Photomedicine and Laser Surgery
Volume 27, Number 3, 2009
© Mary Ann Liebert, Inc.
Pp. 425–433
DOI: 10.1089/pho.2008.2316
Optimum Wavelength for the Differentiation of Brain Tumor
Tissue Using Autofluorescence Spectroscopy
Ariya Saraswathy, MPhil,1 R.S. Jayasree, Ph.D.,2 K.V. Baiju, MPhil,3
Arun Kumar Gupta, M.D.,2 and V.P. Mahadevan Pillai, Ph.D.1
Abstract
Objective: The role of autofluorescence spectroscopy in the detection and staging of benign and malignant brain
tumors is being investigated in this study, with an additional aim of determining an optimum excitation wavelength for the spectroscopic identification of brain tumors. Materials and Methods: The present study involves
in-vitro autofluorescence monitoring of different human brain tumor samples to assess their spectroscopic properties. The autofluorescence measurement at four different excitation wavelengths 320, 370, 410, and 470 nm,
were carried out for five different brain tumor types: glioma, astrocytoma, meningioma, pituitary adenoma,
and schwannoma. Results: The fluorescence spectra of tumor tissues showed significant differences, both in
intensity and in spectral profile, from those of adjacent normal brain tissues at all four excitation wavelengths.
The data were then subjected to multivariate statistical analysis and the sensitivities and specificities were calculated for each group. Of the four excitation wavelengths being considered, 470 nm appeared to be the optimal wavelength for detecting tissue fluorescence of brain tumor tissues. Conclusions: In conclusion, the spectroscopic luminescence measurements carried out in this study revealed significant differences between tumor
tissue and adjacent normal tissue of human brains for all the tumor types tested, except for pituitary adenoma.
From the results of this study we conclude that excitation wavelengths ranging from 410–470 nm are most suitable for the detection of brain tumor tissue. Moreover, in this particular study, only excitation at 470 nm indicated that samples we considered to be normal tissue were not normal, and that these were indeed pituitary
adenoma tissues. This distinction was not clear at other excitation wavelengths.
Introduction
S
URGERY IS THE MOST COMMON METHOD of treatment for
brain tumors. Surgeons delineate tumor margins via the
use of imaging modalities such as computed tomography
(CT), magnetic resonance imaging (MRI), or ultrasound.
Here the primary aim of the surgeon is to remove as much
of the tumor as possible, with minimal neurological damage
to unaffected areas. But in many cases the surgeon is unable
to visually differentiate the tumor tissue from normal tissue
and they seek pathological support, which is time consuming. Thus there is need for a better method of differentiating
tumor tissues from normal tissues. Recently, fluorescence
spectroscopy has been used in medicine to characterize various metabolic and pathological changes at the cellular and
tissue levels.1,2 Diagnostic techniques based on spectroscopy
have potential use for the assessment of biochemical and
morphological properties of tissues to aid in patient care.
These techniques are fast, noninvasive and quantitative, and
they can be used to elucidate key tissue features. Advances
in the development of laser energy sources, fiberoptics, and
detector technology may soon allow optical diagnosis to be
carried out remotely in real time. Although several optical
techniques such as infrared (IR) spectroscopy, Raman spectroscopy, reflectance spectroscopy, and elastic scattering
spectroscopy have been proposed for use in such an optical
biopsy system, most studies have explored the potential of
laser-induced fluorescence spectroscopy for tissue characterization and imaging.3,4 An important feature of fluorescence is its high intrinsic sensitivity and rapid noninvasive
ability to analyze tissues in the body in situ, with the help of
endoscope-compatible fiberoptic probes. Diagnostics can be
1Department of Optoelectronics, University of Kerala, Kariavattom, 2Department of Imaging Sciences and Interventional Radiology, Sree
Chitra Tirunal Institute for Medical Sciences and Technology, and 3Department of Statistics, University of Kerala, Kariavattom, Trivandrum, Kerala, India.
425
426
SARASWATHY ET AL.
easily automated using this technique and may allow analysis by less skilled personnel and mass screening.5,6 Due to
its intrinsic sensitivity, rapid results, and safe use, fluorescence techniques are becoming widely accepted for quantitative analysis of clinical samples, and the range of applications continues to expand.
In vitro delineation of brain tumor tissue using optical spectroscopy has been reported by Lin et al. using the nitrogen
laser, and later Lin and others reported on the feasibility of
using this method in vivo.7,8 According to those studies, a fluorescence peak was consistently observed around 460 nm
from both normal and tumorous brain tissues using 337-nm
excitation. Another study using three different excitation
wavelengths has been reported by Chung et al., and they suggested three optimal excitation wavelengths for the differentiation between normal and tumor tissues.9 In this study we
present the autofluorescence spectra of brain tumors excited
at four different wavelengths, and we endeavored to find a
single excitation wavelength for maximally effective differentiation of normal from tumorous brain tissues.
Materials and Methods
The study was designed in such a way that the autofluorescence emission from both normal and brain tumor tissue
samples were recorded in vitro, and the nature and extent of
autofluorescence was studied for various tumorous and their
corresponding normal tissues. Informed consent was given
by the patients or their guardians for testing the surgicallyremoved samples.
Freshly removed samples of glioma, astrocytoma, meningioma, pituitary adenoma, and schwannoma tissues were
collected from patients who underwent surgery at our institute. The normal tissue samples were obtained from the uninvolved portions near the resected cancerous specimens. Patient demographics and the details of the samples used for
this study are shown in Table 1.
The spectroscopic analysis was performed within 24 h after
tissue removal. The tissue specimens were cut into pieces 2–3
mm thick with a cross-sectional dimension of ⬃10 mm ⫻ 10 mm.
A Fluorolog-3 Model FL3-22 (Jobin Yvon, Edison, NJ,
USA) spectro-fluorometer was used for recording the fluorescence spectra. Spectroscopic parameters such as slit
width, optimum integration time, and spectral range were
kept constant for all samples. The excitation light from the
450 W xenon lamp was shone perpendicularly to the tissue
surface with a spot size of 2 mm ⫻ 6 mm. The emission from
the sample was collected at a 22.5° angle with respect to the
TABLE 1.
Group
PATIENT DEMOGRAPHIC CHARACTERISTICS
Type of tumor
No. and gender
of patients
excitation beam. The emission light of the selected wavelength range was then passed into the photomultiplier tube
for analysis.
The excitation wavelengths of 320, 370, 410, and 470 nm
were selected using a double-grating monochromator, and
all spectra were recorded using DataMax™ software (DataMax, Round Rock, TX, USA).
Multivariate statistical algorithm development
The primary steps involved in the development of the
multivariate statistical algorithm are preprocessing of spectral data, principal component analysis, selection of the diagnostically most useful principal component, and development of a probability-based classification scheme.
The objective of preprocessing is to calibrate the tissue
spectra for inter-patient and intra-patient variations, and is
achieved by normalization of the spectra. Normalizing a fluorescence spectrum removes absolute intensity information;
algorithms developed from normalized fluorescence spectra
rely on differences in spectral line shapes for diagnosis.
Principal component analysis (PCA) transforms the original variables of the fluorescence spectra into smaller sets
that account for most of the variance of the original data set.
It provides a novel approach of condensing all the spectral
information into a few meaningful components, with minimal information loss. PCA was performed on a normalized
data matrix, and the eigenvalues accounting for 99% of the
variance of the original preprocessed data set were retained.
The corresponding number of eigenvectors were then multiplied by the original data matrix to obtain the principal
component score matrix. Finally, the component loading of
each principal component was calculated.
Logistic discrimination was used to develop a classification based on posterior probabilities, and this was used to
calculate the probability that an unknown sample belonged
to one of the diagnostic categories identified.10–13 The observed proportion of cases in each group was taken as the
estimate of the prior probabilities. The stepwise multiple discriminant analysis was performed using SPSS statistical software (SPSS, Inc., Chicago, IL, USA).14 The sensitivity and
specificity values of the study were evaluated for each tissue type using multivariate analysis.
Results
The emission spectra recorded for the samples analyzed
in this study are shown in Figs. 1, 2, 3, 4, and 5. In all cases
AND
DETAILS
OF THE
SAMPLES USED
IN
THIS STUDY
Age range
(y)
No. of normal
tissue samples
studied
No. of diseased
tissue samples
studied
1
Glioma
9M
12
3F
15–56
14
4
2
Astrocytoma
8M
12
4F
3–43
15
9
3
Meningioma
2M
9
7F
41–62
9
5
4
Schwannoma
2M
7
5F
17–65
4
2
5
Pituitary adenoma
4M
5
1F
28–44
1
4
FLUORESCENCE SPECTROSCOPIC STUDY OF BRAIN
427
we found that the emission intensity of normal tissue was
several times higher than that of tumor tissue at all excitation wavelengths considered. Moreover, a shift in the emission’s maximum wavelength could also be seen in some
cases. In order to accommodate both the intensity values of
normal and tumor tissues in the same graph, arbitrary values of intensities are shown on the y-axis of each wavelengthintensity graph. The peak intensity of the cancerous tissue is
indicated in some cases. The results of multivariate analysis
of all the tissues studied at the various excitation wavelengths is given in Table 2.
For 410 nm excitation, the emission spectrum of normal
tissue shows an intense broad peak at 515 nm, with additional peaks at 556, 585, 630, and 690 nm. The peak at 630
nm has an intensity of 60 ⫻ 105 cps, compared to 14 ⫻ 105
cps of glioma tissues seen at 625 nm. In glioma tissues, a
noisy and weak spectrum is observed with peaks at 510, 560,
and 625 nm. A shift of about 5 nm from each peak of normal tissue is seen here. The peaks at 585 nm and 690 nm are
absent in the spectra of tumor samples.
When excited at 470 nm, the normal tissue spectrum shows
three peaks at 524, 560, and 594 nm, with the first two peaks
showing an intensity of 64 ⫻ 105 cps. The tumor spectrum
also shows three weak but distinct peaks at 517, 560, and
610 nm.
Glioma
Spectral observation. For 320 nm excitation the emission
spectrum from the normal tissue shows a broad band at the
wavelength range of 380–395 nm. A peak with maximum intensity of 13 ⫻ 106 counts per second (cps) is observed at 460
nm. After 500 nm, emission drops gradually with a slight
rise at 558 nm. At 370 nm, the glioma tissues showed a sharp
peak, while the curve was broad in the case of normal tissues. The maximum peak (8 ⫻ 105 cps) at 473 nm continued
up to 500 nm, and then emission drops with a weak rise at
560 nm.
When excited at 370 nm, the normal brain tissues show a
primary peak at 467 nm with a shoulder around 510 nm. The
maximum intensity at 467 nm was found to be 14 ⫻ 106 cps,
compared to 8 ⫻ 105 cps of glioma. Additional weak peaks
are observed at 560 nm, 590 nm, and 628 nm, whereas in the
spectrum of glioma tissue, the primary peak appears at
around 470 nm with a shoulder at 504 nm. The peak of normal tissues at 590 nm is found to be shifted to 608 nm in
glioma tissues.
Multivariate analysis. The PCA of normalized spectra at
320 nm excitation resulted in seven principal components accounting for 99% of the total variance. The first, third, and
fifth principal components demonstrated statistically most
significant differences between normal and diseased tissues
(p values are 0.014, 0.017, and 0.005, less than the required
level of significance of 0.05), and hence were considered for
the development of the classification algorithm. Similarly
PCA was also carried out for the other excitation wavelengths.
Multivariate analysis yielded 100% sensitivity and 100%
specificity for the glioma and adjacent normal tissues for all
excitation wavelengths considered.
Astrocytoma
When excited at 320 nm, the spectra from the normal tissues and astrocytoma tissues show peaks at 383 nm. This
30E ⫹ 06
Normal
Intensity (arbitrary units)
Intensity (arbitrary units)
14E ⫹ 06
8E ⫹ 05
Glioma
35E ⫹ 05
300000
350
400
450
500
550
600
25E ⫹ 06
Normal
20E ⫹ 06
15E ⫹ 06
10E ⫹ 06
50E ⫹ 05
Meningioma
10E ⫹ 05
350
400
450
Wavelength (nm)
500
550
600
550
600
Wavelength (nm)
Normal
Intensity (arbitrary units)
Intensity (arbitrary units)
13E ⫹ 06
11E ⫹ 06
67E ⫹ 05
Astrocytoma
5E ⫹ 05
350
400
450
500
Wavelength (nm)
FIG. 1.
550
600
30E ⫹ 05
Normal
20E ⫹ 05
10E ⫹ 05
2E ⫹ 05
350
6E ⫹ 05
400
Pituitary adenoma
450
500
Wavelength (nm)
Fluorescence emission spectra of tumor tissue and adjacent normal tissue when excited at 320 nm.
428
SARASWATHY ET AL.
14E ⫹ 05
15E ⫹ 06
Intensity (arbitrary units)
Intensity (arbitrary units)
Normal
8E ⫹ 05
Glioma
35E ⫹ 05
Normal
20E ⫹ 05
50E ⫹ 05
Meningioma
5E ⫹ 05
2E ⫹ 05
400
450
500
550
600
650
400
450
550
600
650
14E ⫹ 05
10E ⫹ 06
10E ⫹ 05
Intensity (arbitrary units)
Intensity (arbitrary units)
15E ⫹ 06
Normal
Astrocytoma
50E ⫹ 05
12E ⫹ 05
Normal
10E ⫹ 05
800000
600000
Pituitary adenoma
400000
200000
2E ⫹ 05
400
450
500
550
600
650
400
450
Wavelength (nm)
FIG. 2.
550
600
650
Fluorescence emission spectra of tumor tissue and adjacent normal tissue when excited at 370 nm.
14E ⫹ 05
Intensity (arbitrary units)
40E ⫹ 05
Normal
13E ⫹ 05
Glioma
6E ⫹ 05
450
500
Wavelength (nm)
60E ⫹ 05
Intensity (arbitrary units)
500
Wavelength (nm)
Wavelength (nm)
500
550
600
650
700
Normal
10E ⫹ 05
Meningioma
3E ⫹ 05
750
450
500
550
600
650
700
750
650
700
750
Wavelength (nm)
Wavelength (nm)
11E ⫹ 06
500000
Intensity (arbitrary units)
Intensity (arbitrary units)
Normal
23E ⫹ 05
3E ⫹ 05
Astrocytoma
1E ⫹ 05
400000
Normal
300000
200000
Pituitary adenoma
100000
0
450
500
550
600
Wavelength (nm)
FIG. 3.
650
700
750
450
500
550
600
Wavelength (nm)
Fluorescence emission spectra of tumor tissue and adjacent normal tissue when excited at 410 nm.
FLUORESCENCE SPECTROSCOPIC STUDY OF BRAIN
429
70E ⫹ 05
50E ⫹ 05
Intensity (arbitrary units)
Intensity (arbitrary units)
Normal
45E ⫹ 05
4E ⫹ 05
Glioma
2E ⫹ 05
500
550
600
650
700
40E ⫹ 05
750
Normal
600000
Meningioma
500
550
600
Wavelength (nm)
650
700
750
700
750
Wavelength (nm)
6E
⫹ 05
50E ⫹ 05
350000
Normal
35E ⫹ 05
Intensity (arbitrary units)
Intensity (arbitrary units)
60E ⫹ 05
Astrocytoma
Normal
300000
250000
200000
150000
Pituitary adenoma
100000
50000
500
550
600
650
700
750
500
550
600
Wavelength (nm)
FIG. 4.
Fluorescence emission spectra of tumor tissue and adjacent normal tissue when excited at 470 nm.
8E ⫹ 05
14E ⫹ 06
A
Normal
B
Normal
Intensity (arbitrary units)
Intensity (arbitrary units)
27E ⫹ 06
50E ⫹ 05
650
Wavelength (nm)
Schwannoma
8E ⫹ 05
Schwannoma
44E ⫹ 05
3.5E ⫹ 05
2E ⫹ 05
350
400
450
500
550
600
400
450
64E ⫹ 05
Intensity (arbitrary units)
Intensity (arbitrary units)
30E ⫹ 05
C
Normal
6E
⫹ 05
20E ⫹ 05
Schwannoma
2E ⫹ 05
450
500
550
600
650
Wavelength (nm)
500
550
600
650
Wavelength (nm)
Wavelength (nm)
700
750
20E ⫹ 05
25E ⫹ 05
500
D
Normal
3.5E ⫹ 05
Schwannoma
550
600
650
700
750
Wavelength (nm)
FIG. 5. Fluorescence emission spectra of schwannoma tissue and adjacent normal tissue at all the excitation wavelengths
considered: (A) 320 nm, (B) 370 nm, (C) 410 nm, and (D) 470 nm.
430
SARASWATHY ET AL.
TABLE 2.
MULTIVARIATE ANALYSIS
OF
ALL TISSUES STUDIED
AT THE
VARIOUS EXCITATION WAVELENGTHS
Excitation wavelength
320 nm
370 nm
410 nm
470 nm
Tissue
type
(%)a
(%)b
(%)a
(%)b
(%)a
(%)b
(%)a
(%)b
Normal
Glioma
Normal
Astrocytoma
Normal
Meningioma
Normal
Pituitary adenoma
Normal
Schwannoma
100
––
86.7
––
88.9
––
100
––
100
––
––
100
––
77.8
––
80
––
100
––
100
100
––
100
––
100
––
100
––
100
––
––
100
––
88.9
––
100
––
100
––
100
100
––
93.3
––
100
––
100
––
100
––
––
100
––
100
––
100
––
100
––
100
100
––
93.3
––
100
––
0
––
100
––
––
100
––
66.7
––
100
––
100
––
100
aSpecificity.
bSensitivity.
peak appears to be sharp in the case of astrocytoma, while
it is more broad in normal samples. Another peak was found
at 450 nm in both tissue types. The second peak is more intense than the one at 383 nm for normal tissues, while the
reverse is seen for the tumor tissues.
At 370 nm, the maximum fluorescence intensity of normal
brain tissue samples is at 15 ⫻ 106 cps, and that of astrocytoma tissue samples is 10 ⫻ 105 cps. An intense broad band
in the range of 450–510 nm was observed in both spectra. At
around 630 nm there is a sharp peak with a shoulder at 590
nm in normal tissues but in tumor tissue, a broad peak appears from 590–640 nm.
The mean spectra of normal tissue at 410 nm excitation
show broad peaks at 510 and 586 nm, with an additional
peak of intensity of 11 ⫻ 106 cps at 631 nm and a peak at 694
nm. In the case of diseased tissues, a weak band is observed
at 631 nm. The peak at 586 nm is shifted to 598 nm in the tumor spectrum, and appears as a shoulder compared to the
more marked peaks in the normal spectrum. The three peaks
of the normal spectrum at 586, 630, and 694 nm converge to
form an unresolved broad band in the range from 580–700
nm in the tumor tissues.
Three equally intense peaks are observed at 520, 560, and
598 nm in the spectrum of normal tissue when excited at 470
nm. Correspondingly, the spectrum of tumor tissue also
shows three distinct peaks with the third peak at 609 nm being most intense. The ratio of fluorescence intensity of normal to tumor tissue spectra is 10:1.
Multivariate analysis. The PCA of normalized data reduced
the results of 320, 370, 410, and 470 nm excitation to seven
principal components. The classification results gave specificity values ranging from 86.7–100%, and sensitivity values
from 66.7–100 %.
Meningioma
The emission spectra of normal and meningioma tissues
excited at 320 nm show the same spectral profile with noticeable variations in fluorescence intensity. Emission peaks
at 380 and 445 nm are observed in both cases. The maximum
fluorescence intensity of normal spectra, 30 ⫻ 106 cps, is
found at 380 nm, and that of diseased tissues is 45 ⫻ 105 cps
at the same wavelength.
At 370-nm excitation, the mean spectra of normal tissue
show broad bands centered at 466 nm with shoulders at 510,
560, and 595 nm. This band appears to be shifted to 459 nm
in the tumor spectra.
A primary peak of intensity of 40 ⫻ 105 cps at 508 nm is
observed in the mean spectra of normal tissues at 410-nm
excitation. This peak is shifted to 494 nm with reduced intensity of 6 ⫻ 105 cps in the diseased tissue spectra. The peak
at 590 nm, extending to 633 nm in normal tissue spectra is
replaced by a broad drop in diseased tissues.
An intense peak at 518 nm, and two peaks at 557 and 596
nm were observed for the normal tissue at 470-nm excitation. Corresponding peaks are observed in the spectra of diseased tissues with variations in intensity.
Multivariate analysis. The PCA of normalized data of 320,
370, 410, and 470 nm excitation reduced the entire data set
to five, six, six, and four principal components, respectively.
The classification results gave a sensitivity and specificity
ranging from 80.0% to 100%. All of the 9 normal samples
and 5 diseased samples were correctly classified.
Pituitary adenoma
At 320-nm excitation, peaks are seen at 382 and 444 nm.
The sharp peak that appears at 382 nm in the normal spectrum is found to be broadened in tumor tissues. The small
peaks at 518 and 562 nm of the normal spectrum are retained
without change in the tumor spectrum.
The fluorescence emission spectrum at 370-nm excitation
has a primary peak at 452 nm, with a shoulder peak at 512
nm, and smaller peaks at 556 and 595 nm. The same is seen
in the diseased spectrum, which has a slight shift in the first
peak to 447 nm.
The emission peaks for 410-nm excitation for the normal
spectrum are at 485, 558, and 595 nm. Less intense peaks
with a shift in the last peak to 620 nm is seen in diseased
tissues.
FLUORESCENCE SPECTROSCOPIC STUDY OF BRAIN
431
The emission spectrum of normal tissue shows peaks at
519, 557, and 593 nm at 470-nm excitation. The third peak is
shifted to 615 nm and is more intense than the other two
peaks in the spectrum of diseased tissues, whereas in normal tissues the first peak is more intense.
from tumor tissue, and the maximum difference was at
320 nm.
For excitation at 320 nm, the normal and the tumor tissues
consistently showed two peaks near 380 and 450 nm. In the
spectra of the normal tissue adjacent to pituitary adenomas,
meningiomas, and schwannomas, the intensity of the peak
near 380 nm was significantly higher than that near 450 nm,
while it was reversed for the tissues adjacent to gliomas. In
the normal tissue adjacent to astrocytomas, both peaks had
the same intensity. The spectrum of meningioma tissue alone
retained had a spectral profile similar to that of normal tissue. For gliomas, a considerable change could be seen, with
the occurrence of three distinct peaks around 380, 470, and
560 nm, with the center one being most intense and broad.
The spectra of astrocytomas, pituitary adenomas, and
schwannomas showed general broadness of the band and
stronger intensity of the peak around 450 nm compared to
the one at 370 nm. Tryptophan, NADH, and collagen are believed to be responsible for these emission bands.9
At 370-nm excitation, except for the general decrease in
intensity and broadening of the peaks, no major changes
were seen for the tissues from pituitary adenomas, schwannomas, and meningiomas compared to their corresponding
normal tissues. The spectra of astrocytomas and gliomas
showed changes in the spectral profile in addition to a decrease in intensity. Fluorophores contributing to the peaks
around 460, 560, 630, and 690 nm are believed to be collagen, flavin adenine dinucleotide (FAD), and porphyrins.15
Similar findings were observed at 410 nm. The spectra of
schwannomas, pituitary adenomas, and meningiomas
showed only reductions in intensity. In the case of gliomas
and astrocytomas, the peak around 690 nm of the normal
spectrum is absent from the spectrum of diseased tissue. In
the case of astrocytomas, an additional shoulder is observed
at 598 nm. The fluorescence from the FAD and porphyrin
components of the tissue are responsible for these peaks.14
The most interesting and useful results for discriminating
tumor tissues from normal tissues were obtained from the
excitation wavelength of 470 nm. In majority of the tumor
types considered (meningioma, glioma, and astrocytoma),
the spectral profile was significantly different from that of
normal spectra. In these cases the peak around 600 nm was
found to be more intense than the other two peaks in the tumor spectra, which is a major difference from normal spectra. The spectral profile of schwannoma did not show a
marked difference from that of normal tissue. No changes
could be seen between the mean spectrum of pituitary adenoma tissue and the corresponding normal tissue. Here the
normal and diseased spectra coincided in the wavelength region from 650–750 nm. The intensity variation in this region
was also minimal compared to that of the other tumor types
tested. Comparison of the statistical results indicated that the
tissue samples that we considered to be normal were not,
and hence they were incorrectly classified as diseased tissue.
Here, the negative results of the spectral findings is well
supported by the statistical results. In all other cases, the statistical analyses were in agreement with the spectral observations, with specificity ranging from 86.7–100%, and sensitivity ranging from 66.7–100%.
From the results of this study it is clear that autofluorescence spectroscopy can be effectively used as a tool for the
diagnosis of various types of brain tumors. Of the different
Multivariate analysis. The PCA of normalized data of 320,
370, and 410 nm reduced the entire data set to three, four,
and four principal components, respectively. The classification results of performing discriminant analysis on the output of PCA for all excitation wavelengths yielded a sensitivity and specificity of 100% for normal tissues and 100%
for pituitary adenoma tissues.
The classification results of performing discriminant analysis on the output of PCA (using two principal components)
at 470 nm was 100% for pituitary adenoma tissues, but the
normal sample was statistically classified as diseased. Of the
4 diseased cases, all were correctly classified.
Schwannoma
The emission spectra at 320 nm excitation shows a primary
peak at 383 nm and secondary peak at 444 nm in the normal
spectrum, and in diseased tissues the primary intense peak
was at 394 nm and a secondary one was at 450 nm. In diseased tissues, the second peak was found to be more intense
than first.
In the normal emission spectrum at 370-nm excitation a
broad peak appears at 469 nm, with a shoulder at 507 nm
and smaller peaks at 560 and 595 nm. In the case of tumor
tissues, the emission peaks are found at 448 nm, with smaller
peaks at the same regions as those of the normal tissues. The
intensity of the spectrum of normal tissue was found to be
more than 17 times that of tumor tissue.
The fluorescence peaks in the normal tissue spectrum at
410 nm excitation are seen at 507 and 591 nm, with smaller
peaks at 559 and 648 nm. In the tumor tissues the first broad
peak is shifted to 514 nm, with smaller peaks at 558 and the
591 nm.
The emission spectra of normal brain tissues at 470-nm excitation show peaks at 522, 556, and 590 nm. The spectrum
of tumor tissues showed similar peaks at these same wavelengths, though they were of lower intensity.
Multivariate analysis. The PCA of the normalized data at
320 and 410 nm reduced the entire data set to four principal
components, and on normalized data at 370 and 470 nm to
five principal components. The classification results of performing discriminant analysis on the output of PCA yielded
100% specificity and sensitivity for normal tissues, as well as
100% specificity and sensitivity for schwannoma tissues
(both of the 2 diseased cases and all 4 normal cases were correctly classified).
Discussion
From the autofluorescence spectra of the five types of
tumor tissues considered, it is well demonstrated that the
spectra of tumor tissue can easily be distinguished from
the spectra of the corresponding adjacent normal tissue.
This result was true for all four excitation wavelengths
considered. In all cases, the autofluorescence from the normal tissue was found to be several times higher than that
432
excitation wavelengths considered, 470 nm appears to be the
optimum wavelength for analysis of tissue fluorescence of
brain tumor tissues. Even though all wavelengths considered
yielded good spectral discrimination when the intensity parameters are considered, the spectral signature at 470-nm excitation was found to be more prominent, with the appearance of an intense porphyrin peak at around 600 nm in the
tumor tissue.
Chung et al. performed laser-induced autofluorescence assessment of the brain to distinguish brain tumors from normal tissues, and the results are in good agreement with our
results.9 Laura et al. reported lower NADH fluorescence intensity in brain tumor tissue than in normal tissue.16 Bottiroli
et al. also reported significant differences in autofluorescence
emission properties between normal and brain tumor tissues, in terms of both spectral shape and signal amplitude.17
Advanced studies in this area reveal that even the concentration of components differs from one type of tumor to another.8,9,17–21
Most screening and diagnostic algorithms developed from
fluorescence spectroscopy of tissues incorporate qualitatively or statistically selected variables, which are evaluated
using a binary or probability-based classification scheme. Ramanujam et al. developed an algorithm that uses statistically
selected spectral variables and probability-based classification for cervical pre-cancer detection.22 This multivariate statistical algorithm employs PCA to reduce pre-processed tissue fluorescence emission spectra into orthogonal principal
components. Probability-based classification is then developed using the diagnostically relevant principal components. By using statistical analysis of tissue fluorescence the
entire spectral data content is utilized.
Ramanujam et al. developed an algorithm using biochemical and/or morphological features related to the tissue
fluorescence spectra, coupled with a probability-based classification, to discriminate between normal coronary arteries
and non-calcified and calcified atherosclerotic plaques in
vitro.23 The advantage of this physically-based model is that
it provides insight into the biochemical and morphological
features of diseased and normal tissues.24 But the development of such an algorithm has been hampered by the fact
that fluorescence spectroscopy of human tissue is greatly affected by the absorption and scattering of the excitation light
and the emitted light, making the interpretation of the measured spectral data challenging.
Conclusion
In conclusion, spectroscopic luminescence measurements
carried out in the present study revealed significant differences between tumor and adjacent normal tissues of the human brain for all the tumor types tested, except for pituitary
adenoma. For pituitary adenoma, the spectral differences between the tumor and normal tissues were minimal, and the
two spectra even coincided at wavelengths. This is reinforced
by the statistical findings of zero specificity at the excitation
wavelength of 470 nm. In all other cases the spectral findings were well supported by the multivariate analysis for all
the tissue types and excitation wavelengths considered.
From this study we conclude that excitation wavelengths
ranging from 410–470 nm are more suitable for the discrimination of brain tumor tissue than those below 350 nm, as
SARASWATHY ET AL.
has been proposed in earlier works. Moreover, in our study
we observed that the results of 470-nm excitation yielded results that incorrectly indicated that the sample was normal
tissue. This was not seen at the other excitation wavelengths
tested.
Acknowledgements
The financial support received from Board of Research in
Nuclear Sciences (BRNS), Department of Atomic Energy
(DAE), Government of India is sincerely acknowledged. We
also acknowledge Prof. K. Mohandas, Director, of the Sree
Chitra Tirunal Institute for Medical Sciences and Technology (SCTIMST) for providing the facilities needed to complete this work, and Dr. K.N. Bhattacharya, Professor and
Head, Department of Neurosurgery, for providing the samples used in this study.
Disclosure Statement
No competing financial interests exist.
References
1. Lakowicz, J.R. (1983). Principles of Fluorescence Spectroscopy,
Vol. 1. New York: Plenum Press.
2. Dinh, T.V. (2003). Biomedical Photonics Handbook, Vol. 1. New
York: CRC Press Inc. York.
3. Kantelhardt, S.R., Leppert, J., Krajewski, J., et al. (2007).
Imaging of brain and brain tumor specimens by time-resolved multiphoton excitation microscopy ex vivo. Neurooncol. 9, 103–112.
4. Ronald, W.W. (2002). Lasers in Medicine, Vol. 1. Boca Raton,
FL: CRC Press Inc.
5. Yang, V.X.D., Muller, P.J., Herman, P., and Wilson, B.C.
(2003). A multispectral fluorescence imaging system: Design
and initial clinical tests in intra-operative Photofrin-photodynamic therapy of brain tumours. Lasers Surg. Med. 32,
224–232.
6. Gaigalas, A.K., Li, L., Henderson, O., Vogt, R., and Barr, J.
(2001). The development of fluorescence intensity standards.
J. Res. Natl. Inst. Stand. Technol. 106, 381–389.
7. Lin, W.C., Toms, S.A., Motamedi, M., Jansen, E.D., and
Jansen, A. (2000). Brain tumor demarcation using optical
spectroscopy: an in vitro study. J. Biomed. Opt. 5, 214–220.
8. Lin, W.C., Toms, S.A., Johnson, M., Jansen, E.D., and Jansen,
A.D. (2001). In vivo brain tumor demarcation using optical
spectroscopy. Photochem. Photobiol. 73, 396–402.
9. Chung, Y.G., Schwartz, J.A., Gardner, C.M., Sawaya, R.E.,
and Jacques, S.L. (1997). Diagnostic potential of laser induced auto fluorescence in brain tissue. J. Korean Med. Sci.
121, 135–142.
10. De Coursey, W.J. (2003). Statistics and Probability for Engineering Applications, Vol. 1. Woburn, MA: Elsevier Science.
11. Box, G.E.P., Leonard, T., Chien-Fu, W. (1983). Scientific Inference, Data Analysis and Robustness, Vol. 1. New York: Academic Press.
12. Brown, B.W., and Hollander, M. (1977). Statistics: A Biomedical Introduction, Vol. 1. New York: John Wiley & Sons.
13. Ross, S.M. (1987). Introduction to Probability and Statistics for
Engineers and Scientists, Vol. 1. New York: John Wiley & Sons.
14. Marques, J.P. (2003). Applied Statistics Using SPSS, Statistica,
& Matlab, Vol. 1. New York: Springer.
15. Zuluaga, A.F., Utzinger, U.R.S., Durkin, A., Fuchs, H., Gillenwater, A., Jacob, R., Kemp, B., Fan, J., Richards-Kortum, R.
FLUORESCENCE SPECTROSCOPIC STUDY OF BRAIN
433
(1999). Fluorescensce excitation emission matrices of human
tissue: A system for in vitro measurement and method of
data analysis. Appl. Spectrosc. 53, 302–311.
Laura, M., Javier, A.J., Pramod, V.B., William, H.Y., Brian,
K.P., Keith, L.B., and Reid, C.T. (2004). Fluorescence life time
spectroscopy of glioblastoma multiforme. Photochem. Photobiol. 75, 1–10.
Bottiroli, G., Croce, A.C., Locatelli, D., Nano, R.,
Giombelli, E., Messina, A., and Benericetti, E. (1998). Brain
tissue autofluorescence: an aid for intraoperative delineation of tumour resection margins. Cancer Detect. Prev.
2, 330–339.
Madhuri, S., Venkadesan, N., Aruna, P., Kodeeswaran, D.,
Venkatesan, P., and Ganesan, S. (2003). Native fluorescence
spectroscopy of blood plasma in the characterization of oral
malignancy, Photochem. Photobiol. 78, 197–204.
Cubillos, S., Obregon, F., Vargas, M.F., Salazar, L.A.,
and Lima, L. (2006). Taurine concentration in human
gliomas and meningiomas: tumoural, peritumoural,
and extratumoural tissue. Adv. Exp. Med. Biol. 583,
419–422.
Miller, C.R., Dunham, C.P., Scheithauer, B.W., and Perry, A.
(2006). Significance of necrosis in grading of oligodendroglial neoplasms: a clinicopathologic and genetic study of
newly diagnosed high-grade gliomas. J. Clin. Oncol. 24,
5419–5426.
21. Kortum, R.R., and Muraca, E.S. (1996). Quantitative optical
spectroscopy for tissue diagnosis. Ann. Rev. Phys. Chem. 47,
556–606.
22. Ramanujam, N., Mitchell, M.F., Mahadevan, A., et al. (1996).
Development of multivariate statistical algorithms to
analyse human cervical tissue-fluorescence spectra acquired
in vivo. Lasers Surg. Med. 19, 46–82.
23. Ramanujam, N., and Kortum, R.R. (1996). Cervical precancer detection using a multivariate statistical algorithm based
on laser induced fluorescence spectra at multiple excitation
wavelengths. Photochem. Photobiol. 64, 720–735.
24. Ramanujam, N. (2000). Fluorescence spectroscopy in vivo, in:
Encyclopedia of Analytical Chemistry. R.A. Mayers, (ed.).
Chichester: John Wiley & Sons, pp. 20–56.
16.
17.
18.
19.
20.
Address reprint requests to:
Dr. R.S. Jayasree, Ph.D.
Scientist
Department of Imaging Sciences and Interventional Radiology
Sree Chitra Tirunal Institute for Medical Sciences and
Technology
Medical College P.O.
Trivandrum, 695011, Kerala, India
E-mail: [email protected]