Biopolymers - 2006 - Chowdary

M. V. P.
Chowdary1,2
K. Kalyan Kumar1,2
Jacob Kurien2
Discrimination of Normal,
Stanley Mathew3 Benign, and Malignant Breast
C. Murali Krishna1
1
Tissues by Raman Spectroscopy
Center for Laser
Spectroscopy, Manipal
Academy of Higher Education,
Manipal-576 104,
Karnataka, India
2
Department of Surgical
Oncology, Shirdi Sai Baba
Cancer Hospital and Research
Center, Manipal Academy of
Higher Education, Manipal-576
104, Karnataka, India
3
Department of General
Surgery, Kasturba Medical
College, Manipal Academy of
Higher Education, Manipal-576
104, Karnataka, India
Received 3 May 2006;
revised 15 July 2006;
accepted 2 August 2006
Published online 8 August 2006 in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/bip.20586
Abstract: Breast cancers are the leading cancers among females. Diagnosis by fine needle aspira-
tion cytology (FNAC) is the gold standard. The widely practiced screening method, mammography,
suffers from high false positive results and repeated exposure to harmful ionizing radiation. As with
all other cancers survival rates are shown to heavily depend on stage of the cancers (Stage 0, 95%;
Stage IV, 75%). Hence development of more reliable screening and diagnosis methodology is of
considerable interest in breast cancer management. Raman spectra of normal, benign, and malig-
nant breast tissue show significant differences. Spectral differences between normal and diseased
breast tissues are more pronounced than between the two pathological conditions, malignant and
benign tissues. Based on spectral profiles, the presence of lipids (1078, 1267, 1301, 1440, 1654,
1746 cm–1) is indicated in normal tissue and proteins (stronger amide I, red shifted DCH2, broad
and strong amide III, 1002, 1033, 1530, 1556 cm–1) are found in benign and malignant tissues. The
major differences between benign and malignant tissue spectra are malignant tissues seem to have
an excess of lipids (1082, 1301, 1440 cm–1) and presence of excess proteins (amide I, amide III, red
shifted DCH2, 1033, 1002 cm–1) is indicated in benign spectra. The multivariate statistical tool,
principal components analysis (PCA) is employed for developing discrimination methods. A score
of factor 1 provided a reasonable classification of all three tissue types. The analysis is further fine-
tuned by employing Mahalanobis distance and spectral residuals as discriminating parameters.
This approach is tested both retrospectively and prospectively. The limit test, which provides the
Correspondence to: C. Murali Krishna; e-mail: cm.krishna@

manipal.edu, [email protected]
Biopolymers, Vol. 83, 556–569 (2006)
# 2006 Wiley Periodicals, Inc.
556
10970282, 2006, 5, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/bip.20586 by Agricultural University of Faisalabad, Wiley Online Library on [16/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Breast Tissue Discrimination Using Raman Spectroscopy 557
most unambiguous discrimination, is also considered and this approach clearly discriminated all
three tissue types. These results further support the efficacy of Raman spectroscopic methods in dis-
criminating normal and diseased breast tissues. # 2006 Wiley Periodicals, Inc. Biopolymers 83:
556–569, 2006
This article was originally published online as an accepted preprint. The ‘‘Published Online’’ date
corresponds to the preprint version. You can request a copy of the preprint by emailing the
Biopolymers editorial office at [email protected]
Keywords: Raman spectroscopy; PCA; breast cancer
INTRODUCTION larger region. Thus, the results can be extrapolated to

in vivo/in situ conditions.
Breast cancer is the most commonly diagnosed malig- Raman spectroscopy offers certain distinct advan-
nancy of women (23% of all cancers), ranking overall tages over other optical diagnostic techniques such as
second when both sexes are considered.1 The incidence fluorescence and FTIR. The advantages include high
of breast carcinoma is increasing at the rate of 1–2% spatial resolution (down to 1 m) in the microspec-
annually. Prognosis and survival is shown to heavily troscopic model, use of less harmful NIR radiation,
depend on the stage of diagnosis. Conventional diag- less or no sample preparation, and no influence of
nostic methods, clinical examination, and mammog- water bands, which facilitates in vivo/in situ measure-
raphy have well-described limitations. These include ments. Despite these advantages not many Raman
poor detection of cancerous lesions in younger women spectroscopic studies pertaining to diagnosis of can-
or those with radiologically dense breast and a risk of cer are reported in the literature.
repeated exposure to the harmful ionizing radiations. As can be seen from the literature, mentioned above,
Mammography may detect 70 to 90% of patients with few groups have investigated the feasibility of diagnos-
breast cancer2 with a false negative rate of up to ing breast cancers by Raman spectroscopy.10–17 How-
31%.3,4 Fine needle aspiration cytology (FNAC) is the ever, for clinical application, it is necessary that a meth-
gold standard for confirmation of the diagnosis. Breast odology is verified and validated by different groups to
biopsy involves surgical excision (removal of entire establish the approach before employing it for routine
lesion), core biopsy, and FNAC to ensure proper sam- use. In view of these considerations, Raman spectros-
pling.5 Thus the present clinical diagnostic regime, at copy of normal, benign, and malignant breast tissues of
times, involves multiple biopsies with time delayed an Indian population are carried out. To the best of our
results. Therefore, there exists considerable interest in knowledge, this is the first report on Raman spectros-
developing fast, less invasive, objective methods for copy–based diagnosis of breast cancers among an In-
diagnosis and screening of breast cancers. dian population. In the present study, Raman spectra of
Optical spectroscopic techniques6–17 are fast normal, benign, and malignant tissues are recorded.
emerging as promising alternatives in biology and PCA is used as the multivariate statistical tool for the
medicine, including cancer diagnosis. Breast cancer data analysis. Scores of factors, Mahalanobis distance,
diagnosis by spectroscopic methods such as fluores- and spectral residuals are used as discriminating para-
cence,6–8 reflectance,9 and Raman10–17 are reported meters. Further a multiparametric ‘‘limit test’’ ap-
in the literature. Many of the Raman spectroscopy proach is also considered to develop more objective
studies that appeared in the literature have employed and unambiguous discrimination. The results obtained
a microspectroscopic approach.10–15 These studies are presented and discussed in the paper.
indicated very good discrimination of normal and
pathological conditions. In most of these investiga-
tions, normal spectra are dominated by lipids whereas MATERIALS AND METHODS
an abundance of proteins is seen in pathological tis-
Breast tissue samples, routine surgical resections, or biop-
sues. Microspectroscopy provides better understand-
sies were obtained in saline from the Department of Surgi-
ing of biochemical composition as spectra can be
cal Oncology, Shirdi Sai Baba Cancer Hospital, and
recorded from the selected regions.18 On the other Research Center, MAHE, Manipal, India and Department
hand, conventional Raman spectroscopy has the of General Surgery, Kasturba Hospital, MAHE, Manipal,
advantage of simulating in situ/in vivo conditions India. Tissue from uninvolved areas from the same subjects
using ex vivo samples. In this approach spectra that was used as control/normal. A mirror image of each sample
are representative as signals are collected from a was sent for histopathological certification. Spectra were
Biopolymers DOI 10.1002/bip

558 Chowdary et al.
Table I Details of Samples Used in the Study logical conditions, benign and malignant. The spectral
features of normal tissue, bands at around 1078, 1267,
Standard Samples 1301, 1440, 1654, and 1746 cm–1, were assigned to
vibrational modes of lipids (Figure 1a). The spectral
Spectrum Number Nature Histopathology
profiles of benign and malignant tissues, strong amide
I, broad amide III, and peaks at 1002, 1083, 1530, and
1–105 Normal Normal
106–206 Malignant Malignant 1556 cm–1 were attributed to vibrational modes of
207–258 Fibroadenoma Fibroadenoma proteins (Figure 1b and 1c). The major differences
between benign spectra with respect to malignant
spectra are: red shifted and broad DCH2 peak, rela-
recorded at six or more different sites on each tissue sam- tively strong amide I, and stronger and broader amide
ple. Only good spectra from clear normal, malignant, and III (Figure 1c). All of the observed spectral features
benign tissues were employed for analysis. A total of 258
may provide vital clues in understanding the differen-
sites/spectra (105 normal, 101 malignant, 52 benign) from
29 normal, 24 malignant, and 7 benign subjects were
ces in the biochemical composition of tissues.
recruited in the study. Details of the samples used in the To bring out the differences in spectral profiles
study are shown in the Table I. more clearly, difference spectra were computed by
subtracting mean normal spectrum from mean malig-
nant and benign spectra, respectively (Figure 2a and
Laser Raman Spectroscopy
b). All of the negative peaks, 1066, 1301, 1437, and
Raman spectra were recorded using the set up assembled by 1744 cm–1, were due to normal spectrum and these
us.19–21 This setup consisted of a diode laser (785 nm, SDL spectral features can be assigned to lipids. The posi-
diode laser 8530, 150 mW) for excitation. Raman signals tive peaks in Figure 2a and b were due to malignant
were detected by combination of spectrometer (Spex Triax
and benign spectra, respectively. Most of the positive
320, 600 g/mm, blazed at 900 nm) and Spectrum one liquid
nitrogen cooled CCD (Jobin Yvon-Spex, Instruments S.A.,
bands seen around 857, 940, 1002, 1033, 1169, 1244,
Inc.). A holographic filter (HLBF-785.0, Kaiser Optics, 1281, 1341, 1381, 1417, 1465, 1530, 1556, 1587,
Ann Arbor, MI, USA) was used to reject unwanted lines 1637, and 1666 cm–1 can be assigned to different
from the excitation source. Rayleigh scattering was
removed by a notch filter (HSPF-5812, Kaiser Optics, Ann
Arbor, MI, USA). Other experimental conditions were inte-
gration time – 30 s and number of accumulations – 20.
These settings were kept constant for all the measurements.
The recorded spectra were calibrated with a cubic fit to
known frequencies of Tylenol (4-acetamidophenol).
Data Analysis
Baseline correction, smoothing, calibration, and normaliza-
tion with respect to DCH2 band were carried out using algo-
rithms of Grams 32 (Galactic Industries Corporation,
Salem, NH, USA). PCA was carried out using Grams PLS
Plus (Galactic Industries Corporation). PCA was carried out
under different conditions: different number of factors,
entire range (800–1800 cm–1), several selected regions, de-
rivative spectra to explore the best discrimination. In our
analysis, the 1400–1750 cm–1 spectral region with nine fac-
tors provided the best discrimination. Further analysis was
carried out under these conditions.
RESULTS AND DISCUSSION
The mean Raman spectra of normal, malignant, and

benign tissues showed very significant differences
(Figure 1). Differences between spectra profiles of
normal and pathological conditions were more pro- FIGURE 1 Mean Raman spectra of breast tissues: (a)
nounced compared with those between the two patho- normal; (b) malignant; (c) benign.
in malignant tissues, corroborate findings of earlier

studies.10–17 To verify the variability within the same
tissue, i.e., normal, benign, and malignant, spectra from
the same tissue corresponding to respective tissue type
were overlapped as shown in Figure 3. As can be seen
from the figure, the variability within a particular tissue
type is very subtle, as expected, it is limited to minor
intensity variations without any significant change in
the profile of the spectra.
One of the major advantages of spectroscopic diag-
nosis is high objectivity. This is facilitated by the fact
that the spectra are amenable to multivariate statistical
tools such as PCA, artificial neural network (ANN),
and hierarchial cluster analysis (HCA). In the present
studies, two approaches were considered to obtain dis-
crimination among the three classes of samples,
namely, normal, benign, and malignant. In the first
approach, spectra of all three classes of samples were
pooled and analyzed by PCA. Scores of factors were
explored for discrimination. In PCA, large spectral
data are decomposed into small number of independent
variations known as Factors or Principal Components
and contributions of these factors are known as Scores.
Scores of factors is one of the widely used parameters
for classification. As mentioned earlier, the selected
region of 1400–1750 cm–1 gave the best discrimina-
tion. Plots of % variation versus factors and Eigen val-
FIGURE 2 Difference spectra of normal, malignant, and
benign breast tissue spectra: (a) malignant – normal; (b) be-
nign – normal; (c) malignant – benign.
vibrational modes of backbone and amino acid resi-

dues of proteins. As mentioned earlier, the differences
between spectral profiles of benign and malignant are
less pronounced compared with those between normal
and pathological tissues. This is clearly evident from
the difference spectra (Figure 2). The bands in differ-
ence spectrum Figure 2c, which was computed by sub-
tracting the mean benign spectrum from the mean ma-
lignant spectrum, were relatively weak compared with
the other two difference spectra (Figure 2a and b). In
this spectrum, Figure 2c, all of the negative peaks were
due to the benign spectrum. Many of these features,
bands around 857, 940, 1002, 1033, 1163, 1245, 1279,
1347, 1390, 1468, 1556, 1634, and 1673 cm–1 can be
assigned to proteins, which indicates the presence of
excess proteins in the benign spectrum. The positive
peaks around 839, 879, 1082, 1301, and 1440 cm–1,
which were contributed by the malignant spectrum,
could be due to lipids. All of the vibrational modes
observed in Raman spectra of normal, malignant, and
benign tissues were assigned based on available litera-
ture data.22,23 These spectral features, dominance of FIGURE 3 Typical Raman spectral variations within a tis-
lipids in the normal spectrum and high protein content sue in breast tissues: (a) normal; (b) malignant; (c) benign.


FIGURE 4 PCA of normal, malignant, and benign Raman spectra of breast tissue: (a) Eigen
value plot; (b) total % variance; (c) loadings of factor 1; (d) loadings of factor 2; (e) loadings
of factor 3; (f) loadings of factor 8; (g) loadings of factor 9.
Chowdary et al.
560
Table II PCA of Raman Spectra of Normal, The classification based on scores of factor though
Malignant, and Benign Breast Tissues: a widely used approach could be subjective as dis-
Eigen values, Total % Variance criminating components are selected. Generally algo-
rithms of discriminant analysis are used to identify
Factor Number Eigen Value Total % Variance the most discriminating factors or principal compo-
nents.18 However, this may not provide a very good
1 1047.349 93.67731
2 27.7394 96.15838
classification under all circumstances.19–21 In view of
3 22.31309 98.15412 this we have considered another approach. In this sec-
4 7.245638 98.80218 ond approach, standard sets of each class, normal,
5 4.583775 99.21217 malignant, and benign, using pathologically certified
6 3.097776 99.48924 samples were developed. Spectral residuals (sum of
7 2.279846 99.69315 the squares of the difference between observed and
8 1.869763 99.86039 simulated spectra) and Mahalanobis distance (meas-
9 1.560891 100 ured in terms standard deviation from mean of the
training set) were computed for the spectra against
ues versus factors are shown in Figure 5. The first three the standard sets in both retrospective as well as pro-
factors accounted for 98% variation in the spectra (Ta- spective modality. And these parameters have been
ble II). Profiles of factors, ‘‘factor loads,’’ of the first explored to achieve better discrimination of tissue
three factors and the last two factors are shown in Fig- type. This approach has been successfully employed
ure 4. In our analysis, scores of factor 1 provided a rea- by us for discrimination of oral and cervical tis-
sonable classification among three groups (Figure 4). sues.19–21 The mathematical basis of the Mahalanobis
As can be seen from the figure, the contribution of fac- distance calculation is well known24–26 and widely
tor 1 is negative for normal spectra and largely positive used for spectral discrimination. It is very sensitive to
for benign and malignant spectra. Cluster of malignant intervariable changes in the calibration data and it is
spectra occupy between clusters corresponding to nor- measured in terms of standard deviations from the
mal and benign tissues. The mean and standard devia- mean of the training samples. In addition to providing
tions of score of factor 1 for normal, malignant, and spectral discrimination, it also gives a statistical mea-
benign spectra are 0.068 6 0.006, 0.028 6 0.023, sure of how well the unknown sample spectrum
and 0.083 6 0.007, respectively. Thus a reasonable matches or does not match. Thus it is a statistical
classification of three clusters can be seen up to mean measure of proximity of two spectra. In the present
6 1 standard deviation, which indicates 75% sensitiv- analysis, Mahalanobis distance is coupled with PCA,
ity/specificity. which is a very effective data reduction technique for
FIGURE 5 PCA of Raman spectra of normal, malignant, and benign breast tissues. Normal;
* Malignant; ~ Benign.

562 Chowdary et al.
spectroscopic data and thus allows full spectral cover- nign standard sets. Randomly picked spectra based on
age for all samples. In our analysis, Mahalanobis dis- scores of factor 1 and pathological certifications are
tance is computed using the expression employed to develop standard calibration sets for nor-
mal (36 spectra), malignant (35 spectra), and benign
D2 ¼ ðStest Þ M1 ðStest Þ0 (21 spectra) conditions. The standard sets were veri-
Where Stest is the vector of the scores and sum fied retrospectively where spectra from the standard
of squared spectral residuals for a given test sample, sets are matched to compute Mahalanobis distance
and M is an f by f Mahalanobis matrix given by and spectral residuals. It is expected that if spectra and
0
M ¼ ðnSS1Þ, where S contains the corresponding para- standard sets are of same class, values for Mahalano-
meters for the calibration set (n standards). bis distance and spectral residuals are low and vice
The advantages of using D2 as a discriminating pa- versa. As an example, retrospective analysis against
rameter are described elsewhere.23–25 Mahalanobis normal standard is shown in Figure 6. As can be seen
distance and spectral residuals are computed by from the figure, very good classification of tissue type
matching spectra against normal, malignant, and be- is achieved. Mahalanobis distance for all normal spec-
FIGURE 6 PCA of Raman spectra of breast tissue—retrospective study against normal standard
set: (a) plot of spectral residual versus Mahalanobis distance; (b) plot of sample number versus

Mahalanobis distance. Normal; * Malignant; ~ Benign.

tra are around 0.881 6 0.44, and values for malignant þ 1.24 and 10.52 þ 1.75, respectively. Spectral resid-
and benign spectra are around 104.42 6 28.64 and ual values for malignant, benign and normal spectra
259.85 6 28.43, respectively. Spectral residual values were around 0.65 þ 0.40, 2.60 þ 0.56 and 5.23 þ
for normal spectra are around 0.09 6 0.05 and for ma- 1.24, respectively.
lignant and benign spectra are around 8.40 6 2.27 and The above approach was further extended to com-
20.76 6 2.26, respectively. The approach was eval- pute match/mismatch tables or a ‘‘Limit test,’’ which
uated prospectively, matching all normal, benign, and is considered to obtain unambiguous discrimination
malignant spectra that were not included in the stand- of tissue types. This methodology is successfully ap-
ard sets, against standard sets. Typical results obtained plied to discriminate normal, inflammatory, premalig-
in prospective analysis against malignant standard sets nant, and malignant oral tissues and normal and ma-
are shown in Figure 7. The clear clustering, barring lignant conditions of the cervix.19–21 In this method-
minor overlap among the clusters, which could not be ology, spectra are matched against standard sets with
assigned pathological status, obtained in the analysis inclusions/exclusions set for three discrimination pa-
validates standard sets. Mean Mahalanobis distance rameters: scores of factors, Mahalanobis distance,
for all malignant spectra were around 1.19 þ 0.64, and spectral residuals. If the values of the given spec-
and values for benign and normal spectra around 5.53 trum fall within set limits, spectra show YES or PASS,
FIGURE 7 PCA of Raman spectra of breast tissues— prospective study against malignant stand-
ard set: (a) plot of spectral residual versus Mahalanobis distance; (b) plot of sample number versus

Mahalanobis distance. Normal; * Malignant; ~ Benign.

564 Chowdary et al.
Table III Multiparametric Match/Mismatch Table Compared Against Benign Standard Set
Sample Number Match Mahalanobis Distance Limit Tests Spectral Residual
1 No 154.52246 Fail (PFF#) 24.151834

2 No 166.45626 Fail (PFF#) 26.039245
3 No 165.6464 Fail (PFF#) 25.893308
4 No 162.32982 Fail (PFF#) 25.382414
5 No 134.07467 Fail (PFF#) 20.986553
6 No 162.64674 Fail (PFF#) 25.428597
7 No 157.47355 Fail (PFF#) 24.635371
8 No 158.44057 Fail (PFF#) 24.779253
9 No 159.83266 Fail (PFF#) 24.994801
10 No 159.37699 Fail (PFF#) 24.921713
11 No 162.6142 Fail (PFF#) 25.426248
12 No 157.60521 Fail (PFF#) 24.652342
13 No 131.16563 Fail (PFF#) 20.562932
14 No 151.691 Fail (PFF#) 23.748195
15 No 156.93794 Fail (PFF#) 24.567031
16 No 176.3405 Fail (PFF#) 27.573818
17 No 148.19333 Fail (PFF#) 23.214416
18 No 161.72258 Fail (PFF#) 25.300577
19 No 164.22728 Fail (PFF#) 25.689614
20 No 165.62013 Fail (PFF#) 25.903564
21 No 167.03384 Fail (PFF#) 26.124753
22 No 164.3466 Fail (PFF#) 25.70915
23 No 157.53945 Fail (PFF#) 24.651057
24 No 149.76166 Fail (PFF#) 23.433988
25 No 144.80625 Fail (PFF#) 22.665443
26 No 143.732 Fail (PFF#) 22.498237
27 No 133.58811 Fail (PFF#) 20.925826
28 No 149.64087 Fail (PFF#) 23.404721
29 No 133.97002 Fail (PFF#) 20.984014
30 No 154.35693 Fail (PFF#) 24.150431
31 No 147.04683 Fail (PFF#) 23.013104
32 No 154.66193 Fail (PFF#) 24.196715
33 No 149.92767 Fail (PFF#) 23.461226
34 No 122.29226 Fail (PFF#) 19.194846
35 No 125.49122 Fail (PFF#) 19.682771
36 No 111.86007 Fail (PFF#) 17.568149
37 No 135.13666 Fail (PFF#) 21.177365
38 No 115.80469 Fail (PFF#) 18.170427
39 No 136.49853 Fail (PFF#) 21.387595
40 No 157.53945 Fail (PFF#) 24.651057
41 No 166.14668 Fail (PFF#) 25.986709
42 No 158.07804 Fail (PFF#) 24.706919
43 No 157.88281 Fail (PFF#) 24.675314
44 No 156.44336 Fail (PFF#) 24.455824
45 No 154.94271 Fail (PFF#) 24.222595
46 No 158.96237 Fail (PFF#) 24.845705
47 No 156.19518 Fail (PFF#) 24.415974
48 No 157.04216 Fail (PFF#) 24.550775
49 No 156.82689 Fail (PFF#) 24.515457
50 No 168.44621 Fail (PFF#) 26.315474
51 No 152.94638 Fail (PFF#) 23.907725
52 No 152.60167 Fail (PFF#) 23.851668
53 No 153.20453 Fail (PFF#) 23.947045
54 No 151.95531 Fail (PFF#) 23.753628
(Continued )
Table III (Continued from the previous page.)
55 No 153.19782 Fail (PFF#) 23.94453

56 No 153.76303 Fail (PFF#) 24.033135
57 No 154.52246 Fail (PFF#) 24.151834
58 No 169.30983 Fail (PFF#) 26.457941
59 No 168.55814 Fail (PFF#) 26.350382
60 No 163.98942 Fail (PFF#) 25.638539
61 No 166.07914 Fail (PFF#) 25.984536
62 No 168.5125 Fail (PFF#) 26.341428
63 No 155.9306 Fail (PFF#) 24.393914
64 No 171.39882 Fail (PFF#) 26.796937
65 No 167.83373 Fail (PFF#) 26.246916
66 No 168.16938 Fail (PFF#) 26.298465
67 No 172.01048 Fail (PFF#) 26.872543
68 No 166.62991 Fail (PFF#) 26.060579
69 No 166.28094 Fail (PFF#) 26.011062
70 No 139.91296 Fail (PFF#) 21.921439
71 No 154.52246 Fail (PFF#) 24.151834
72 No 129.80808 Fail (PFF#) 20.356054
73 No 128.06179 Fail (PFF#) 20.099773
74 No 166.45626 Fail (PFF#) 26.039245
75 No 165.6464 Fail (PFF#) 25.893308
76 No 156.15936 Fail (PFF#) 24.432674
77 No 158.69514 Fail (PFF#) 24.823467
78 No 155.20604 Fail (PFF#) 24.28653
79 No 162.66834 Fail (PFF#) 25.432695
80 No 160.21848 Fail (PFF#) 25.053339
81 No 162.9888 Fail (PFF#) 25.483686
82 No 161.12346 Fail (PFF#) 25.183213
83 No 161.04602 Fail (PFF#) 25.170668
84 No 162.56282 Fail (PFF#) 25.404399
85 No 162.56282 Fail (PFF#) 25.404399
86 No 178.19598 Fail (PFF#) 27.830032
87 No 157.94923 Fail (PFF#) 24.693513
88 No 173.60286 Fail (PFF#) 27.127591
89 No 173.66463 Fail (PFF#) 27.138237
90 No 173.3279 Fail (PFF#) 27.085481
91 No 173.15514 Fail (PFF#) 27.057981
92 No 169.39683 Fail (PFF#) 26.478632
93 No 174.12861 Fail (PFF#) 27.209825
94 No 165.04634 Fail (PFF#) 25.787529
95 No 168.10887 Fail (PFF#) 26.268448
96 No 168.20897 Fail (PFF#) 26.286019
97 No 166.96542 Fail (PFF#) 26.091291
98 No 167.70928 Fail (PFF#) 26.207637
99 No 167.06289 Fail (PFF#) 26.107231
100 No 155.3238 Fail (PFF#) 24.285134
101 No 128.60533 Fail (PFF#) 20.147898
102 No 142.68863 Fail (PFF#) 22.318912
103 No 148.00253 Fail (PFF#) 23.152701
104 No 153.89957 Fail (PFF#) 24.071532
105 No 149.80185 Fail (PFF#) 23.438586
106 No 16.244087 Fail (PFF#) 2.7172556
107 No 11.58526 Fail (PFF#) 2.0316373
108 No 14.110687 Fail (PFF#) 2.4257418
(Continued )
566 Chowdary et al.
109 No 10.228449 Fail (PFF#) 1.7486183

110 No 7.8725281 Fail (P?F#) 1.4200823
111 No 6.9843291 Fail (P?F#) 1.2732269
112 No 11.302798 Fail (PFF#) 1.968387
113 No 10.228449 Fail (PFF#) 1.7486183
114 No 14.110687 Fail (PFF#) 2.4257418
115 No 11.58526 Fail (PFF#) 2.0316373
116 No 16.244087 Fail (PFF#) 2.7172556
117 No 15.681282 Fail (PFF#) 2.6517327
118 No 28.406374 Fail (FFF#) 4.6707034
119 No 12.422363 Fail (PFF#) 2.1603494
120 No 24.794465 Fail (PFF#) 4.0635876
121 No 11.385221 Fail (PFF#) 2.0023184
122 No 12.422363 Fail (PFF#) 2.1603494
123 No 24.794465 Fail (PFF#) 4.0635876
124 No 30.49831 Fail (PFF#) 4.876836
125 No 18.086489 Fail (PFF#) 3.0130556
126 No 16.471775 Fail (PFF#) 2.7447988
127 No 78.166704 Fail (PFF#) 12.306064
128 No 32.993165 Fail (PFF#) 5.2987373
129 No 24.840821 Fail (PFF#) 4.0333362
130 No 6.1116584 Fail (P?F#) 1.1238708
131 No 11.773123 Fail (PFF#) 2.0228927
132 No 13.904603 Fail (PFF#) 2.3101694
133 No 13.904603 Fail (PFF#) 2.3101694
134 No 5.7014316 Fail (P?F#) 1.0943288
135 No 24.831926 Fail (PFF#) 3.9955121
136 No 62.714877 Fail (PFF#) 9.8949026
137 No 17.185709 Fail (PFF#) 2.836669
138 No 56.838813 Fail (PFF#) 8.9967588
139 No 29.018779 Fail (PFF#) 4.6566396
140 No 22.232862 Fail (PFF#) 3.6031636
141 No 26.440053 Fail (PFF#) 4.2671934
142 No 14.782007 Fail (PFF#) 2.5178466
143 No 13.243161 Fail (PFF#) 2.2740299
144 No 20.545113 Fail (PFF#) 3.3621383
145 No 12.250009 Fail (PFF#) 2.0864864
146 No 78.166704 Fail (PFF#) 12.306064
147 No 32.993165 Fail (PFF#) 5.2987373
148 No 24.840821 Fail (PFF#) 4.0333362
149 No 45.012375 Fail (PFF#) 7.173343
150 No 51.733377 Fail (PFF#) 8.2105943
151 No 15.969781 Fail (PFF#) 2.5826447
152 No 13.280632 Fail (PFF#) 2.1621049
153 No 19.125931 Fail (PFF#) 3.0794604
154 No 11.054709 Fail (PFF#) 1.8187381
155 No 8.3843428 Fail (P?F#) 1.47462
156 No 11.802658 Fail (PFF#) 2.0178502
157 No 65.01548 Fail (PFF#) 10.267177
158 No 25.134037 Fail (PFF#) 4.0931434
159 No 5.8629188 Fail (P?F#) 1.1354316
160 No 26.016332 Fail (PFF#) 4.2442661
161 No 83.282837 Fail (PFF#) 13.105609
162 No 20.958097 Fail (PFF#) 3.3950776
(Continued )
163 No 30.111839 Fail (PFF#) 4.8673111

164 No 45.195132 Fail (PFF#) 7.2167438
165 No 24.757509 Fail (PFF#) 4.0127864
166 No 30.412939 Fail (PFF#) 4.8772424
167 No 37.083359 Fail (PFF#) 5.9257475
168 No 46.100841 Fail (PFF#) 7.3430807
169 No 17.349995 Fail (PFF#) 2.8247216
170 No 17.185709 Fail (PFF#) 2.836669
171 No 17.349995 Fail (PFF#) 2.8247216
172 No 56.838813 Fail (PFF#) 8.9967588
173 No 17.349995 Fail (PFF#) 2.8247216
174 No 17.349995 Fail (PFF#) 2.8247216
175 No 11.90357 Fail (PFF#) 2.0174714
176 No 44.056272 Fail (PFF#) 7.0136503
177 No 11.90357 Fail (PFF#) 2.0174714
178 No 16.609869 Fail (PFF#) 2.7551161
179 No 8.5902449 Fail (P?F#) 1.5159437
180 No 11.824622 Fail (PFF#) 1.996328
181 No 31.888074 Fail (PFF#) 5.1364167
182 No 58.025759 Fail (PFF#) 9.1835259
183 No 72.881859 Fail (PFF#) 11.491003
184 No 31.888074 Fail (PFF#) 5.1364167
185 No 89.840098 Fail (PFF#) 14.146692
186 No 28.38357 Fail (PFF#) 4.5966694
187 No 28.034792 Fail (PFF#) 4.5960093
188 No 29.390382 Fail (PFF#) 4.7889831
189 No 41.017168 Fail (PFF#) 6.5747806
190 No 23.861237 Fail (PFF#) 3.9285166
191 No 31.046557 Fail (PFF#) 5.0192138
192 No 9.7566777 Fail (PFF#) 1.7094075
193 No 42.29662 Fail (PFF#) 6.7052732
194 No 40.909812 Fail (PFF#) 6.5159057
195 No 45.355118 Fail (PFF#) 7.2081141
196 No 62.20789 Fail (PFF#) 9.8502354
197 No 42.29662 Fail (PFF#) 6.7052732
198 No 60.30373 Fail (PFF#) 9.5272639
199 No 42.29662 Fail (PFF#) 6.7052732
200 No 6.9953473 Fail (P?F#) 1.2255174
201 No 10.188441 Fail (PFF#) 1.7553443
202 No 11.574023 Fail (PFF#) 1.9788702
203 No 6.9953473 Fail (P?F#) 1.2255174
204 No 10.188441 Fail (PFF#) 1.7553443
205 No 11.574023 Fail (PFF#) 1.9788702
206 No 6.9953473 Fail (P?F#) 1.2255174
207 Possible 1.2339715 Pass (PP?#) 0.1075388
208 Yes 0.3428911 Pass (PPP#) 0.1512337
209 Yes 0.9878245 Pass (PPP#) 0.04488258
210 Yes 0.5677136 Pass (PPP#) 0.1707019
211 Yes 0.4586411 Pass (PPP#) 0.1704782
212 Yes 0.5472146 Pass (PPP#) 0.2208394
213 Possible 1.2339715 Pass (PP?#) 0.1075388
214 Yes 0.5472146 Pass (PPP#) 0.2208394
215 Yes 0.7996501 Pass (PPP#) 0.2643451
216 Yes 0.8596138 Pass (PPP#) 0.08203353
(Continued )

568 Chowdary et al.
217 Possible 1.1667353 Pass (PP?#) 0.2849572

218 Possible 1.2261093 Pass (PP?#) 0.07781574
219 Possible 2.8754408 Pass (PP?#) 0.5887623
220 Yes 0.7996501 Pass (PPP#) 0.2643451
221 Possible 1.0471809 Pass (PP?#) 0.1386804
222 Yes 0.8659271 Pass (PPP#) 0.07630237
223 Yes 0.4925431 Pass (PPP#) 0.12101
224 Yes 0.8596138 Pass (PPP#) 0.08203353
225 Yes 0.8691097 Pass (PPP#) 0.0751665
226 Possible 1.2889638 Pass (PP?#) 0.02532688
227 Possible 1.4717093 Pass (PP?#) 0.4113595
228 Possible 1.1667353 Pass (PP?#) 0.2849572
229 Yes 0.2196176 Pass (PPP#) 0.1913339
230 Yes 0.7425372 Pass (PPP#) 0.3260033
231 Possible 1.1876442 Pass (PP?#) 0.3833324
232 Yes 0.8892672 Pass (PPP#) 0.1231935
233 Possible 1.022601 Pass (PP?#) 0.1281673
234 Yes 0.3626659 Pass (PPP#) 0.1543471
235 Yes 0.555973 Pass (PPP#) 0.2657551
236 Yes 0.555973 Pass (PPP#) 0.2657551
237 Yes 0.555973 Pass (PPP#) 0.2657551
238 Possible 1.7473039 Pass (PP?#) 0.4759303
239 Yes 0.7996501 Pass (PPP#) 0.2643451
240 Possible 1.1505599 Pass (PP?#) 0.01640513
241 Possible 1.1876442 Pass (PP?#) 0.3833324
242 Possible 1.1104419 Pass (PP?#) 0.1842548
243 Possible 1.2339715 Pass (PP?#) 0.1075388
244 Yes 0.3428911 Pass (PPP#) 0.1512337
245 Yes 0.9878245 Pass (PPP#) 0.04488258
246 Possible 1.1876442 Pass (PP?#) 0.3833324
247 Yes 0.8892672 Pass (PPP#) 0.1231935
248 Possible 1.022601 Pass (PP?#) 0.1281673
249 Yes 0.68485 Pass (PPP#) 0.08948508
250 Possible 1.2261093 Pass (PP?#) 0.07781574
251 Possible 1.4559582 Pass (PP?#) 0.1709246
252 Possible 1.7274048 Pass (PP?#) 0.3727497
253 Yes 0.8892672 Pass (PPP#) 0.1231935
254 Possible 1.1505599 Pass (PP?#) 0.01640513
255 Possible 1.022601 Pass (PP?#) 0.1281673
256 Possible 1.4222331 Pass (PP?#) 0.293526
257 Yes 0.8419798 Pass (PPP#) 0.07460492
258 Yes 0.8419798 Pass (PPP#) 0.07460492
otherwise NO or FAIL. Thus any given spectrum is and malignant, show no match (NO/FAIL). Thus the
matched against all the available standard sets and spec- results obtained in the study further support the efficacy
tra to be classified as a particular class should show a of Raman spectroscopic methods in discrimination of
match (YES/PASS) against that particular standard set normal, malignant, and benign breast tissues.
and no match (NO/FAIL) against all other sets. Thus,
this methodology provides the most unambiguous dis- CONCLUSION
crimination. A typical match/mismatch table compared
against benign standard sets is given in Table III. All Raman spectra obtained from normal and pathologi-
spectra belonging to the benign condition show match cal tissues show significant differences. The spectral
(YES/PASS) and all other classes of spectra, normal profile of normal tissues is indicative of higher levels
of lipids. In comparison, the spectral profile of patho- 7. Majumder, S. K.; Gupta, P. K.; Jain, B.; Uppal, A.
logical tissues, both benign and malignant, indicates Lasers Life Sci 1998, 8, 249–264.
the presence of more proteins and fewer lipids. Further, 8. Yang, Y.; Katz, A.; Celmer, E. J.; Zurawska-Szczepa-
among the pathological tissues, malignant tissue con- niak, M.; Alfano, R. R.; Photochem Photobiol 1997,
66, 518–522.
tains relatively more lipids in comparison to benign tis-
9. Bigio, I. J.; Bown, S. G.; Briggs, G.; Kelley, C.;
sue. These findings are in corroboration with our basic
Lakhani, S.; Pickard, D.; Ripley, P. M.; Rose, I. G.;
understanding that neoplastic processes are fundamen- Saunders, C. J. Biomed Opt 2000, 5, 221–228.
tally more cellular and hence would biochemically be 10. Haka, A. S.; Volynskaya, Z.; Gardecki, J. A.; Nazemi,
observed to have higher levels of protein. Also, benign J.; Hicks, D.; Fitzumaurice, M.; Dasari, R. R.; Crowe,
tissue having more stromal components relative to the J. P.; Feld, M. S. Cancer Res 2006, 66, 3317–3322.
more cellular components of malignant tissue would 11. Abigail, S. H.; Shaefer-Peltier, K. E.; Maryann, F.; Jo-
explain our observation of relatively more lipid in ma- seph, C.; Dasari, R. R.; Feld, M. S. Proc Natl Acad Sci
lignant tissue in comparison to benign tissue. PCA is U S A 2005, 102, 12371–12376.
employed for developing discrimination methods. 12. Abigail, S. H.; Shaefer-Peltier, K. E.; Maryann, F.;
Scores of factor 1 provided a reasonable discrimination. Crowe, J.; Dasari, R. R.; Feld, M. S. Cancer Res 2002,
The analysis is further finetuned by employing Mahala- 62, 5375–5380.
nobis distance and spectral residuals as discriminating 13. Shaefer-Peltier, K. E.; Haka, A. S.; Fitzumaurice, M.;
Crowe, J.; Myles, J.; Dasari, R. R.; Feld, M. S. J Raman
parameters. Limit tests or computation of match–mis-
Spectrosc 2002, 33, 552–563.
match tables showed most unambiguous discrimina-
14. Shaefer-Peltier, K. E.; Haka, A. S.; Motz, J. T.; Fitzu-
tion. Thus the results obtained in the study further sup- maurice, M.; Dasari, R. R.; Feld, M. S. J Cell Biochem
ports the efficacy of Raman spectroscopic methods in 2002, 87, S125–S137.
discrimination of normal, malignant, and benign breast 15. Kneipp, J.; Tom, B. S.; Kliffen, M.; Marian, M. P.;
tissues. Perceptively, these results are significant as Puppels, G. Vib Spectrosc 2003, 32, 67–74.
with improvement of technique (development of fiber- 16. Manoharan, R.; Shafer, K.; Perelman, I.; Wu, J.; Chen,
optic probes), reproducibility, and data analysis, develop- K.; Deinum, G.; Fitzumaurice, M.; Myles, J.; Crowe,
ment of objective, rapid, and minimally invasive Raman J.; Dasari, R. R.; Feld, M. S. Photochem Photobiol
spectroscopic technique for breast cancer screening and 1998, 67, 15–22.
diagnosis seems quite feasible. 17. Hanlon, E. B.; Manoharan, R.; Koo, T. W.; Shafer, K.
E.; Motz, J. T.; Fitzumaurice, M.; Kramer, J. R.; Itzkan,
This work is carried out under the project entitled: ‘‘Devel- I.; Dasari, R. R.; Feld, M. S. Phys Med Biol 2000, 45,
opment of laser spectroscopy techniques for early detection R1–R59.
and follow up of therapy in breast malignancy’’ No. 5/13/ 18. Murali Krishna, C.; Sockalingam, G. D.; Kurien, J.;
23/2003-NCD-III, Indian Council for Medical Research, Lakshmi, R.; Venteo, L.; Pluot, M.; Manfait, M.; Kar-
Government of India. The authors (M.V.P.C. and K.K.K.) tha, V. B. Appl Spectrosc 2004, 58, 107–114.
are thankful to ICMR for senior research fellowships. Ms. 19. Murali Krishna, C.; Prathima, N. B.; Malini, R.; Vad-
Keerthi and Mr. Chethan N. Anand are acknowledged for hiraja, B. M.; Rani, Bhatt, A.; Donald, J. F.; Pralhad,
their technical assistance in the sample handling/preserva- K.; Vidyasagar, M. S.; Kartha, V. B. Vib Spectrosc
tion and data acquisition. 2006, 41, 136–141.
20. Malini, R.; Venkatakrishna, K.; Kurien, J.; Keethilatha
Pai, M.; Kartha, V. B.; Murali Krishna, C. Biopolymers
REFERENCES 2006, 81, 179–193.
21. Manjunath, B. K.; Kurien, J.; Rao, L.; Murali Krishna,
C.; Chidananda, M. S.; Venkatakrishna, K.; Kartha, V.
1. Max, P. D.; Freddie, B.; Ferlay, J.; Poala, P. A Cancer J
B. J Photochem Photobiol B: Biology, 2004, 73, 49–58.
Clin 2005, 55, 74–108.
2. Ciatto, S.; Rosselli, D. T. M.; Zappa, M. Br J Cancer 22. Clark, R. J. H.; Hester, R. E. In Advances in Spectros-
1995, 71, 337–339. copy; New York: Wiley, 1993.
3. Esserman, L.; Cowley, H.; Eberie, C.; Eberle, C.; Kirkpa- 23. Parker, E. S. Applications of Infrared and Raman and
trick, A.; Chang, S.; Berbaum, K.; Gale, A. J Natl Cancer Resonance Raman Spectroscopy in Biochemistry;
Inst 2002, 369–375. New York: Plenum Press, 1983.
4. Destounis, S. V.; DiNitto, P.; Logan-Young, W.; 24. Mahalanobis, P. C. Proc Natl Inst Sci India, 1936, 2, 49.
Bonaccio, E.; Zuley, M. L.; Willison, K. M. Radiology 25. Mark, H. L. Anal Chem 1987, 59, 790.
2004, 232, 578–584. 26. PLSplus/IQ User’s Guide; Galactic Industries Corpora-
5. Liberman, L.; Dershaw, D. D.; Rosen, P. P.; Abramson, A. F.; tion: Salem, New Hampshire, USA, 1999.
Deutch, B. M.; Hann, L. E. Radiology 1994, 192, 793–795.
6. Gupta, P. K.; Majumder, S. K.; Uppal, A. Lasers Surg
Med 1997, 21, 417–422. Reviewing Editor: Laurence Nafie

Biopolymers - 2006 - Chowdary

Uploaded by

Copyright:

Available Formats

Biopolymers - 2006 - Chowdary

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Biopolymers - 2006 - Chowdary

Uploaded by

Copyright:

Available Formats

M. V. P.

Correspondence to: C. Murali Krishna; e-mail: cm.krishna@

Keywords: Raman spectroscopy; PCA; breast cancer

INTRODUCTION larger region. Thus, the results can be extrapolated to

Biopolymers DOI 10.1002/bip

RESULTS AND DISCUSSION

The mean Raman spectra of normal, malignant, and

in malignant tissues, corroborate ﬁndings of earlier

vibrational modes of backbone and amino acid resi-

Biopolymers DOI 10.1002/bip

Biopolymers DOI 10.1002/bip

Biopolymers DOI 10.1002/bip

Biopolymers DOI 10.1002/bip

Biopolymers DOI 10.1002/bip

Sample Number Match Mahalanobis Distance Limit Tests Spectral Residual

1 No 154.52246 Fail (PFF#) 24.151834

Table III (Continued from the previous page.)

Sample Number Match Mahalanobis Distance Limit Tests Spectral Residual

55 No 153.19782 Fail (PFF#) 23.94453

Table III (Continued from the previous page.)

Sample Number Match Mahalanobis Distance Limit Tests Spectral Residual

109 No 10.228449 Fail (PFF#) 1.7486183

Table III (Continued from the previous page.)

Sample Number Match Mahalanobis Distance Limit Tests Spectral Residual

163 No 30.111839 Fail (PFF#) 4.8673111

Biopolymers DOI 10.1002/bip

Table III (Continued from the previous page.)

Sample Number Match Mahalanobis Distance Limit Tests Spectral Residual

217 Possible 1.1667353 Pass (PP?#) 0.2849572

Biopolymers DOI 10.1002/bip

You might also like