changes of cells have been studied at the molecular level in cervical cancer1-4. The structural changes are produced from carcinogenesis. Different modes of vibration in the molecules of the cells and tissues are existed if the cells and tissues are induced by infrared light.

1877-0509 © 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of organizing committee of the International Conference on Computer Science and Computational Intelligence (ICCSCI 2015) doi:10.1016/j.procs.2015.07.524

134

Yessi Jusman et al. / Procedia Computer Science 59 (2015) 133 – 141

Based on the vibrations, the existed unique vibrational frequencies of major functional groups are characterized by the changes in the FTIR spectra. Thus, the FTIR spectra could show normal or malignant cells with their spectral characteristic appearance 2. Besides the capability of the FTIR technique, measurement results in form of spectra data are disturbed by noise twice in the data acquisition step. Due to these noises, the features extraction and analyzing purpose can be corrupted5-7. The noises usually appear as curves dinky and short peak. For features extraction purpose, the noise which exists in real peak absorbance and slope of cervical cell FTIR spectra can heckle in computing features related with the high of peaksand/or slope in features extraction process. Based on the literature, the peak and slope could be significant features for FTIR spectra. Based on the literature review, Savitzky-Golay (SG) filter has been used by almost all spectroscopic software packages for standard smoothing technique8-10. The SG filter uses the Least-Squares Polynomial (LSP) method to calculate its filter coefficients11, 12.Many researchers have used and provided the testimonial over the capability of the SG filters to increase the performance of signal smoothing. The SG filter has been used in many applications (i.e. cervix, breast, skin xerosis, aortic, parathyroid, forensic investigation)by Raman, FTIR and FTIR micro spectroscopies5-7, 13-16. In these aforementioned applications, the SG filter can emphasize the steep edges and increase the Signal to Noise Ratio (SNR) of the spectra. However, in detail of other applications, the additional false negative signals have been existed in SG smoothing filter at the shoulders of each vibrating band 10. The SG smoothing filter also can lead to the loss of weak signals and the reduction of spectral resolution17. Therefore, based on the SG filter frailness, an alternative method (i.e. Chebyshev filter) has been applied for smoothing the optic spectra18. However, the Chebyshev filter provides the highest cut-oơ frequency among the five points of the SG filter. Therefore, due to the limitations of the smoothing filters, the proposed signal preprocessing technique to smooth the FTIR spectra appears as contribution in this study. The proposed signal smoothing filter will apply as preprocessing technique of the cervical cell FTIR spectra which are used in this study. 2. Proposed Signal Smoothing Technique The SG, Chebyshev, and Binomial filters are examples of the direct forms that are normally used for the smoothing technique in spectral data. Details of these smoothing techniques and applications are presented 11, 18-21. In the current paper, new signal smoothing techniques is proposed for the preprocessing stage to remove the noise from the FTIR spectra as shown in Figure 1. As presented in Figure 1, a new direct form filter namely Quadratic of Half Ellipse (QHE) filter is proposed for smoothing of FTIR spectrum. For the first signal smoothing technique, this study proposes a new direct (singular) form filter namely a Quadratic of Half Ellipse (QHE) filter as shown in Figure 1. The QHE filter is proposed to enhance the conventional smoothing filters (i.e. SG, Chebyshev, and Binomial filters). The basic idea of the QHE filter is ellipse and the square out of the ellipse equation as defined in Equations (1) and (2), respectively.

Fig. 1. The proposed smoothing filter namely QHE filter.F

135


n2 x2 n

f ( x)

( f ( x)) 2

n2 x2 n

(1)

(2)

As presented in20-22, there are three conditions which have to be fulfilled by a coefficient filter. From the three conditions, the binomial filter fulfills all conditions of a smoothing filter. This filter produces less undesirable effects on the signal and it is easy to be implemented in the hardware23. This filter is proven to work as fast as the SG filter21. However, according to Battistoni et al. (1995), this filter was also proven to have better performance than the SG filter when the filter width did not exceed 17 points21. Thus, inspired by the requirement of the coefficient filter to fulfill those three conditions, this study proposes QHE filter which the coefficients are obtained by using Equation (3).

b( k )

2 2 §1· n x n¨ ¸ n ©n¹

(3)

where b(k) is the QHE coefficient filter, n is the order of the coefficient filter and k is the range of the filter from –n to n.

3. Results and Discussions The ellipse is one of the most important curves in mathematics. By using analysis of the equation and the geometry of the curve, when the QHE coefficient filter is plotted in x and y axes, it can be observed that the curve is similar to that of the Binomial coefficient filter as shown in Figure 2. Thus, it is believed that the QHE could be used as a good filter in this study. In this Section, analyses of smoothing techniques to prove the noise reduction capability are presented both qualitatively and quantitatively. For qualitative analysis, the capability of the proposed filters are illustrated using three different cervical cell FTIR spectra namely Spectrum 1 (normal cell spectrum), Spectrum 2 (LSIL cell spectrum), and Spectrum 3 (HSIL cell spectrum) corrupted by Gaussian noise with 20, 30, and 40 dB (S/N), respectively. The reference, noisy and smoothed spectra are plotted to show the smoothing effect on the spectra. Then, the SNR values are compared for all methods for the quantitative analysis. Furthermore, the capabilities of the smoothing techniques are tested using 850 cervical spectra corrupted by six levels of Gaussian noise (i.e. 20, 25, 30, 35, 40, and 45 dB of S/N). For the qualitative analysis, the reference spectra are shown in Figure 3(a), 4(a), and 5(a). Figure 3(b), 4(b), and 5(b) are the noisy spectra corrupted with 20, 30, and 40 dB of S/N, respectively. In each Figure, spectra (c) to (f) represent the smoothed spectra by using direct smoothing filters (i.e. SG, binomial, Chebyshev, and QHE filters, respectively).

136


Fig. 2. The coefficient filters of the QHE and the binomial filters which are plotted in x and y axes.

Generally, the results depict that the resultant spectra produced by all direct smoothing filters in Figures 3(c) to 3(f), 4(c) to 4(f), and 5(c) to 5(f) could smooth the noisy spectra in Figures 3(b), 4(b), and 5(b), respectively. However, the proposed QHE smoothing filter (i.e. as shown in Figures 3(f), 4(f), and 5(f)) depicts the best result compared to the other three direct smoothing filters (i.e. SG, binomial, and Chebyshev filters) as shown in Figure 3(c) to 3(e), 4(c) to 4(e), and 5(c) to 5(e), respectively. The resultant spectra produced by the other three direct smoothing filters are still influenced by noise, which indicate that these algorithms are less robust to the noise mixture. Thus, it is proven that the proposed QHE algorithm can remove the effect of noise better than the other conventional direct smoothing filters. Therefore, the QHE filter is chosen to be used in the automated cervical precancerous screening system based on the cervical cell FTIR spectra. In this section, the results of quantitative evaluation of the smoothing results for Figures 3(c) to 3(f), 4(c) to 4(f), and 5(c) to 5(f) are tabulated in Table 1. Higher value of SNR indicates better smoothing results. The best results are made bold. From Table 1, the SNR values obtained by the direct smoothing filters (i.e. SG, Binomial, Chebyshev, and QHE filters) are compared for the three selected spectra corrupted in different noise levels. Overall, the QHE smoothing filter produces the highest SNR value as compared to other direct smoothing filters for all selected spectra. This finding successfully proves the best visual evaluation obtained by the proposed QHE filter in previous section. Table 1. Quantitative evaluations on smoothing the three selected spectra corrupted with noise until 20 dB, 30 dB, and 40 dB

Spectra names Spectrum 1 Spectrum 2 Spectrum 3

SNR 20 30 40

SG 20.93 24.72 38.76

Direct smoothing filters Binomial Chebyshev 24.95 20.89 30.69 26.55 43.84 37.77

QHE 27.38 33.93 48.02


137

Fig. 3. Resultant FTIR Spectrum 1 after applying: (a) original spectrum 1, (b) noisy spectrum, (c) SG filter, (d) binomial filter, (e) Chebyshev filter, (f) QHE filter.

Total 850 cervical cell FTIR spectra are then corrupted with noise ranging from 20 to 45 dB of S/N for further testing. Using the aforementioned quantitative analysis, the average performance of direct smoothing filters on the tested spectra is shown in Figure 6. From Figure 6, it can be seen that the proposed QHE smoothing filter provides the highest value of SNR which demonstrates its capability in reducing the noise effect of the FTIR spectra. Overall, it indicates that the QHE smoothing filter is the best direct smoothing filter in this study. The capability of the QHE smoothing filter is due to the coefficient filter of the QHE technique fulfilled the certain technique to be used as a smoothing filter20, 22. Based on the presented results, the spectrum of SG and Chebyshev filters appear as failed smoothing techniques for the cervical cell FTIR spectra. Binomial and the QHE filters have good performances as a smoothing filter. As stated in the literature review, the five point coefficients of the SG, Chebyshev, Binomial, and the QHE filters are

138


tabulated in Table 2. The last coefficients (Np, -Np) of the five points’ Chebyshev filter shown in Table 2 are negative, similar to the last coefficients of the SG filter. As indicated by Bromba and Ziegler (1983), Marchand and Marmet (1983), Bromba and Ziegler (1983), Battistoni et al. (1995), the following conditions for coefficients filter (i.e. bk) must be fulfilled for a certain technique to be used as a smoothing filter20, 22: (1) The sum of the coefficients must be equal to 1; (2) The filter coefficients must be symmetrical with the central coefficient b0, whereby bk = -bk; (3) The sequence of the filter’s coefficients should be b > b > .. > b > 0.

Fig. 4. Resultant FTIR Spectrum 2 after applying: (a) original spectrum 2, (b) noisy spectrum, (c) SG filter, (d) binomial filter, (e) Chebyshev filter, (f) QHE filter


139

Condition (1) ensures the conservation of the peak area and a constant background. Meanwhile, conditions (2) and (3) avoid a phase shift between input and output data, and avoid undesired oscillations at both sides of the peak, known as the wing effects. The last condition is not completely obeyed by the SG and Chebyshev filters, as the last coefficients (Np, -Np) are always negative. These negative coefficients introduce some small oscillations at the peak sides20. The QHE filter performs the best smoothing filter among the SG, Chebyshev, and Binomial filters. In addition, the QHE coefficient filter condition (i.e. center coefficient filter is smaller than the Binomial filter) can improve the QHE filter results due to the wing effects is pressed and the goal point of the spectrum is appeared in high performances due to the center coefficient filter condition. Therefore, the results of the QHE filter is the better than the Binomial filter.

Fig. 5. Resultant FTIR Spectrum 3 after applying: (a) original spectrum 3, (b) noisy spectrum, (c) SG filter, (d) binomial filter, (e) Chebyshev filter, (f) QHE filter.

140


Table 2. Filter coefficients for different smoothing filters using five data points as well as the respective transfer function

Filters

Coefficients

Transfer function

LSP degree 2 (SG filter)

[-3, 12, 17, 12, -3]/35

ሾଵ଻ାଶସ ୡ୭ୱሺఠሻି଺ ୡ୭ୱሺଶఠሻሿ

Binomial filter

[1, 4, 6, 4, 1]/16

ሾ଺ା଼ ୡ୭ୱሺఠሻାଶ ୡ୭ୱሺଶఠሻሿ

Chebyshev filter

[-1, 4, 10, 4, -1]/16

ሾହାସ ୡ୭ୱሺఠሻିୡ୭ୱሺଶఠሻሿ

The QHE filter

[1, 4, 5, 4, 1]/15

ሾହା଼ ୡ୭ୱሺఠሻାଶ ୡ୭ୱሺଶఠሻሿ

ଷହ ଵ଺ ଼ ଵହ

Fig. 6. Performance comparison of different direct smoothing filters for 850 noisy cervical cell FTIR spectra.

4. Conclusions In this paper, a smoothing technique which is called Quadratic of Half Ellipse (QHE) filter has been presented for cervical cells FTIR spectra. The QHE filter has achieved better smoothing results than the other direct forms filter which are used for comparison in this paper. In the further application, the QHE filter can also be used for smoothing technique for other types of spectrum. Acknowledgements This research is supported by University of Malaya Postgraduate Research Fund PG083-2013B


141

References 1. Wong, PTT, Wong, RK, Caputo, TA, Godwin, TA and Rigas, B, Infrared spectroscopy of exfoliated human cervical cells: Evidence of extensive structural changes during carcinogenesis, Proc Natl Acad Sci USA 1991; 88:10988-10992 2. Fung, MFK, Senterman, M, Eid, P, Faught, W, Mikhael, NZ and Wong, PTT, Comparison of fourier-transform infrared spectroscopic screening of exfoliated cervical cells with standard papanicolaou screening, Gynecol Oncol 1997;66:10-15 3. El-Tawil, SG, Adnan, R, Muhamed, ZN and Othman, NH, Comparative study between Pap smear cytology and FTIR spectroscopy: A new tool for screening for cervical cancer, Pathology 2008;40:600-603 4. Purandare, NC, Patel, II, Trevisan, J, Bolger, N, Kelehan, R, von Bunau, G, Martin-Hirsch, PL, Prendiville, WJ and Martin, FL, Biospectroscopy insights into the multi-stage process of cervical cancer development: probing for spectral biomarkers in cytology to distinguish grades, Analyst 2013;138:3909-3916 5. Majumder, SK, Kanter, E, Viehoever, AR, Jones, H and Mahadevan-Jansen, A, Near-infrared Raman spectroscopy for in-vivo diagnosis of cervical dysplasia - A probability-based multi-class diagnostic algorithm, Proc SPIE 2007; 6430: 6. Baheri, N, Miranbaygi, M and Malekfar, R, Improved skin xerosis detection by combining extracted features from raman spectra, ISABEL 2009 7. Banas, K, Banas, A, Moser, HO, Bahou, M, Li, W, Yang, P, Cholewa, M and Lim, SK, Multivariate analysis techniques in the forensics investigation of the postblast residues by means of fourier transform-infrared spectroscopy, Anal Chem 2010;82:3038-3044 8. Hargittai, S, Savitzky-Golay least-squares polynomial filters in ECG signal processing, Computers in Cardiology 2005; 32:763-766 9. Luo, J, Ying, K and Bai, J, Savitzky-Golay smoothing and differentiation filter for even number data, Signal Processing 2005;85:1429-1434 10. Člupek, M, Matějka, P and Volka, K, Noise reduction in Raman spectra: Finite impulse response filtration versus Savitzky–Golay smoothing, J Raman Spectrosc 2007;38:1174-1179 11. Savitzky, A and Golay, MJE, Smoothing and differentiation of data by simplified least squares procedures, Anal Chem 1964;36:1627-1639 12. Staggs, JEJ, Savitzky-Golay smoothing and numerical differentiation of cone calorimeter mass data, Fire Safety Journal 2005;40:493-505 13. Majumder, SK, Keller, MD, Boulos, FI, Kelley, MC and Mahadevan-Jansen, A, Comparison of autofluorescence, diffuse reflectance, and Raman spectroscopy for breast tissue discrimination, J of Biomed Opt 2008;13:054009 14. Fabian, H, Thi, NAN, Eiden, M, Lasch, P, Schmitt, J and Naumann, D, Diagnosing benign and malignant lesions in breast tissue sections by using IR-microspectroscopy, Biochimica et Biophysica Acta - Biomembranes 2006;1758:874-882 15. Bonnier, F, Rubin, S, Ventéo, L, Krishna, CM, Pluot, M, Baehrel, B, Manfait, M and Sockalingum, GD, In-vitro analysis of normal and aneurismal human ascending aortic tissues using FT-IR microspectroscopy, Biochimica et Biophysica Acta - Biomembranes 2006;1758:968973 16. Das, K, Stone, N, Kendall, C, Fowler, C and Christie-Brown, J, Role of Fourier transform infrared spectroscopy (FTIR) in the diagnosis of parathyroid pathology, Photodiagnosis and Photodynamic Therapy 2007;4:124-129 17. Battistoni, C, Mattogno, G and Righini, G, Spectral noise removal by new digital smoothing routine, J Electron Spectrosc 1995;74:159-166 18. Palacio, C, Pascual, C, Suarez, F and Lloret, I, Smoothing of digital spectroscopic data by using a Chebyshev filter, Vacuum 2002;64:481485 19. Gorry, PA, General least-squares smoothing and differentiation by the convolution (Savitzky-Golay) method, Analytical Chemistry 1990;62:570-573 20. Marchand, P and Marmet, L, Binomial smoothing filter: A way to avoid some pitfalls of least-squares polynomial smoothing, Rev Sci Instrum 1983;54:1034-1041 21. Battistoni, C, Kaciulis, S, Mattogno, G and Righini, G, Noise removal from Auger images by using adaptive binomial filter, Journal of Electron Spectroscopy and Related Phenomena 1995;76:399-404 22. Bromba, MUA and Ziegler, H, Digital filter for computationally efficient smoothing of noisy spectra, Anal Chem 1983;55:1299-1302 23. Aubury, M, Binomial Filters, Journal of VLSI Signal Processing 1995;i:1-8 Van der Geer J, Hanraads JAJ, Lupton RA. The art of writing a scientific article. J Sci Commun 2000;163:51-9.