Anda di halaman 1dari 20

Intraoperative Raman

S p e c t ro s c o p y
Michelle Brusatori, PhDa,b,c, Gregory Auner, PhDa,b,c,d, Thomas Noh, MDe,f,
Lisa Scarpace, MSe, Brandy Broadbent, MSa,b,c, Steven N. Kalkanis, MDe,f,*

KEYWORDS
 Diagnostics  Raman spectroscopy  In vivo  Molecular signature  Intraoperative

KEY POINTS
 Surgical excision of brain tumors provides a means of cytoreduction and diagnosis while minimizing
neurologic deficit and improving overall survival.
 Raman spectroscopy is a promising investigative and diagnostic tool for neurosurgery, and pro-
vides rapid, nondestructive molecular characterization in vivo or in vitro for biopsy, margin assess-
ment, or laboratory uses.
 A review is presented of Raman spectroscopy techniques for clinical practice, particularly in distin-
guishing between normal and diseased tissue without the assistance of a pathologist, or to
augment pathologic findings.

INTRODUCTION The estimated incidence of new primary brain


and central nervous system (CNS) tumors for
Raman spectroscopy has been identified as a 2016 is 77,670 with 16,616 predicted deaths
promising investigative and diagnostic tool in resulting from malignant tumors.1 With advances
neurosurgery because it provides rapid, non- in intraoperative imaging, surgical excision is
destructive molecular characterization in vivo or becoming more relevant for brain tumors,
in vitro for biopsy, margin assessment, or la- providing a means of cytoreduction and diag-
boratory use. Spectroscopic information re- nosis while minimizing neurologic deficit and
sulting from Raman active functional groups of improving overall survival. Even in primary malig-
nucleic acids, proteins, lipids, and carbohydrates nant tumors such as high-grade gliomas, for
in tissue allows evaluation and characterization. which prognosis is poor, the extent of resection
This article highlights the need for such an is associated with improved mortality and out-
instrument, summarizes neurosurgical Raman comes.2 Glioblastoma multiforme (GBM), an
work to date, and discusses what the future of extremely aggressive primary brain tumor with
neurosurgical Raman spectroscopy might look an average life expectancy of approximately 12
like. to 18 months, accounts for 15.4% of all CNS

Disclosure: The authors have no relationship to any commercial company with direct financial interest in the
subject material discussed in this article.
a
Department of Surgery, Wayne State University, 5050 Anthony Wayne Drive, Detroit, MI 48202, USA;
b
Department of Biomedical Engineering, Wayne State University, 5050 Anthony Wayne Drive, Detroit, MI
neurosurgery.theclinics.com

48202, USA; c Department of Smart Sensors and Integrated Microsystems, Wayne State University, 5050 An-
thony Wayne Drive, Detroit, MI 48202, USA; d The Detroit Institute of Ophthalmology, Henry Ford Health Sys-
tem, 15415 E. Jefferson Avenue, Grosse Pointe Park, MI 48230, USA; e Department of Neurosurgery, Hermelin
Brain Tumor Center, Henry Ford Health System, 2799 West Grand Boulevard, Detroit, MI 48202, USA;
f
Josephine Ford Cancer Center, Henry Ford Health System, 2799 West Grand Boulevard, Detroit, MI
48202, USA
* Corresponding author. Department of Neurosurgery, Henry Ford Health System, 2799 West Grand Boule-
vard, Detroit, MI 48202.
E-mail address: skalkan1@hfhs.org

Neurosurg Clin N Am 28 (2017) 633–652


http://dx.doi.org/10.1016/j.nec.2017.05.014
1042-3680/17/Ó 2017 Elsevier Inc. All rights reserved.
634 Brusatori et al

and primary brain tumors. A recent study blood. Also, there is no clear-cut interpretation
reviewed tumor volumetrics in 500 newly diag- of the data in terms of individual components.
nosed patients with supratentorial GBM and Bridging the gap between maintaining neuro-
found a statistically significant difference in sur- logic function and comprehensive tumor cell
vival with subtotal resections as low as 78% for removal is critical. Direct histologic evaluation
survival benefit.3 This work, among growing liter- through techniques such as mass spectrometry
ature, shows that early maximal resection while is gaining momentum but many techniques are
preserving a patient’s functional status is critical severely limited by their time-based nature.8
for optimal patient outcome,4,5 and this suggests Raman spectroscopy has potential to be an impor-
a need for new technologies designed to achieve tant modality for intraoperative evaluation of tissue
maximal safe resections. during surgical resection. It is a reagentless,
Many tools have been developed to aid ne- nondestructive optical technique with demon-
urosurgeons with intraoperative tumor identifica- strated ability to detect changes in the chemical
tion and delineation. Image-guided surgical composition and/or molecular structure between
techniques include intraoperative MRI, fluores- diseased and healthy tissue. Although most
cence-guided surgery, neuronavigation, and Raman spectroscopy studies are exploratory,
ultrasonography. As an example, fluorescence- in vitro experimental studies have shown that this
based imaging methodologies that rely on endog- technique can be used to differentiate normal
enous autofluorescence are briefly highlighted. brain tissue from tissue with infiltrating cancer cells
Autofluorescence is a well-known phenomenon and dense cancerous masses with high speci-
in which a biological substrate is excited with a ficity. Further, this technique has potential to be
suitable wavelength of light. The emission of light used in conjunction with imaging modalities such
in the ultraviolet (UV), visible and near-infrared as MRI and ultrasonography for more in-depth
(NIR) range results from the presence of fluoro- intraoperative characterization of residual tumor.
phores that naturally occur in tissues such as pro- This article provides contemporary Raman spec-
teins (elastin and collagen), flavins, vitamin A, and troscopy data and further understanding of the
fatty acids. However, steady state measurements Raman effect on biological tissues.
that rely solely on fluorescence emission intensity
are affected by numerous factors. These factors RAMAN SPECTROSCOPY
include irregular tissue surfaces, nonuniform sam-
ple illumination, and the presence of endogenous Interest in clinical spectroscopy is increasing
absorbers in the biological substrate. Fluores- because of the potential of vibrational spectro-
cence lifetime imaging microscopy (FLIM) re- scopic techniques for noninvasive tissue diag-
solves some of the difficulties in quantifying nosis. Raman spectroscopy and infrared (IR)
fluorescence intensities. FLIM produces spatially spectroscopy probe molecular vibrations associ-
resolved images of fluorophore lifetime. The tech- ated with chemical bonds in a sample to obtain in-
nique involves exciting the sample with repetitive formation on molecular structure, composition,
light pulses and observing the fluorescence and intermolecular interactions.
response. Sun and colleagues6 performed a pilot The Raman effect was discovered in 1928 by
study using an endoscopic fluorescence lifetime C.V. Raman when he observed that light traveling
imaging for intraoperative identification of GBM through various liquids scatters differently, in a
and showed the contrast between normal and tu- behavior distinct from fluorescence.9 The basic
mor tissue. Time-resolved fluorescence spectros- process of scattering is the absorption of and
copy (TR-LIFS) is also used to extract fluorophore reemission of electromagnetic radiation by mole-
lifetime but, instead of image acquisition, the cules. The mechanism associated with scattering
method records the time-resolved spectrum at a is induced electric dipole radiation. Light incident
single point. Butte and colleagues7 used on matter can scatter at the same optical fre-
TR-LIFS to detect the fluorescence lifetime differ- quency as the incident radiation (ie, the induced
ences between normal brain tissue and glioma dipole moment in a molecule oscillates and radi-
in vivo.7 Both time-resolved fluorescence spec- ates at the same frequency as the incident light).
troscopy and imaging techniques have complex This process is termed elastic scattering (Ray-
instrumental setups making the translation to a leigh scattering). Light can also scatter at fre-
clinical setting difficult. In addition, because of quencies that differ from the original radiation.
the presence of multiple fluorescence molecules, This process is called inelastic scattering and,
the measured signal from the tissue is a distribu- unlike the elastic process, it involves an energy
tion of fluorescence decays, making the methods transfer between the incident photons and a ma-
susceptible to background interference such as terial. An inelastic process, termed the Raman
Intraoperative Raman Spectroscopy 635

effect, occurs in approximately 1 in 107 photon coincides with an electronic transition of the mole-
interactions with matter10 and results from mod- cule a resonance effect is observed and the inten-
ulations in the induced dipole moment within sity of some Raman active vibrations can be
molecules. Changes in polarizability with molec- increased by a factor of 102 to 106.
ular vibration cause the induced dipole moment There are 4 primary components in a conven-
to oscillate at frequencies other than the incident tional Raman spectroscopy system: (1) the light
light.11 Radiation emitted at longer wavelengths source (usually a laser), (2) the spectrometer, (3)
(downshifted frequencies) than the incident light a filter to block the laser line and allow Raman
is termed Stokes scattering, which occurs when scatter to pass, and (4) a detector.10 Fig. 2 shows
a molecule is promoted from the ground state a basic Raman spectrometer. The recorded in-
to a virtual energy level (an intermediate electron elastic scattering results in a Raman spectrum or
state with energy lower than a real electronic fingerprint, which can be used in imaging and di-
transition) then relaxes to a vibrational state. Ra- agnostics. The x-axis corresponds with the
diation emitted at shorter wavelengths (upshifted change in energy of the scattered photon; the y-
frequencies) than the original radiation is known axis corresponds with the photon count. Energy
as anti-Stokes scattering. This process occurs shifts correlate to types of bonds, such as C5H,
when a molecule is initially in a vibrational state C-C, or CH2 twist, based on quantized energy
and relaxes to the ground state after scattering. levels of each molecular bond. The amount, loca-
With conventional Raman spectroscopy, the ef- tion, and intensity of peaks vary based on the num-
fect is independent of wavelength because no real ber of vibrational bonds in a molecule and
energy states are involved (only virtual states), and their ensemble interactions with each other.12
this is termed nonresonance Raman. However, Biological spectra can be complex. For example,
certain substances when exposed to electromag- L-phenylalanine (C6H5CH2CHCOOH), an important
netic radiation can produce a strong fluorescence tumor marker, shows 63 peaks in the Raman
signal that overlaps the Raman signal. Raman fingerprint (Fig. 3).13
scattering and fluorescence are competing phe- When using Raman with large molecules, the
nomena that have a similar origin. The Raman ef- group contribution approach is used whereby
fect corresponds with the absorption and the largest peaks come from the bonds that
subsequent emission of a photon via an intermedi- occur most frequently.12 The distinctive peak for
1
ate electron state, as shown in an energy level di- L-phenylalanine between 1000 and 1008 cm re-
agram (Fig. 1). Molecules are excited to a virtual sults from C-C bonds and ring breathing modes of
energy level for a short period, on the order of pi- benzene, which are the most prevalent bond vibra-
coseconds, before an emitted photon results; tions.14 Some molecular peaks may not be
whereas in fluorescence the incident light is apparent in the spectrum depending on bond
absorbed by a molecule and reemitted from elec- strength and shape.15 If the bond does not change
tronically excited states after a resonance time on its polarizability, it is not easily detectable. For
the order of nanoseconds. example, water has a weak Raman signal because
In contrast, resonance Raman spectroscopy, a of its symmetric polarity. In tissue diagnostics, this
variant of conventional Raman, measures molecu- can be advantageous. In contrast, infrared spec-
lar vibrations in a wavelength-dependent manner. troscopy signals can be overwhelmed by water
When the wavelength of the exciting source emissions.
The Raman spectrum of tissue induces many
peaks representative of the plethora of cellular
constituents (see Fig. 3). Distinct differences be-
tween pathologic conditions are shown as
different intensities of the same peak or peak
shifting based on binding conformation. Occa-
sionally, there is a unique peak from a different
moiety, such as calcifications or hemorrhage.
Biological tissues are largely made up of the
same molecules, but the quantity and interac-
tions of the molecules vary between tissue types.
For example, tumors use glycolysis to generate
ATP, which produces more pyruvate, which con-
verts to lactate and subsequently to the amino
acid alanine, resulting in stronger protein
Fig. 1. Energy transition of a molecule. peaks.16,17
636 Brusatori et al

Fig. 2. Raman spectrometer. A narrow line–width laser source at a particular wavelength, and optimized to the
material under investigation, is focused on the sample. Scattered light is collected by the microscope lens and
passes through a notch or edge filter, which eliminates light of the original wavelength, leaving only inelastically
scattered light. A grating is used to separate the different wavelengths of scattered light, such that the different
wavelengths fall on different areas of a charged coupled device (CCD) or other detector, such as a photomulti-
plier tube.

EARLY APPLICATIONS OF NEUROLOGIC Similarly, the neurinoma and neurocytoma were


RAMAN SPECTROSCOPY characterized by specific spectral bands, corre-
sponding with carotenoids and hydroxyapatite,
In 1990, Raman spectroscopy was first applied to respectively.20
characterize parenchymal water content in situ.18 Koljenovic and colleagues21 created Raman im-
Normal and edematous brain tissue showed ages of sectioned GBM from 20 patients and
similar peak intensities attributed to protein CH compared them with hematoxylin and eosin
bonds. Although water produces a weak Raman (H&E) staining. They showed an ability to distin-
signal, when coupled with the OH water peak, guish vital tumor from necrotic tissue given the
the OH/CH ratio gives a relative comparison of wa- increased presence of cholesterol and its esters,
ter content.
Later, Mizuno and colleagues used Raman
spectroscopy to differentiate gray from white mat-
ter in a rat model by examining the spectral ratio of
protein to lipid content. In this case, gray matter
has an abundance of proteins, whereas white mat-
ter has increased phospholipid levels because of
its myelination. Numerous studies have contrib-
uted a wealth of information to the molecular fin-
gerprints of biological tissues, and several
Raman peak assignment tables are available.14,19
Table 1 provides a summary of peak information
for brain diagnostics.
Follow-up work showed differentiating spectral
characteristics between excised grades II and III
glioma, neurinoma, and neurocytoma of choroid
Fig. 3. Raman spectra typical of white matter and tu-
plexus.20 Although the spectrum for the grade II mor necrosis show features similar to purified mea-
glioma was similar to that of normal cortical tissue, surements of their known molecular constituents
grade III glioma had unique molecular structural (methyl-docosahexaenoate [methylDHA] and phenyl-
changes, namely a band at 1245 cm 1, an intensi- alanine, respectively). Spectra are vertically offset
fied 1130 cm 1 band, and a band at 856 cm 1. and scaled for visualization purposes only.
Intraoperative Raman Spectroscopy 637

Table 1
Raman peak assignments from neurosurgical literature

Raman
Peak (cmL1) Assignment Citation
58–61
427–430 Cholesterol/cholesterol ester
59
Higher in white matter than gray matter
61
Cholesterol deposits found in tumor sections
61
429 Bending PO4 in hydroxyapatite, deposits found in tumor tissues
58
457 Melanin, found in metastatic melanoma
62
467–475 Glycogen and polysaccharides seen in PA and MG with SERS
63,64
482 Glycogen, seen in metastatic renal cell carcinoma
60
491 Cholesterol
60,61
498–500 Nucleic acids/nucleotides
65
525 S-S stretch mode of gauche-gauche-trans form, higher in
synaptosomal fraction than myelin
61
538 Cholesterol ester, cholesterol deposits found in tumor sections
58–60
544–548 Cholesterol
59
Higher in white matter than gray matter
65
553 S-S trans-gauche-trans mode, higher in synaptosomal fraction than
myelin
62
558–566 Tryptophan, higher in GBM, ODG tumor when measured with SERS
61
583 Bending of PO4 in hydroxyapatite, deposits found in tumor tissues
58,66
597–599 Melanin, seen in metastatic melanoma
64,67
607–608 607: glycerol
39,58–60,68
608: cholesterol
59
Higher in white matter than gray matter
68
Cholesterol deposits found in tumor sections
39,61
613–622 Cholesterol ester
61
Cholesterol deposits found in tumor sections
60,62–64,68
622/624: phenylalanine (protein) associated with necrosis
62
Present in SERS, higher in ODG tumor
63,64,68
642–645 Tyr (protein) associated with necrosis
62
648–655 C-C and C-S stretch of protein (SERS)
Ratio of 724:655 is significantly different between tumor/normal 62
61,63,64
661–669 666: Heme associated with hemorrhage
61,63,68
667, 669: nucleic acid/nucleotide associated with necrosis, tumor
60,61
680–683 Nucleic acids
37,65
700, 703/704 C-N stretch mode from choline head in phosphatidylcholine and
sphingomyelin
59
720:701 ratio increases in GBM
23,58–61,63,64,68–71
Cholesterol
63
Variable in normal tissue
23,59
Higher in white matter than gray matter
65
Distinguishes myelin fraction from synaptosomal
61,68
Deposits found in tumor sections
67
Cholesterol content decreases from corpus callosum to cortex
Decreased cholesterol/phospholipid content in tumor compared 67,71
with normal
65
C-S stretch from methionine, cysteine, and cysteine residues of
proteins (less likely)
716–719 Choline in head group of sphingomyelin and phosphatidylcholine, 23,58–61,63,64,67–69,71
phosphatidylethanolamine
63,64,67,71
Higher in normal than tumor
59
720:701 ratio increases in GBM
61
Stronger in GBM than MG
(continued on next page)
638 Brusatori et al

Table 1
(continued )

Raman
Peak (cmL1) Assignment Citation
722–728 DNA, RNA base ring breathing mode, C-S of proteins, CH2 rocking 60–62,68,72
of adenine, NADH, FADH (SERS), nucleotides
62
Ratio of 724:655 signifies difference between tumor/normal
72
Seen in cell nuclei
67
Increased DNA/RNA in tumor with respect to normal
58
727 Melanin, seen in metastatic melanoma
23,61,63,64,66
739–759 Heme (blood)
65
746 Phosphorus-oxygen-phosphorus bonds symmetric stretch of
phospholipids (predominant in white matter and myelin
fraction)
61,68
749–751 Nucleotides
60,63,64,71
757–760 Tryptophan
63,64
Associated with necrosis
60–63,68,72,73
781–795 O-P-O stretch of DNA/nucleic acids, seen in cell nuclei
63,73
Associated with tumor
Nucleic acid band with the least overlap with lipid, cholesterol, 68
and protein, making it a strong marker for mitosis index
69
800 Quartz substrate (removed via processing)
63,64,68,71
826–830 Tyr, proline
61
830 Phosphate backbone of b-DNA conformation
63,64,67
Higher in tumor, necrosis than in normal
60
845 Cholesterol
63,64,68
851–852 Tyr, associated with necrosis
850–855 Glycogen, seen in metastatic renal cell carcinoma and high-grade 63,73
tumors
20
856 Polysaccharides
22,61
Collagen
20
Strong in glioma
22
Seen in dura mater but not meningioma
73
865 Phosphatidylethanolamine, lower in tumor than normal
64
867 Glycogen, seen in metastatic renal cell carcinoma
68
873 Phosphatidylcholine
31
875–878 Tyr, higher in astrocytoma cells than astrocyte cells
59,60
Choline group in PC and sphingomyelin
62
905–908 Glucose level, increased in PA, as measured by SERS
67,74
925–926 C-C in peptide backbone
60
Cholesterol
22,61,68,71
935–939 C-C of peptide backbone, collagen
22
Seen in dura mater but not meningioma
63
941 Glycogen, seen in metastatic renal cell carcinoma
956 Carotenoid (present in schwannoma, not present in normal tissue) 20
35,62
C-C vibration of protein, carotenoids, hydroxyapatite, collagen
62
Increased in lung metastases tumor, GBM, ODG, PA, EP, MG as
measured by SERS
958–960 Stretching vibrations of PO4 in hydroxyapatite, deposits found in 20,22,61,66,75
neurocytoma, metastatic melanoma, psammoma body of
meningioma
75
Stronger in necrosis than tumor
74
961 Higher cholesterol content in white matter than gray matter
(continued on next page)
Intraoperative Raman Spectroscopy 639

Table 1
(continued )

Raman
Peak (cmL1) Assignment Citation
32,61
969–977 Tricalcium phosphate Ca3(PO4) calcification seen in schwannoma,
necrosis
58,66
976 Melanin, seen in metastatic melanoma
33
985 Calcification
60
990 Phosphate ions (in PBS)
20,23,31–33,35,58,60–69,
1002–1006 Ring breathing mode of phenylalanine of protein, heme
71,73,74,76–78
23,31,32,62–64,74
Higher in lung met tumor, GBM, EP (SERS), astrocytoma, necrosis
than in normal
32
Stronger in necrosis than tumor
73
Increased in high-grade tumors
79
Slightly increased after radiation exposure
78
Weaker after brain injury
20
Carotenoid (present in schwannoma, not present in normal tissue)
61
Heme/hemorrhage
32,33,35,60,63,64,68,78
1032 C-H of phenylalanine
32,63,64
Increased in necrosis
78
Weaker after tissue fixation
35
Higher in glioma tissue and cells than normal tissue and astrocyte
cells
62
1042–1049 C-O and C-N stretch of protein, higher in lung met tumor, MG
measured by SERS
69
1050 Quartz substrate (removed via processing)
20,23,32,33,58–61,63–65,
1060–1066 C-O stretch and C-O-C symmetric stretch, C-C stretch of
67–69,72–75,77
phospholipids (side chains specifically) and cholesterol
20,23,65,72,74,79
Strongest in white matter and myelin fraction
32,33,63,64,67,73,74,79
Strong in normal, weaker in tumor/necrosis
77
Higher in Alzheimer hippocampus than normal hippocampus
75
Stronger in necrosis than tumor
61
Cholesterol deposits found in tumor sections
61
1071 Stretching vibration of PO4 in hydroxyapatite, deposits found in
tumors
1080–1090 C-C stretch and PO2 symmetric stretching of phospholipids and 20,32,35,60,63,65,72,
74,75,78
nucleic acids
72
C5O vibration of ester/aldehyde
20,65,72
Higher in gray matter than white matter
32
Higher in white matter than gray matter and tumor
63
Seen in metastatic tumors
72
Seen in cell nuclei
75
Stronger in necrosis than tumor
74
Stronger in gray matter than tumor
62
1088–1097 C-N stretch of protein, higher in GBM, ODG, PA, EP, MG when
measured by SERS
59,64,73,77
1088 Lipid/glycogen
64,73
Seen in metastatic renal cell carcinoma, high-grade tumors
77
Higher in Alzheimer hippocampus than normal hippocampus
33,35,60,61,68
1094–1100 Phosphate backbone of b-DNA conformation
68
Marker for mitotic figure
Higher in glioma tissue and cells than normal tissue and astrocyte 35
cells
63
1122 Glycogen, seen in metastatic renal cell carcinoma
(continued on next page)
640 Brusatori et al

Table 1
(continued )

Raman
Peak (cmL1) Assignment Citation
58
1124 Melanin, seen in metastatic melanoma
61
Heme/hemorrhage
20,32,33,35,58,60,61,64,
1126–1133 C-C stretch of phospholipids (side chains), cholesterol
65,67,69,72,74,75,78
35,60,77
C-C and C-N stretch of protein/glucose
59
CH2 vibration
20,65,72,74
Stronger in white matter and myelin fraction
20,74
Stronger in glioma, suggesting higher concentration of trans
conformation hydrocarbon chains in tumor lipids
10,33
Stronger cholesterol/phospholipid content in normal than tumor
77
Stronger in Alzheimer hippocampus than normal hippocampus
75
Stronger in necrosis than tumor
61
Cholesterol deposits found in tumor sections
37
1142 Lipid
20,32,33,59,60,75
1157–1159 Carotenoid, present in schwannoma and some GBMs, not present
in normal tissue
32,75
Stronger in necrosis than tumor
62,68,74
1174–1175 Protein, higher in lung met tumor (measured by SERS), and in GBM
than normal
78
Cholesterol/lipid band appearing after brain injury
62
1206–1208 Amide III, higher in ODG, MG as measured by SERS
33,63,64,68,71
Phenylalanine
32,63,64
Highest in necrosis, lowest in normal, tumor intermediate
66
1212 Oxygenated hemoglobin
78
1228 Cholesterol/lipid band appearing after brain injury
64
Seen in lung carcinoma met
20,31–33,35,60–69,72–74,
1225–1300 Amide III band (collagen/protein, amino acids)
77,78
20,65
Mainly a-helix conformation in normal brain
20
Shifts to lower wavenumber in glioma, suggesting random coil
structure
62,63
Intensity and position are variable in normal brain
31,62
Higher in lung met tumor, GBM, ODG, PA, astrocytoma cells
32,63,64
Associated with necrosis
32
Shoulder in necrosis, weaker shoulder in tumor
35
Higher in glioma tissue and cells than normal tissue and astrocyte
cells
23,33,61,63,64
1231–1258 Heme (blood)
22,32,33,58,63,64,67,
1260 Overlaps with unsaturated fatty acids/phospholipids
69,74
22
Lipids higher in dura than meningioma
59,66,68,72–74
1268–1270 d5CH unsaturated fatty acids
73
Diagnostically significant for normal tumor and tumor grade
72
Decreases with increasing saturation
59
Higher in gray matter than white matter
79
Higher in white matter than gray matter
79
Slightly increased after radiation exposure
(continued on next page)
Intraoperative Raman Spectroscopy 641

Table 1
(continued )

Raman
Peak (cmL1) Assignment Citation
20,22,23,33,35,58–61,
1296–1302 CH2 twist and wag of phospholipids, fatty acid, cholesterol
63–69,72–75,77,78

67,73
Not a diagnostically significant/variable intensity/position
72
Increases with increasing saturation
8,22,23,67,72,74
Stronger in white matter and myelin fraction than gray matter
and tumor
31,63
Seen in metastatic tumors, astrocytoma cells
77
Higher in Alzheimer hippocampus than normal hippocampus
75
Stronger in necrosis than tumor
61
Cholesterol deposits found in tumor sections
64
1302–1305 Lipid/glycogen (renal cell carcinoma)
60
Cholesterol ester
32,33,35
1313 CH3/CH2 deformation of lipids and proteins
32
Strongest in necrosis, intermediate in tumor, weakest in normal
62
1320–1330 C-H deformation or CH2 bend (protein) increased in lung met
tumor, GBM, ODG, EP, MG (SERS)
23,32,33,60,61,63,64,
1333–1343 Aliphatic amino acids, including tryptophan, nucleic acids
68,71,74
63,73
Glycogen
63,64
Associated with necrosis
63,73
Increased in high-grade tumors and metastatic renal cell
carcinoma
61
1346 Heme/hemoglobin
78
CH deformation of lipid/protein
62
1366 Higher in GBM (SERS)
60,61,72
1372–1376 DNA bases, seen in cell nuclei
32,33
1397 CH2/CH3 deformation of lipids and proteins
58,66
1400–1404 Melanin, seen in metastatic melanoma
69
Quartz substrate (removed via processing)
68
1422 Nucleotides
20,22,23,31–33,35,58–69,
1439–1455 CH2/CH3 deformation of lipids side chains, proteins, amino acids,
71,73,74,76–78
cholesterol/cholesterol ester
67
Intensity and position are variable in normal brain
23,59,74
Higher in white matter than gray matter and tumor
77
Higher in Alzheimer hippocampus than normal hippocampus
22
Lipids higher in normal than meningioma
61,68
Cholesterol deposits found in tumor sections
31,62,63
Higher in ODG, EP, MG renal cell carcinoma met, astrocytoma
73
Variable in malignant tissues
64
Associated with necrosis
35
Higher in glioma cells than astrocytes
61
1454 Heme/hemoglobin
63
1460 Glycogen, seen in metastatic renal cell carcinoma
77
1466 CH2 bend of proteins
60,61,72
1483–1488 DNA bases, seen in cell nuclei
20,33,60,75
1518–1527 Carotenoid
20,59
Present in schwannoma, some GBMs, not present in normal
tissue
32,75
Stronger in necrosis than tumor
37
1540 Nucleic acid, higher in tumor
(continued on next page)
642 Brusatori et al

Table 1
(continued )

Raman
Peak (cmL1) Assignment Citation
60,63,68,71
1550–1555 Tryptophan
32,63
Associated with necrosis, slightly weaker in normal, weakest in
tumor
66,69,71
1546–1568 Hemoglobin
60–63,68,72
1575–1588 C-C stretch of protein, nucleic acids (SERS)
62,63
Higher in ODG, PA, EP, MG
68,72
Seen in cell nuclei, marker for mitosis
78
Sharp band appears after brain injury
33,62,64
1585 Heme (blood)
58,66
1592–1595 Melanin (metastatic melanoma)
61,66
1603–1605 Oxygenated hemoglobin
68
1614 Aromatic amino acids (protein)
23,66
1619–1626 Oxygenated hemoglobin
78
Sharp band appears after brain injury
23,60,66,69,71,73
1635–1640 Water
73
Glioma has higher water content than normal (but normal tissue
may be edematous)
20,23,31–33,37,58,60,61,
1656–1668 Amide I band (protein) mainly a-helix in normal brain
64–69,71,73,74,76–78
20,65
Also assigned to unsaturated fatty acids (nC5C)
22,23,32,33,58,59,63,64,
(1645–1675) Intensity and position are variable in normal brain
66–69,73,74,78
63,67
Associated with necrosis, not as strong in tumor
63,64
Higher in normal than tumor
22
Higher in tumor than normal
31,37
Lower unsaturated lipid content as malignancy increases
73
Lower unsaturation in white matter
73
Higher in gray matter than white
59
Higher in white matter than gray matter
61
Shoulder at 1670 in Alzheimer hippocampus (more deposition of
b-amyloid protein?)
77,78
Weaker after brain injury
23,32,33,59–61,63,68,
1669–1674 Steroid ring of cholesterol/cholesterol ester
69,73
61,68
Cholesterol deposits found in tumor sections
64
1675 Protein (lung carcinoma met)
61
1684 Collagen
62
1688 Higher in EP (SERS)
22,23,32,33,59–61,75,78
1731–1739 C5O of ester groups, cholesterol ester
75
Stronger in necrosis than tumor
22,67
Lipids higher in dura, cortex than tumor
61
Deposits found in tumor sections
78
Not seen in fixed tissues
23,58,63–65,72,73
2850–2853 CH2 symmetric stretch of lipid side chain
65,73
Higher in white matter, lower in gray matter, lowest in tumor
64
Seen in metastatic renal cell carcinoma
23,58,63–65,69,72,73
2880–2886 CH2, CH3 stretch found in lipid (side chain), cholesterol
23,65
Higher in white matter, lower in gray matter
73
Lower in tumor
64
2895 Glycogen/lipid (metastatic renal cell carcinoma)
(continued on next page)
Intraoperative Raman Spectroscopy 643

Table 1
(continued )

Raman
Peak (cmL1) Assignment Citation
23,58,63,65,66,72
2929–2938 CH2/CH3 symmetric stretch of lipid side chains, cholesterol
23,63,66
Aliphatic amino acids (associated with necrosis)
65,72
2958–2960 CH3 asymmetric stretch
73
3010–3015 Unsaturated lipids, lower in tumor
23,58,66,80
3100–3600 OH stretch in water
80
3300 Weak amide band, water
23
Lower in gray matter than white matter

Abbreviations: EP, ependymoma; FADH, flavin adenine dinucleotide; MG, meningioma; NADH, nicotinamide adenine
dinucleotide; ODG, oligodendroglioma; PA, Pilocytic Astrocytoma; SERS, surface-enhanced Raman spectroscopy.

carotenoids, and calcification peaks measured in astrocytoma, and GBM. IDH1 staining was posi-
necrotic tissue. In another study, they differenti- tive for 60% of LA with an overall positive rate of
ated 8 microcystic, 7 transitional, and 2 syncytial 33% for all gliomas. Raman spectroscopic anal-
meningioma tissues from 3 healthy dura mater tis- ysis showed significant (P.01) differences in the
sues with 100% accuracy using linear discriminant spectra between all 3 grades, suggesting that
analysis. Although the spectrum for dura mater spectroscopic biomarkers may provide more
had Raman signatures typical of collagen, the dif- robust delineation of tissue than staining alone.
ference spectrum (meningioma minus dura) was
characteristic of lipids, indicating their relative EARLY APPLICATIONS OF IN VIVO RAMAN
abundance in tumors. A partial least squares algo- SPECTROSCOPY
rithm was used to examine the real-world scenario
of investigating a biopsy specimen likely contain- Small pilot studies have shown the feasibility of
ing mixed dura and tumor tissue.22 This work Raman spectroscopy for in vivo tissue assess-
showed that given a normal distribution of error ment. In vivo instrumentation, facilitated by tech-
the algorithm estimated with 68.3% certainty that nological advancements in spectrometer design,
a specimen was 13% tumor. This certainty in- has enabled rapid Raman collection in a mobile
creases to 95% if the estimate is 26% tumor. setup. Standard screening for neoplastic pro-
Krafft and colleagues23 examined nondried, cesses typically involves gross inspection and
unfixed brain tissue using Raman spectroscopy multiple biopsies of aberrant tissue. The develop-
with a laser spot diameter of 4 mm, 80 mW power, ment of fiber optic Raman probes that fit through
and 15 seconds laser exposure time. Lipid/protein the instrument port of an endoscope has allowed
and cholesterol/protein ratios were more signifi- researchers and clinicians to obtain spectroscopic
cant in distinguishing white from gray matter and information, in a minimally invasive manner, for
distinguishing normal tissue from glioma and me- in vivo evaluation. Bergholt and colleagues27
ningioma tumors, consistent with reports that de- showed significant spectral differences between
gree of malignancy in gliomas is associated with normal esophageal tissue and cancerous tissue
changes in the membrane lipid composition.24 in vivo with 96% diagnostic accuracy. The Raman
Comparative mapping studies showed that signal capture time ranged from 0.1 to 0.5 sec-
Raman spectra acquired from thawed frozen tis- onds. Similar results were shown for benign and
sue yielded similar information to fresh tissue. malignant gastric ulcers.28
Contributions from hemoglobin and its byproducts Further studies evaluated cervical and breast
were the main difference. cancer tissues in vivo. Researchers obtained
Mutations in isocitrate dehydrogenase 1 (IDH1) similar peak intensities in vivo to prior in vitro
and TP53 have been found to be prevalent in studies for distinguishing between benign, low-
70% to 80% of gliomas, showing a direct relation- grade, and high-grade precancerous cervical le-
ship with their grading.25 Gajjar and colleagues26 sions with sensitivities of 95.7%, 82.8%, and
showed that Raman spectroscopy could be used 81.3%, and specificities of 100%, 92.3%, 88.5%,
in vitro to differentiate various grades of glioma. respectively, with 1-second acquisition time.29
Gliomas were subdivided into 3 pathologic sub- In a breast cancer study, a Raman probe was
types: low-grade astrocytoma (LA), anaplastic used intraoperatively to evaluate the margins
644 Brusatori et al

following primary tumor excision with 1-second and colleagues36 addressed finding the optimal
acquisition and diagnosis time. These margins camera temperature and lighting conditions for us-
were then excised per standard surgical practice. ing a Raman spectroscopy probe during neurosur-
Using a previously validated algorithm, 1 malig- gery. This group showed that the signal-to-noise
nant specimen was identified from among the ratio increased as the camera temperature
other benign breast tissue margins. This specimen decreased and integration time increased. In addi-
was grossly normal; however, on histopathologic tion, they identified several light sources, including
confirmation reoperation was necessary. This sec- microscope light, directed surgical lighting (low or
ond procedure may have been prevented had a high), and fluorescent light, that prohibited accu-
Raman probe been used to guide excision.30 rate Raman measurement of the sample.36 How-
ever, lighting from exterior lights, indirect low
DEVELOPMENTS TOWARD IN VIVO surgical lighting, and LCD (liquid crystal display)
NEUROSURGICAL DIAGNOSTICS monitors did not adversely affect the Raman spec-
troscopy measurement.36 Furthermore, this study
The applicability of Raman spectroscopy to neuro- and previous work by the same group showed
surgery continues to evolve as its effectiveness in that a spectroscopic probe can be used to detect
distinguishing disparate tissues is revealed. While necrotic tissue in vivo using inelastic scat-
analyzing single cells, Banerjee and colleagues31 tering.36,37 Of note, the probe used the Medtronic
showed that Raman spectra were very similar SureTrak neuronavigation system to coregister the
across the GBM cell lines from 4 different patients. site with preoperative MRI.36 There are reports of
On a larger scale, recent work from Kalkanis and the British Columbia Cancer Agency spin-off com-
colleagues32 further shows the viability of Raman pany Verisante performing intraoperative neuro-
spectroscopy as a neurosurgical adjunct in tumor surgical diagnostics in collaboration with Imperial
resection. Focusing on GBM, they developed College Healthcare NHS Trust, although published
Raman spectral maps for normal brain, GBM, and findings are not yet available.38
necrosis from 40 tissue specimens, with a total of In a recent investigation, our group developed a
17,138 spectra from 95 distinct regions.32 These custom-built portable Raman spectrometer with
were examined along with adjacent H&E-stained probe using a 785-nm laser as the excitation
sections. Normal brain tissue samples, consisting source to acquire 66 Raman spectra of tissue
primarily of gray matter, were higher in lipid content samples from 18 patients comprising 9 women
than tumor and necrosis. Necrosis showed higher and 9 men with a mean age of 54 years in the oper-
protein and nucleic acid content, whereas GBM ating room immediately following surgical resec-
was between the two tissue types. Freeze artifacts tion. From 2 to 5 tissue samples per patient were
affected the discriminant function analysis accu- examined with Raman spectroscopy in a light-
racy in distinguishing tissue types in the validation tight chamber, with 1 site per sample being
data set, from 97.8% to 77.5%. However, this fac- recorded. After spectral acquisition, the speci-
tor is moot in the intraoperative setting. This group mens were submitted to pathology for evaluation.
subsequently published studies showing that the The Raman spectra were preprocessed to mini-
same label-free imaging can distinguish boundaries mize any fluorescence contributions and to pro-
of infiltrating tumor and areas of hemorrhage.33,34 vide better interpretability. Fig. 4 shows the
Tanahashi and colleagues35 used an ex vivo
platelet-derived growth factor subunit B–induced
infiltrative glioma murine model in their Raman
studies. It was designed to replicate World Health
Organization grade II oligodendrogliomas and the
less distinct margins between infiltrative glioma
and normal tissue. They examined cultured cells
as well as native brain tissue, identifying spectral
peaks that characterized glioma, which were pri-
marily derived from phenylalanine, lipid, DNA, and
protein. Tumor prediction was reported to have a
sensitivity of 98.3% and specificity of 75.0%. The
sensitivity and specificity of whole tissue samples
were 58.8% and 76.4%, respectively. Fig. 4. Mean Raman spectra of normal tissue, solid tu-
Recently 2 research groups, in addition to our mor, necrosis, and hemorrhage normalized to the
group, reported taking Raman spectroscopy 1003 cm 1 band. Spectra are vertically offset and
into the neurosurgical suite. In 2015, Desroches scaled for visualization purposes only.
Intraoperative Raman Spectroscopy 645

Raman spectra of tissue normalized to the peak at 1296 cm 1. Many lipids found in both gray and
1003 cm 1. This band is the phenyl ring breathing white matter have Raman bands at w1064 and
mode of phenylalanine in the protein and is not w1299 cm 1.39 Necrosis, an indicator of malig-
sensitive to conformational changes of protein. nancy in astrocytoma and common in GBM, re-
The averaged spectrum from the tumor group con- sults in cell membrane breakdown. Magnetic
sists of spectra obtained from GBM (grade 4), me- resonance spectroscopy studies have shown
ningioma (grade 1), anaplastic oligoastrocytoma that high levels of mobile lipids (cholesterol ester
(grade 3), ependymoma (grade 2), anaplastic oli- and triacylglycerol associated with the presence
godendroglioma (grade 3), and oligodendroglia of lipid droplets) correlate with necrosis.40 Bands
(grade 2). For normal tissue, the spectrum may at 1064 and 1299 cm 1 in the spectra of necrotic
have contributions from both gray matter and tissue may indicate the presence of triacylglycerol,
white matter. In adults, white matter, which primar- which has spectral features at these wave
ily consists of glial cells and myelinated axons, has numbers; however, further investigation is war-
higher total lipid concentration, lower water con- ranted. In addition, the spectrum of necrotic tissue
tent, and lower capillary density than gray matter, also shows observable bands at 985 cm 1 corre-
which is composed of neuronal cell bodies. sponding with calcifications and 1155 and
Because white matter and gray matter differ in 1515 cm 1 with carotenoids. The spectrum from
their protein/lipid contents, further investigation a region of hemorrhage shows bands at 1003,
(neuropathology identification) is warranted for 1122, 1224, 1458, and 1561 to 1636 cm 1 that
further differentiation. are consistent with hemoglobin and a band at
Distinct spectral differences between the mean 1342 cm 1 that indicates fibrin.41 This work re-
spectra of normal, tumor, and necrotic tissue sam- flects a more direct interoperational study that
ples as deemed by pathology as well as hemor- closely resembles in vivo measurements.
rhage are evident. In general, characteristic Although the end goal is for Raman spectroscopy
peaks between 1760 and 1500 cm 1 are attributed to have a place in standard clinical practice, con-
to amide I vibrations (C5O), with contributions ventional Raman spectroscopy has encountered
of water, nucleic acids, and lipids (C5C, C5O). challenges such as weak signals, long acquisition
Note that the amide II band at w1550 cm 1 times, background fluorescence, and data process-
(C-N, N-H) is weak and cannot be observed in ing and instrument costs. Advancements in instru-
the absence of resonance enhancement. The re- mentation designed to improve sensitivity and
gion between 1500 and 1400 cm 1 shows spectral reduce the size and cost of Raman spectrometers,
contributions caused by C-H, CH2, and CH3 vibra- as well as strategies to promote signal enhance-
tions characteristic of lipids and protein. Bands in ment (ie, reduce acquisition time) and mitigate fluo-
the region between 1400 and 1200 cm 1 result rescence are currently under development.
from amide III vibrations (C-N, N-H) with contribu- In many cases, the measured spectra consist of
tions from CH2/CH3 vibrations of protein and a small component of Raman scatter on top of a
lipids (phospholipids) as well as from nucleic large fluorescence background, resulting in a
acids. The region between 1200 and 800 cm 1 poor signal-to-noise ratio. Although all light causes
contains spectral contributions from nucleic acids, the Raman effect, NIR light reduces interference
lipids (C-C, C-O), proteins (C-N, C-C), and C-O-C from fluorescence in biological samples because
stretching of carbohydrates (saccharides/polysac- of its lower excitation energy. In contrast, shorter
charides). The region between 400 and 800 cm 1 wavelengths induce larger Raman signals but
is associated with nucleic acids and nucleotide stimulate more fluorescent background. Complex
conformation; however, bands arising from spectral processing algorithms may be used to
glucose, glycogen, cholesterol, and lipids are assist in analyzing biological Raman data.
also present in this spectral area. Not only can Although it is generally believed that NIR
the spectral differences be used for discriminatory (785 nm) or IR (1064 nm) wavelengths frequently
purposes they can provide insight into tumor used in biological Raman spectroscopy impart
growth, compositional changes, and metastasis. no damage to surrounding tissue, assurance of
The lipid composition of brain tumors differs this technology’s nondestructiveness must be
from that of normal tissue. One such difference is shown. There is a dearth of research examining
noted. Compared with that of normal and necrotic the damage potential on in vivo brain tissue, and
tissue, the spectrum associated with tumor shows our group is currently studying this aspect.
reduced spectral features of lipid bands at 1064 For improved signal-to-noise ratio, there are
and 1296 cm 1. These bands arise from skeletal several complementary techniques that have
trans C-C stretching and CH2 twisting of acyl been developed based on the Raman effect.
chains, respectively, with protein band overlap at Table 2 identifies several of these methods, in
646 Brusatori et al

Table 2
Variations of Raman spectroscopy

SERS
Method Signal enhancement is implemented through the use of metallic surfaces
(nanoparticles/nanostructures). Electromagnetic enhancement is considered
the dominant contributor to most SERS processes and involves the interaction
of surface plasmons (generated by incident light) on metallic nanostructures
with Raman active molecules. The surface plasmons on the nanostructures
transfer energy to the vibrational modes of molecules that are in close
proximity, thus providing a means for signal enhancement53
Advantage vs Significant enhancement of Raman signal is reported by a factor of 1010
standard Raman
Disadvantage for Requires additional steps during surgery, such as adding a nanoprobe molecule
intraoperative to the tissue of interest for enhancement,81 SERS tag must be biocompatible,
use Raman measurement must be in close proximity to tag. The analysis is
confined to tens of nanometers from the nanoparticle or probe. The
variability of the nanoparticles creates nonreproducible results
SORS
Method Scatter photons are collected at the laser spot and laterally offset sites. Using
scaled subtraction algorithms, surface and subsurface spectra can be derived50
Advantage vs Using an offset collection point allows data to be collected from deeper within
standard Raman the area of interest; up to 4 mm was shown.50 By comparison, standard Raman
only penetrates a few hundred micrometers. Reduces tissue fluorescence
Disadvantage for Interrogation and collection offset of at least 3.5 mm are recommended, tumor-
intraoperative thickness detection limitation of 2 mm (breast tissue)51; complex hardware
use requirements
CARS
Method Typically uses 2 pulsed trains, a pump beam at frequency up and a Stokes beam
at frequency us that interact with a sample. When energy difference up us
matches a molecular vibration of the sample, a resonantly enhanced anti-
Stokes signal is generated at a frequency uas 5 2up us82
Advantage vs Background fluorescence does not interfere with the sample, and the signal is 4
standard Raman orders of magnitude stronger than standard Raman82
Disadvantage for Requires tunable pulsed lasers to probe different molecules in the sample.
intraoperative Difficult to effectively couple and synchronize the lasers into a handheld or
use portable intraoperative device
SRS
Method SRS, like CARS, uses 2 pulsed trains (a pump beam at frequency up and a Stokes
beam at frequency us) that coincide on the sample. When energy difference
up us matches a molecular vibration of the sample, the scattering process is
resonantly enhanced, However, unlike CARS, the SRS signal occurs at the
frequency of the excitation beams83
Advantage vs Greater signal strength of approximately 4 orders of magnitude
standard Raman
Disadvantage for Requires tunable pulsed lasers to probe different molecules in the sample.
intraoperative Difficult to effectively couple and synchronize the lasers into a handheld or
use portable intraoperative device84
RRS
Method Signal enhancement with resonance Raman is achieved when frequency of
incident radiation coincides with the frequency of an electronic transition of a
molecule, and provides energy to excite electrons to a higher electronic state.
This technique can selectively augment signals affiliated with chromophores
and other large conjugated molecules
(continued on next page)
Intraoperative Raman Spectroscopy 647

Table 2
(continued )

Advantage vs Increased signal strength by a factor of 102 to 106


standard Raman
Disadvantage for Provides more limited/selective molecular information. Non–resonance-
intraoperative enhanced bands may seemingly disappear under the intensity of resonance-
use enhanced spectral peaks. Requires a tunable laser to selectively isolate the
contributions from different chromophores. Carotenoids show enhancement
in the visible region of the spectra, whereas DNA is enhanced in the UV
region. UV laser sources can cause cellular damage. Fluorescence backgrounds
can be significant because of excitation coinciding with UV-visible absorption

Abbreviations: CARS, coherent anti-Stokes Raman spectroscopy; RRS, resonance Raman scattering; SORS, spatially offset
Raman spectroscopy; SRS, stimulated Raman scattering.

contrast with standard Raman spectroscopy, and could detect residual microscopic tumor in resec-
lists the challenges to incorporating them into an tion beds that were not detectable with visual in-
intraoperative surgical tool. spection. Another strategy for SERS diagnostics
of tissue involves the use of tailored nanoparticles
SURFACE-ENHANCED RAMAN labeled with specific antibodies designed to
SPECTROSCOPY adsorb to selective antigens in tissue. This
approach has been shown to enable signal-
Surface-enhanced Raman spectroscopy (SERS) is
enhanced Raman measurement of an antigen un-
a method to enhance the weak signals in sponta-
der investigation. Schütz and colleagues48
neous Raman spectroscopy that uses metallic
showed SERS labels for imaging of the tumor sup-
nanostructures capable of providing surface-
pressor p63 in prostate biopsies.
enhanced Raman scattering with potential
Often there are differences in the number of
enhancement factors of w105 to 1010. There are
modes present in SERS spectrum compared
3 mechanisms of enhancement described in the
with a traditional Raman spectrum; that is, the
literature: an electromagnetic effect, molecular
SERS spectrum may have additional bands (or
resonance within the molecule, and charge trans-
the absence of bands) compared with those found
fer resonance between the molecule and sub-
in a conventional Raman spectrum. This difference
strate.42 The electromagnetic effect is dominant,
results from slight changes in the symmetry of a
with signal enhancement resulting from the reso-
molecule caused by surface adsorption.
nant interaction of light with plasmons excited at
the surface of the structure. An enhanced electric SPATIALLY OFFSET RAMAN SPECTROSCOPY
field is created that is localized to a region of a
few nanometers from the surface that subse- Traditional Raman spectral acquisition of tissue is
quently can lead to an increase in the signal typically obtained using a 180 backscatter geom-
of Raman bands for molecules located in the vicin- etry and is limited to near-surface measurements
ity of the metallic substrate.43 Kircher and col- at depths within the first few hundred micrometers
leagues44 reported using SERS imaging in a of the surface. Spatially offset Raman spectros-
mouse model for in vivo tumor resection. In copy (SORS) enables measurements from subsur-
this study, gadolinium-DOTA (1,4,7,10–tetraaza- face layers in diffusely scattering media,49 in
cyclododecane–1,4,7,10–tetraacetic acid)–coated volumes as deep as 4 mm into the sample.50 As
silica-gold nanoparticles were injected into the tail opposed to traditional Raman, in which laser illu-
vein of an orthotopic GBM mouse model. The mination and collection are from the same area
nanoparticles were shown to accumulate in cells of the sample, SORS involves collecting the scat-
within the tumor, but not in normal tissue, without tered light from a point that is away (laterally offset)
necessitating a specific targeting mechanism. This from the laser illumination. For a 2-layered sample,
uptake is generally attributed to the enhanced 2 measurements are required to recover the
permeability and retention effect.45–47 Postinjec- Raman spectra of the individual layers. One spec-
tion visualization of the tumor was achieved with trum is typically taken at zero offset, whereas the
Raman imaging through the skin and skull of the other is taken at nonzero offset. For this case, a
live mice because of the unique and enhanced scaled subtraction of the two spectra may be suf-
signal of the nanoparticles. Twenty-four hours af- ficient to recover the spectrum of the sublayer.
ter nanoparticle injection, sequential tumor resec- However, for a multilayered system, more sophis-
tion was performed on live mice. SERS imaging ticated methods may need to be used. Clinical
648 Brusatori et al

applications of this technique can be extended to broadband CARS, the pump pulse has a narrow
bone50 and breast tumor margin evaluation.51 bandwidth and defines the spectral resolution,
whereas the Stokes pulse is spectrally broad (usu-
TRANSMISSION RAMAN SPECTROSCOPY ally in the femtosecond regime). Multiple Raman
transitions within the bandwidth of the Stokes pulse
Transmission Raman spectroscopy is considered a are excited and are probed. This method allows the
form of SORS, with collection and illumination entire spectrum of excited states to be obtained at
points being on opposite sides of the sample. How- once and has been extended in the fingerprint re-
ever, unlike SORS it is not able to provide the signa- gion of the spectra to allow imaging of biological tis-
tures of individual layers within the sample. Instead, sue. Camp and colleagues53 showed high-speed
it provides information on the entire sample vol- chemical imaging in two-dimensional and three-
ume. For tissue interrogation, the coupling of laser dimensional views of interfaces between xenograft
radiation into deep tissue layers is hindered by los- brain tumors in mice and the surrounding healthy
ses of laser radiation at the surface of the sample tissue using color-coded images corresponding
from scattering as well as the diffuse nature of with Raman bands at 2850 and 2944 cm 1 as well
photon propagation through tissue.52 However, as those in the fingerprint region, 730, 785, 1004,
by using a dielectric filter on the surface of the tis- 1547, and 1564 cm 1.
sue, Stone and Matousek52 detected Raman sig-
nals from depths of up to 2.7 cm within a breast STIMULATED RAMAN SPECTROSCOPY
phantom made up of porcine tissues.
Stimulated Raman spectroscopy (SRS) signals
COHERENT ANTI-STOKES RAMAN are typically generated by the use of 2-pico-
SPECTROSCOPY second pulsed trains that coincide on a sample
(a pump beam at a frequency up and a Stokes
Coherent anti-Stokes Raman spectroscopy (CARS) beam at a frequency us). By tuning the frequency
is a third-order nonlinear process that typically uses difference between the pump and Stokes beams
picosecond pulsed lasers, in part because of the to match the frequency of a molecular vibration,
linewidth of most Raman active transitions being uvib 5 up – us, stimulated excitation of the vibra-
around 10 cm 1. With this technique, a pump laser tional transition occurs. This nonlinear process
(at a frequency up), a probe (at a frequency upr), causes an intensity loss in the pump beam and
and a Stokes laser (at a frequency us) interact with an intensity gain in the Stokes beam. By modu-
a sample via a wave mixing process to lating 1 of the beams, say the Stokes beam,
generate an anti-Stokes field at a frequency of and measuring the signal of the pump beam at
uas 5 up us 1 upr. Commonly the probe beam the frequency of modulation, the intensity loss of
is the same as the pump, resulting in an anti- the pump beam caused by excitation of molecular
Stokes frequency of uas 5 2up us. To significantly vibrations can be distinguished from noise to
enhance the CARS signal, the pump beam is tuned generate a high-speed image of a selected
while the Stokes beam is held fixed so that the fre- Raman band (vibrational transition). Freudiger
quency difference (beat frequency) matches the fre- and colleagues54 imaged mouse models of 4
quency of a Raman active vibrational mode of the common brain disorders: GBM, metastasis,
molecule under study (uvib 5 up – us). This process demyelination, and stroke. Using 2 colored im-
results in a signal yield that is 4 to 5 orders of magni- ages corresponding with CH2 and CH3 stretching
tude higher than that obtained with conventional vibrations at 2845 and 2940 cm 1 of lipids and
Raman. CARS is typically used for video-rate imag- protein, they were able to delineate brain tumor
ing of single Raman bands, with many studies margins. For this work, imaging speed was limited
focusing on the CH-/OH stretch region of the by the speed of the lock-in amplifier with signal
spectra for tissue analysis (2500–3500 cm 1). Nar- collection in sample transmission mode. Using
row laser bandwidth, speed of laser tuning rates, SRS, Ji and colleagues55 imaged biopsies from
and nonresonant background interference limit neurosurgical patients with CNS tumors at 2
this technique to species with high oscillator density Raman frequencies (2845 and 2930 cm 1) to
and uniquely isolated Raman peaks.53 This limita- compare with H&E microscopy and found SRS
tion prevents access to Raman biomarkers in the to detect tumor infiltration in agreement with stan-
fingerprint region (500–1800 cm 1) of the spectra.53 dard H&E light microscopy (k 5 0.86). They devel-
Multiplex techniques have been developed to oped a classifier based on cellularity, axonal
simultaneously excite multiple Raman transitions, density, and protein/lipid ratio in SRS images
providing a more complete vibrational picture than that detects tumor infiltration with 97.5% sensi-
is found with the single-frequency method. With tivity and 98.5% specificity.
Intraoperative Raman Spectroscopy 649

For margin assessment during surgical resec- The road to clinical application also involves mar-
tion, it is desired to obtain high-resolution Raman ket assessment, preclinical validation, first-in-
spectra over a large sample area. Raster scanning human, and phase I safety studies; clinical trials to
with traditional Raman has a long data acquisition support regulatory guidelines; regulatory approval;
time (whours). Coherent techniques such as as well as quality assurance under Good
CARS and SRS allow much more rapid image Manufacturing Practices through the lifecycle of
acquisition than is afforded by spontaneous product development. Because there is no Raman
Raman imaging techniques. However, CARS and predicate device for neurosurgery, a careful clinical
SRS systems are larger and more complex setups study under US Food and Drug Administration
that are difficult to transition to an intraoperative guidance is required. The intraoperative tool under-
environment. going experimental evaluation by the authors has a
small footprint suitable for space constraints in the
RESONANCE RAMAN SCATTERING operating room and can measure and record
Raman spectra in milliseconds with resolution com-
With resonance Raman scattering (RRS), the parable with current bench-top systems.
wavelength of the laser beam incident on a sample
is tuned to (or near) an electronic transition of a SUMMARY AND FUTURE DIRECTIONS
molecule under study. The vibrational modes
associated with a particular transition show greatly Noninvasive or minimally invasive in vivo tools
increased Raman scattering intensity (by factors of that can provide rapid tissue assessment and/or
102-106). This intensity typically overshadows monitor treatment therapies have potential appli-
Raman signals from all other transitions and can cations in many fields of medicine. Raman spec-
provide signal enhancement of particular species troscopy can assist in uncovering the molecular
under investigation. Even in a complex sample basis of disease and provide objective, quantifi-
with numerous vibrational modes, RRS allows able molecular information for diagnosis and
few vibrational modes to be studied at a time. treatment evaluation. Numerous experimental
This limitation can reduce the complexity of the studies have shown the capability of Raman
spectrum to allow easier identification. However, spectroscopy for tissue characterization. Raman
RRS is often impaired by fluorescence back- spectroscopy may complement or possibly sup-
ground, which can obscure the Raman signals plant current methods of surgical guidance in tu-
but may be avoided using short (deep UV) wave- mor resection. The continued development of
lengths.56 UV RRS has typical application in study- Raman spectral databases for various brain disor-
ing biological assemblies containing DNA or RNA, ders, tissue classification methodologies, and in-
whereas visible RRS has typically been used for strument designs trending toward data with
analyzing strongly absorbing organic components greater resolution, shorter collection times, and
such as carotenoids. Using a confocal micro- higher accuracy will ensure that Raman spectros-
Raman system with 532-nm excitation, Zhou and copy becomes a powerful in vivo tool in neurosur-
colleagues57 used RR spectroscopy to discrimi- gery. The prospect of near–real-time imaging
nate between normal brain tissues, cancerous combined with accuracy in characterizing malig-
brain tumors, and benign brain lesions. They nancies and distinguishing them from normal
used statistical methods of principal component brain tissue and tissue with infiltrating cancer cells
analysis and the support vector machine to can be a revolutionary step forward in how physi-
analyze the RR spectral data collected from cians practice medicine.
meningeal tissues with a sensitivity of 90.9% and
specificity of 100%. ACKNOWLEDGMENTS
In addition to methodologies to enhance the The authors would like to acknowledge Rachel E.
weak Raman signals associated with spontaneous Kast and James Tseng for their collaboration and
Raman scattering and reduce spectral acquisition work in this field.
time, the translation for clinical use also involves
the development of comprehensive spectral data- REFERENCES
bases and tissue classification methodologies that
can be compared with current gold standards. 1. Central Brain Tumor Registry of the United States.
Validation studies must be performed to confirm 2016 CBTRUS fact sheet. 2016. Available at: http://
that algorithms developed on ex vivo specimens www.cbtrus.org/factsheet/factsheet.html. Accessed
are applicable to in vivo tissues. Best-practice January 26, 2017.
techniques for data processing, acquisition, and 2. Marko NF, Weil RJ, Schroeder JL, et al. Extent of
classification need to be developed and adopted. resection of glioblastoma revisited: personalized
650 Brusatori et al

survival modeling facilitates more accurate survival prostate cancer. Cancer Metastasis Rev 2014;
prediction and supports a maximum-safe-resection 33(2–3):673–93.
approach to surgery. J Clin Oncol 2014;32(8): 20. Mizuno A, Kitajima H, Kawauchi K, et al. Near-
774–82. infrared Fourier transform Raman spectroscopic
3. Sanai N, Polley MY, McDermott MW, et al. An extent study of human brain tissues and tumours.
of resection threshold for newly diagnosed glioblas- J Raman Spectros 1994;25(1):25–9.
tomas. J Neurosurg 2011;115(1):3–8. 21. Koljenovic S, Bakker Schut TC, Wolthuis R, et al. Tis-
4. Jakola AS, Myrmel KS, Kloster R, et al. Comparison sue characterization using high wave number Raman
of a strategy favoring early surgical resection vs a spectroscopy. J Biomed Opt 2005;10(3):031116.
strategy favoring watchful waiting in low-grade gli- 22. Koljenovic S, Schut TB, Vincent A, et al. Detection of
omas. JAMA 2012;308(18):1881–8. meningioma in dura mater by Raman spectroscopy.
5. Stummer W, Pichlmeier U, Meinel T, et al. Fluores- Anal Chem 2005;77(24):7958–65.
cence-guided surgery with 5-aminolevulinic acid 23. Krafft C, Miljanic S, Sobottka SB, et al. Near-infrared
for resection of malignant glioma: a randomised Raman spectroscopy to study the composition of hu-
controlled multicentre phase III trial. Lancet Oncol man brain tissue and tumors. Proc SPIE Int Soc Opt
2006;7(5):392–401. Eng 2003;5141(Diagnostic optical spectroscopy in
6. Sun Y, Hatami N, Yee M, et al. Fluorescence lifetime biomedicine II):230–6.
imaging microscopy for brain tumor image-guided 24. Campanella R. Membrane lipids modifications in hu-
surgery. J Biomed Opt 2010;15(5):056022. man gliomas of different degree of malignancy.
7. Butte PV, Mamelak AN, Nuno M, et al. Fluorescence J Neurosurg Sci 1992;36(1):11–25.
lifetime spectroscopy for guided therapy of brain tu- 25. Zhang C, Moore LM, Li X, et al. IDH1/2 mutations
mors. Neuroimage 2011;54(Suppl 1):S125–35. target a key hallmark of cancer by deregulating
8. Santagata S, Eberlin LS, Norton I, et al. Intraopera- cellular metabolism in glioma. Neuro Oncol 2013;
tive mass spectrometry mapping of an onco- 15(9):1114–26.
metabolite to guide brain tumor surgery. Proc Natl 26. Gajjar K, Heppenstall LD, Pang W, et al. Diagnostic
Acad Sci U S A 2014;111(30):11121–6. segregation of human brain tumours using Fourier-
9. Raman CV. A new radiation. Indian J Phys 1928;2: transform infrared and/or Raman spectroscopy
387–98. coupled with discriminant analysis. Anal Methods
10. Chase B. A new generation of Raman instrumenta- 2012;5(1):89–102.
tion. Appl Spectrosc 1994;48(7):14–9. 27. Bergholt MS, Zheng W, Lin K, et al. In vivo diagnosis
11. Bernath BF. Light scattering and the Raman effect. of esophageal cancer using image-guided Raman
Spectra of atoms and molecules. Oxford (United endoscopy and biomolecular modeling. Technol
Kingdom): Oxford University Press; 2005. p. Cancer Res Treat 2011;10(2):103–12.
293–320. 28. Bergholt MS, Zheng W, Lin K, et al. Raman endos-
12. Campbell D, Pethrick RA, White JR. Polymer charac- copy for in vivo differentiation between benign and
terization: physical techniques. 2nd edition. Chelten- malignant ulcers in the stomach. Analyst 2010;
ham (United Kingdom): Stanley Thornes; 2000. 135(12):3162–8.
13. Caroline ML, Vasudevan S. Growth and characteriza- 29. Duraipandian S, Mo J, Zheng W, et al. Near-infrared
tion of L-phenylalanine nitric acid, a new organic Raman spectroscopy for assessing biochemical
nonlinear optical material. Mater Lett 2009;63(1):41–4. changes of cervical tissue associated with precar-
14. Movasaghi Z, Rehman S, Rehman IU. Raman spec- cinogenic transformation. Analyst 2014;139(21):
troscopy of biological tissues. Appl Spectrosc Rev 5379–86.
2007;42(5):493–541. 30. Haka AS, Volynskaya Z, Gardecki JA, et al. In vivo
15. Mayo DW, Miller FA, Hannah RW. Course notes on margin assessment during partial mastectomy
the interpretation of infrared and Raman spectra. breast surgery using Raman spectroscopy. Cancer
Hoboken (NJ): John Wiley; 2004. Res 2006;66(6):3317–22.
16. Nelson D, Cox M. Lehninger principles of biochem- 31. Banerjee HN, Zhang L. Deciphering the finger prints
istry. 4th edition. New York: WH Freeman and Com- of brain cancer astrocytoma in comparison to astro-
pany; 2005. cytes by using near infrared Raman spectroscopy.
17. Warburg O, Wind F, Negelein E. The metabolism Mol Cell Biochem 2007;295(1–2):237–40.
of tumors in the body. J Gen Physiol 1927;8(6): 32. Kalkanis SN, Kast RE, Rosenblum ML, et al. Raman
519–30. spectroscopy to distinguish grey matter, necrosis,
18. Tashibu K. Analysis of water content in rat brain us- and glioblastoma multiforme in frozen tissue sec-
ing Raman spectroscopy. No To Shinkei 1990; tions. J Neurooncol 2014;116(3):477–85.
42(10):999–1004 [in Japanese]. 33. Kast RE, Auner GW, Rosenblum ML, et al. Raman
19. Kast RE, Tucker SC, Killian K, et al. Emerging tech- molecular imaging of brain frozen tissue sections.
nology: applications of Raman spectroscopy for J Neurooncol 2014;120(1):55–62.
Intraoperative Raman Spectroscopy 651

34. Kast R, Auner G, Yurgelevic S, et al. Identification of immuno-SERS microscopy. Chem Commun
regions of normal grey matter and white matter from (Camb) 2011;47(14):4216–8.
pathologic glioblastoma and necrosis in frozen sec- 49. Matousek P, Morris MD, Everall N, et al. Numerical
tions using Raman imaging. J Neurooncol 2015; simulations of subsurface probing in diffusely scat-
125(2):287–95. tering media using spatially offset Raman spectros-
35. Tanahashi K, Natsume A, Ohka F, et al. Assessment copy. Appl Spectrosc 2005;59(12):1485–92.
of tumor cells in a mouse model of diffuse infiltrative 50. Buckley K, Kerns JG, Parker AW, et al. Decomposi-
glioma by Raman spectroscopy. Biomed Res Int tion of in vivo spatially offset Raman spectroscopy
2014;2014:860241. data using multivariate analysis techniques.
36. Desroches J, Jermyn M, Mok K, et al. Characteriza- J Raman Spectros 2014;45(2):188–92.
tion of a Raman spectroscopy probe system for in- 51. Keller MD, Wilson RH, Mycek MA, et al. Monte Carlo
traoperative brain tissue classification. Biomed Opt model of spatially offset Raman spectroscopy for
Express 2015;6(7):2380–97. breast tumor margin analysis. Appl Spectrosc
37. Jermyn M, Mok K, Mercier J, et al. Intraoperative 2010;64(6):607–14.
brain cancer detection with Raman spectroscopy 52. Stone N, Matousek P. Advanced transmission Raman
in humans. Sci Transl Med 2015;7(274):274ra219. spectroscopy: a promising tool for breast disease
38. Verisante Technology, Verisante Technology, Inc diagnosis. Cancer Res 2008;68(11):4424–30.
announces brain tumour study in the UK 2015. Avail- 53. Camp CH Jr, Lee YJ, Heddleston JM, et al. High-
able at: http://verisante.com/news/143/verisante- speed coherent Raman fingerprint imaging
technology-inc-announces-brain-tumour-study-in- of biological tissues. Nat Photonics 2014;8(8):627–34.
the-uk-/. Accessed January 26, 2017. 54. Freudiger CW, Pfannl R, Orringer DA, et al. Multicol-
39. Krafft C, Neudert L, Simat T, et al. Near infrared ored stain-free histopathology with coherent Raman
Raman spectra of human brain lipids. Spectro- imaging. Lab Invest 2012;92(10):1492–502.
chim Acta A Mol Biomol Spectrosc 2005;61(7): 55. Ji M, Lewis S, Camelo-Piragua S, et al. Detection of
1529–35. human brain tumor infiltration with quantitative stim-
40. Delikatny EJ, Chawla S, Leung DJ, et al. MR-visible ulated Raman scattering microscopy. Sci Transl Med
lipids and the tumor microenvironment. NMR Bio- 2015;7(309):309ra163.
med 2011;24(6):592–611. 56. Asher SA. UV resonance Raman studies of molecu-
41. Lemler P, Premasiri WR, DelMonaco A, et al. NIR lar structure and dynamics: applications in physical
Raman spectra of whole human blood: effects of and biophysical chemistry. Annu Rev Phys Chem
laser-induced and in vitro hemoglobin denaturation. 1988;39(1):537–88.
Anal Bioanal Chem 2014;406(1):193–200. 57. Zhou Y, Liu CH, Sun Y, et al. Human brain cancer
42. Muehlethaler C, Leona M, Lombardi JR. Review of studied by resonance Raman spectroscopy.
surface enhanced Raman scattering applications J Biomed Opt 2012;17(11):116021.
in forensic science. Anal Chem 2016;88(1):152–69. 58. Krafft C, Kirsch M, Beleites C, et al. Methodology for
43. Kambhampati P, Child CM, Foster MC, et al. On the fiber-optic Raman mapping and FTIR imaging of
chemical mechanism of surface enhanced Raman metastases in mouse brains. Anal Bioanal Chem
scattering: experiment and theory. J Chem Phys 2007;389(4):1133–42.
1998;108(12):5013–26. 59. Kohler M, Machill S, Salzer R, et al. Characterization
44. Kircher MF, de la Zerda A, Jokerst JV, et al. A brain of lipid extracts from brain tissue and tumors using
tumor molecular imaging strategy using a new triple- Raman spectroscopy and mass spectrometry. Anal
modality MRI-photoacoustic-Raman nanoparticle. Bioanal Chem 2009;393(5):1513–20.
Nat Med 2012;18(5):829–34. 60. Bergner N, Krafft C, Geiger KD, et al. Unsupervised
45. Maeda H. The link between infection and cancer: tumor unmixing of Raman microspectroscopic images
vasculature, free radicals, and drug delivery to tumors for morphochemical analysis of non-dried brain tu-
via the EPR effect. Cancer Sci 2013;104(7):779–89. mor specimens. Anal Bioanal Chem 2012;403(3):
46. Maeda H, Nakamura H, Fang J. The EPR effect for 719–25.
macromolecular drug delivery to solid tumors: 61. Krafft C, Sobottka SB, Schackert G, et al. Raman
Improvement of tumor uptake, lowering of systemic and infrared spectroscopic mapping of human pri-
toxicity, and distinct tumor imaging in vivo. Adv mary intracranial tumors: a comparative study.
Drug Deliv Rev 2013;65(1):71–9. J Raman Spectros 2006;37(1–3):367–75.
47. Thakor AS, Gambhir SS. Nanooncology: the future of 62. Aydin O, Altas M, Kahraman M, et al. Differentiation
cancer diagnosis and therapy. CA Cancer J Clin of healthy brain tissue and tumors using surface-
2013;63(6):395–418. enhanced Raman scattering. Appl Spectrosc 2009;
48. Schütz M, Steinigeweg D, Salehi M, et al. Hydrophili- 63(10):1095–100.
cally stabilized gold nanostars as SERS labels for 63. Bergner N, Bocklitz T, Romeike BFM, et al. Identifi-
tissue imaging of the tumor suppressor p63 by cation of primary tumors of brain metastases by
652 Brusatori et al

Raman imaging and support vector machines. Che- soft reference information. Anal Bioanal Chem
mometr Intell Lab Syst 2012;117(0):224–32. 2011;400(9):2801–16.
64. Bergner N, Romeike BFM, Reichart R, et al. Raman 74. Beljebbar A, Dukic S, Amharref N, et al. Ex vivo and
and FTIR microspectroscopy for detection of brain- in vivo diagnosis of C6 glioblastoma development
metastasis. In: Ramanujam N, Popp J, editors. Pro- by Raman spectroscopy coupled to a microprobe.
ceedings of SPIE-OSA Biomedical Optics. Munich Anal Bioanal Chem 2010;398(1):477–87.
(Germany): Optical Society of America; 2011. 75. Koljenovic S, Choo-Smith LP, Bakker Schut TC, et al.
65. Mizuno A, Hayashi T, Tashibu K, et al. Near-infrared Discriminating vital tumor from necrotic tissue in hu-
FT-Raman spectra of the rat brain tissues. Neurosci man glioblastoma tissue samples by Raman spec-
Lett 1992;141(1):47–52. troscopy. Lab Invest 2002;82(10):1265–77.
66. Kirsch M, Schackert G, Salzer R, et al. Raman spec- 76. Abubaker MG, Taylor A, Ferns GA, et al. Differences
troscopic imaging for in vivo detection of cerebral in Raman spectra of aluminum treated brain tissue
brain metastases. Anal Bioanal Chem 2010;398(4): sample. Internet J Toxicol 2008;4(2).
1707–13. 77. Chen P, Shen AG, Zhao W, et al. Raman signature
from brain hippocampus could aid Alzheimer’s dis-
67. Amharref N, Beljebbar A, Dukic S, et al. Discrimi-
ease diagnosis. Appl Opt 2009;48(24):4743–8.
nating healthy from tumor and necrosis tissue in
78. Tay LL, Tremblay RG, Hulse J, et al. Detection of
rat brain tissue samples by Raman spectral imag-
acute brain injury by Raman spectral signature. An-
ing. Biochim Biophys Acta 2007;1768(10):2605–15.
alyst 2011;136(8):1620–6.
68. Krafft C, Belay B, Bergner N, et al. Advances in optical
79. Lakshmi RJ, Kartha VB, Murali Krishna C, et al. Tis-
biopsy–correlation of malignancy and cell density of
sue Raman spectroscopy for the study of radiation
primary brain tumors using Raman microspectro-
damage: brain irradiation of mice. Radiat Res
scopic imaging. Analyst 2012;137(23):5533–7.
2002;157(2):175–82.
69. Krafft C, Sobottka SB, Schackert G, et al. Near
80. Wolthuis R, van Aken M, Fountas K, et al. Determina-
infrared Raman spectroscopic mapping of native
tion of water concentration in brain tissue by Raman
brain tissue and intracranial tumors. Analyst 2005;
spectroscopy. Anal Chem 2001;73(16):3915–20.
130(7):1070–7.
81. Xie W, Schlucker S. Medical applications of surface-
70. Schrader B, Keller S, Löchte T, et al. NIR FT Raman enhanced Raman scattering. Phys Chem Phys
spectroscopy in medical diagnosis. J Mol Struct 2013;15(15):5329–44.
1995;348(0):293–6. 82. Muller M, Zumbusch A. Coherent anti-stokes Raman
71. Leslie DG, Kast RE, Poulik JM, et al. Identification of scattering microscopy. Chemphyschem 2007;8(15):
pediatric brain neoplasms using Raman spectros- 2156–70.
copy. Pediatr Neurosurg 2012;48(2):109–17. 83. Freudiger CW, Min W, Saar BG, et al. Label-free
72. Ong CW, Shen ZX, He Y, et al. Raman microspectro- biomedical imaging with high sensitivity by stimu-
scopy of the brain tissues in the substantia nigra and lated Raman scattering microscopy. Science 2008;
MPTP-induced Parkinson’s disease. J Raman Spec- 322(5909):1857–61.
tros 1999;30(2):91–6. 84. Freudiger CW, Xie XS. In vivo imaging with stimu-
73. Beleites C, Geiger K, Kirsch M, et al. Raman spec- lated Raman scattering microscopy. Opt Photon
troscopic grading of astrocytoma tissues: using News 2011;22(12):27.

Anda mungkin juga menyukai