Raman spectroscopy in neurosurgery
Saga Bergqvist
Engineering Physics and Electrical Engineering, master's level
2020
Luleå University of Technology
Department of Computer Science, Electrical and Space Engineering
Sammanfattning
Hjärntumörer kan drabba människor i alla åldrar, medelåldern för människor som lever med en hjärntumör är 60 år, men det är ett tillstånd som även drabbar barn och unga i stor utsträckning.
Hjärntumörer är den näst vanligaste cancerformen hos barn och är även den främsta orsaken till cancerrelaterad död i den åldergruppen. För att minimera skadorna på hjärnan är det viktigt att en tumör kan lokaliseras och tas bort så tidigt som möjligt. De metoder som används idag bygger framför allt på biopsi, där en del av tumören tas bort och undersöks av en histopatalog. Det är en process som tar lång tid och även påverkas av den mänskliga faktorn, det finns därmed ett behov av en metod som kan avändas in situ som ger ett resultat som inte påverkas av den mänskliga faktorn. En metod som har visat lovande resultat är fotosensibilisering med 5-Aminolevulinsyra (5-ALA). Desvärre har den metoden bara visat sig fungera bra för högmaligna tumörer hos vuxna.
Som ett komplement till fotosensibilisering har Ramanspektroskopi visat lovande resultat i tidigare genomförda studier.
Det här arbetet genomfördes för att undersöka användningen av Ramanspektroskopi som ett verktyg för diagnostisering av hjärntumörer. Som grund användes två tidigare genomförda studier där de undersökte Ramanband från biologiska markörer i hjärnvävnad som ändras i cancerogen vävnad. De undersökte även hur den biokemiska sammansättningen av hjärnvävnaden ändrades genom att jämföra intensiteten av olika Ramanband.
Ett mätsystem för Ramanspektroskopi designades och byggdes upp på Luleå Tekniska Univer- sitet där det även testades på vävnad från kött (fläsk och biff). Därefter transporterades mätsys- temet till Linköpings Universitet för att genomföra mätningar på sex olika vävnadsprov från fem hjärntumörer av olika malignitet. Baserat på en preliminär histopatalogisk bedömning var en av tumörerna högmalignt och de fyra andra tumörerna var antingen lågmalignta eller benigna. Två av proverna som undersöktes kom från den högmalignta tumören som även var fotosensibilierad med 5-Aminolevulinsyra, varav ett av proverna var belyst med blått ljus innan de Ramanspek- troskopiska mätningarna genomfördes.
Innan resultatet från Ramanspektroskopiska mätningarna analyserades behandlades datan med konventionella metoder i MatLab. I de resulterade spektrumen gick det att se tydliga Ramanband associerade med hjärnvävnad. Det gick även att se Ramanband associerade med 5-ALA i de två prover som var fotosensibiliserade och i det provet som var belyst med blått ljus innan de spektroskopiska mätningarna gjordes gick det även att se tydliga Ramanband associerade med hjärnvävnad. När resultatet analyserades gick det även att se spektra associerat med reducerat Neuroglobin (NGB) i ett av proverna. Sammansättningen av NGB är också någonting som ändras i cancerogen vävnad och skulle därför också kunna användas som en bilogisk markör för hjärn- tumörer i framtida studier.
När resultaten från den här studien jämfördes med de tidigare studierna indikerade den ena studien att två av vävnadsproverna kom från en högmalignt tumör och att de resterande fyra från lågmaligna eller benigna tumörer, vilket stämmer överens med den preliminära diagnosticeringen av tumörerna. När resultatet istället jämfördes med den andra studien stämde inte resultatet lika bra med den preliminära diagnosticeringen av tumörerna. Metoden presenterad av Zhou m.fl. in- dikerade att alla tumörer kom från lågmaligna eller benigna tumörer.
Slutsaten av det här arbetet är att Ramanspektroskopi skulle kunna användas som en metod för diagnosticering av hjärntumörer. Metoden skulle även fungera bra som ett komplement till fotosensibilisering med 5-ALA eftersom att det var möjligt att se Ramanband associerade med hjärnvävnad när vävnaden hade belysts med blått ljus.
Sammanfattning
Hjärntumörer kan drabba människor i alla åldrar, medelåldern för människor som lever med en hjärntumör är 60 år, men det är ett tillstånd som även drabbar barn och unga i stor utsträckning.
Hjärntumörer är den näst vanligaste cancerformen hos barn och är även den främsta orsaken till cancerrelaterad död i den åldergruppen. För att minimera skadorna på hjärnan är det viktigt att en tumör kan lokaliseras och tas bort så tidigt som möjligt. De metoder som används idag bygger framför allt på biopsi, där en del av tumören tas bort och undersöks av en histopatalog. Det är en process som tar lång tid och även påverkas av den mänskliga faktorn, det finns därmed ett behov av en metod som kan avändas in situ som ger ett resultat som inte påverkas av den mänskliga faktorn. En metod som har visat lovande resultat är fotosensibilisering med 5-Aminolevulinsyra (5-ALA). Desvärre har den metoden bara visat sig fungera bra för högmaligna tumörer hos vuxna.
Som ett komplement till fotosensibilisering har Ramanspektroskopi visat lovande resultat i tidigare genomförda studier.
Det här arbetet genomfördes för att undersöka användningen av Ramanspektroskopi som ett verktyg för diagnostisering av hjärntumörer. Som grund användes två tidigare genomförda studier där de undersökte Ramanband från biologiska markörer i hjärnvävnad som ändras i cancerogen vävnad. De undersökte även hur den biokemiska sammansättningen av hjärnvävnaden ändrades genom att jämföra intensiteten av olika Ramanband.
Ett mätsystem för Ramanspektroskopi designades och byggdes upp på Luleå Tekniska Univer- sitet där det även testades på vävnad från kött (fläsk och biff). Därefter transporterades mätsys- temet till Linköpings Universitet för att genomföra mätningar på sex olika vävnadsprov från fem hjärntumörer av olika malignitet. Baserat på en preliminär histopatalogisk bedömning var en av tumörerna högmalignt och de fyra andra tumörerna var antingen lågmalignta eller benigna. Två av proverna som undersöktes kom från den högmalignta tumören som även var fotosensibilierad med 5-Aminolevulinsyra, varav ett av proverna var belyst med blått ljus innan de Ramanspek- troskopiska mätningarna genomfördes.
Innan resultatet från Ramanspektroskopiska mätningarna analyserades behandlades datan med konventionella metoder i MatLab. I de resulterade spektrumen gick det att se tydliga Ramanband associerade med hjärnvävnad. Det gick även att se Ramanband associerade med 5-ALA i de två prover som var fotosensibiliserade och i det provet som var belyst med blått ljus innan de spektroskopiska mätningarna gjordes gick det även att se tydliga Ramanband associerade med hjärnvävnad. När resultatet analyserades gick det även att se spektra associerat med reducerat Neuroglobin (NGB) i ett av proverna. Sammansättningen av NGB är också någonting som ändras i cancerogen vävnad och skulle därför också kunna användas som en bilogisk markör för hjärn- tumörer i framtida studier.
När resultaten från den här studien jämfördes med de tidigare studierna indikerade den ena studien att två av vävnadsproverna kom från en högmalignt tumör och att de resterande fyra från lågmaligna eller benigna tumörer, vilket stämmer överens med den preliminära diagnosticeringen av tumörerna. När resultatet istället jämfördes med den andra studien stämde inte resultatet lika bra med den preliminära diagnosticeringen av tumörerna. Metoden presenterad av Zhou m.fl. in- dikerade att alla tumörer kom från lågmaligna eller benigna tumörer.
Slutsaten av det här arbetet är att Ramanspektroskopi skulle kunna användas som en metod för diagnosticering av hjärntumörer. Metoden skulle även fungera bra som ett komplement till fotosensibilisering med 5-ALA eftersom att det var möjligt att se Ramanband associerade med hjärnvävnad när vävnaden hade belysts med blått ljus.
Abstract
Brain tumors or brain cancer is a disease than affects people of all ages. The median age of a person living with a brain tumor is 60 years, it is however a disease that affects children and young adults in high grade. Brain cancer is the second most common type of cancer among children and is also the most common cause of cancer related death among this group. To ensure that the damages on the brain is as small as possible, it is important that a tumor can be diagnosed and removed as early as possible. Previous methods of diagnosis is based on biopsy where a part of the tumor is removed and examinated by a pathologist. This is a time consuming process that also is biased by the human factor, there is therefore a need for a method that can be used in situ with an unbiased result. One method that have shown great promise is photensitation with 5-Aminolevoluic acid (5-ALA). However, this method have shown to only work properly on tumors of high malignancy in adults. As a comlpiment to photosentisation, Raman spectroscopy have shown good promise in previous studies.
This study was conducted to investigate the use of Raman spectroscopy as a tool for in situ brain tumor diagnostics. The use of Raman spectroscopy was tested by comparing two previously performed studies, where they looked at a number of Raman bands from biological markers that are known to change in cancerous tissue as well as the intensity ratio between some Raman bands.
A measurement system for Raman spectroscopy was designed and built at Luleå University of Technology where the system were evaluated on tissue samples from conventional meat i.e. pork and beef to ensure that is was possible to achieve spectroscopic information of protein and lipid content in tissue. The measurement system was then transported to Linköpings University where the measurements on six brain tissue samples where performed. The samples came from five dif- ferent tumors of which one tumor was thought to come from a high malignant tumor based on preliminary histopathological analysis and four from low malignant or benign tumors. Two sam- ples where obtained from the high malignant tumor that was photosentisized with 5-Aminolevoluic acid and one of the samples where illuminated with blue light prior to the Raman spectroscopic measurements.
The spectroscopic data was pre-processed before analysis using conventional methods. The analysed spectra from the brain tissue samples showed presence of the Raman bands associated with brain tissue. It was also possible to see Raman bands associated with 5-ALA in the samples that had been photosentisized, however when the tissue had been illuminated with blue light it was also possible to see distinct Raman bands associated with brain tissue. One tissue sample also showed presence of reduced Neuroglobin (NGB). The composition of NGB is also known to change in tumorous tissue and could therefore be used in future work as a biological marker for brain tu- mors. When comparing the results obtained in this study with the two previously performed, one of the studies showed that two samples were from a tumor of high malignancy and the other from low malignant or benign tumors. This result was in accordance with the preliminary histopatho- logical assessment of the brain tissue samples. When comparing the results to the other study, the results where contradictory and indicated that all tissue samples where from low malignant or benign tumors.
The conclusion of this work is that Raman spectroscopy is possible to use as a tool for brain tumor diagnostics. It would be desirable to use this method in combination with 5-ALA staining since the Raman bands from brain tissue could be resolved when the tissue had been illuminated with blue light.
Acknowledgements
I would like to thank my examinor and supervisor Kerstin Ramser at Luleå University of Tech- nology for theoretical and technical guidance throughout the work, and also for the opportunity to work with this project in my Master thesis. Many thanks also to my co-supervisor Joel Wahl at Luleå University of Technology whom I conducted the measurements in cooperation with, Joel also provided technical guidance throughout the work. I would also like to thank the department of Experimental Mechanics at Luleå University of Technology for letting me use the John Fields laboratory to build the experimental setup as well Per Gren for his help in the lab.
Many thanks also to my supervisor Karin Wårdell at Linköping University for the opportunity to be a part of this project and write my thesis as a part of it. Thanks to Karin also for letting us use their lab to rebuild the measurement system and for guidance on how to perform measure- ments on tissue samples at Linköping University Hospital. Lastly, I would also like to thank the Department of Neurosurgey at Linköping University hospital for their cooperation and interest in the project, without them, the measurements could not have been performed.
Saga Bergvist, Luleå, April 24, 2020
Contents
1 Introduction 1
2 Background 2
2.1 Raman spectroscopy . . . 2
2.2 Anatomy of the human brain and brain cancer . . . 3
2.3 Biochemical composition of malignant and benignant brain tissue . . . 5
2.4 Photosensitation of brain tissue with 5 Aminolevolulinic acid . . . 6
2.5 Collection of Raman spectroscopic data . . . 6
2.6 Noise in Raman spectroscopic data . . . 7
2.6.1 Noise reduction and preprocessing . . . 8
3 Methodology 9 3.1 Description of approach 1 . . . 9
3.2 Description of approach 2 . . . 11
3.3 Raman bands for tumor diagnostics . . . 13
3.4 Materials and experimental setup . . . 13
3.4.1 Tissue samples . . . 13
3.4.2 Materials . . . 15
3.4.3 Experimental setup . . . 15
3.5 Calibration of measurement system . . . 17
3.6 Evaluation of measurement system . . . 17
3.7 Data analysis . . . 17
3.7.1 CCD image . . . 18
3.7.2 Cosmic ray removal . . . 18
3.7.3 Standardization of data . . . 18
3.7.4 Background subtraction . . . 18
3.7.5 Noise reduction . . . 19
3.7.6 Variance between measurement points . . . 19
4 Results 20 4.1 Evaluation measurements . . . 20
4.2 Measurements on brain tissue . . . 20
4.3 Variance of the spectroscopic data from the brain tissue samples . . . 26
5 Discussion 28 5.1 Diagnosis based on approach 1 . . . 28
5.2 Diagnosis based on approach 2 . . . 28
6 Conclusions and Future work 30
A Compilation of spectroscopic data from all brain tissue samples 33 B MatLab pre-processing function for the Raman spectroscopic data 36
Nomenclature
5-ALA 5 Aminolevulinic Acid CCD Charge Coupled Device CNS Central Nervous System CT Computed Tomography Cyt c Cytochrome c
in vitro on glass (test tube experiments) in situ on site
MRI Magnetic Resonance Imaging NAD Nicotinamide adenine dinucleotide NGB Neuroglobin
PET Positron Emission Tomography PNS Peripheral Nervous System PpIX Protoporphyrin IX
RR Resonance Raman SNR Signal to Noise Ratio Trp Tryptophan
US Ultrasound Sonography WHO World Health Organization
1 Introduction
Brain tumors can affect a person in all ages. According to the American Brain Tumor Association [1], over 700 000 people in the United States of America (USA) are living with brain tumors today and approximately 28 000 of them are kids. Although the median age for a person living with a brain tumor is 60 years, brain tumors are the secondary most common type of cancer among children of age 0-14 years and is the leading cause of cancer related death in this group. In addition to that, CNS and brain tumors are the third most comon type of cancer and also the third most common cause of cancer related death among adolescents of age 15-39 years.
A tumor or neoplasm can be classified as either benign or malignant. A benign tumor is nona- gressive, non growing and remains localized and is therefore seen to compose a lower health risk.
A benign tumor is of World Health Organization (WHO) grade I. In contradiction, a malignant or cancerous tumor grows and spreads agressively and therefore composes a larger health risk. [2]
Malignant tumors span from WHO grade II–IV.
The human brain consists of white and gray brain matter which in turn consist of water, pro- teins and lipids. When the normal brain tissue is compromised by neoplastic cells, the composition of the tissue in terms of the ratio between proteins and lipids changes. For example, an increase in malignancy of a glioma, the most common type of malignant brain tumor, is related to an overall reduction in lipids at the same time as the concentration of some lipids increase. The same type of behaviour can be seen when studying ratios and relative contents between other substances as well. This makes it possible to use analysis of the composition of tissue to determine the type and grade of tumors. [3]
The golden standard for tumor analysis and diagnostics is biopsy where a tissue sample is extraced and examined histopathologically. This method can not be perfomed in situ and suffers from long processing times along with variation according to the human factor - the pathologist’s eye and thereby varying levels of presicion. Conventional biopsy is often performed together with in situ imaging techniques to predict location and form of the tumor inside the cranium such as magnetic resonance imaging (MRI), computed tomography (CT) scans, ultrasound sonography (US) and positron emission tomography (PET). [4] One promising technique for more accurate diagnosis that can be performed in situ and in vitro is optical biopsy. This approach is based on studying the changes in biochemical composition of the tissue in terms of fluorescence profiles and Raman frequency peaks and their intensity ratios since the pathologic alteration of the cells that occur with cancer are related to changes in the biochemical composition of the tissue. [4] There are only a few biological molecules that respond well to fluorescence yielding broad, overlapping peaks in the emission spectrum. However, the majority of all biological molecules are Raman active with a vibrational spectrum that works as a spectral fingerprint. Therefore, Raman spectroscopy has potential for diagnosing brain tumors. One drawback with Raman spectroscopy is that the Raman signal is relatively weak compared to the intrinsic fluorescence that occurs intrinsically in biological tissues and therefore, the Raman signal may be difficult to resolve. [5]
2 Background
To be able to use Raman spectroscopy as a tool for diagnosing tumors it is of importance to understand the working principle of Raman spectroscopy as well as the pathological nature of the tumor and the changes that neoplastic cells inflict on the tissue. As mentioned by Zhou et al.
[4], the neoplastic cell changes the biochemical composition of the tissue and these changes can be related to the molecular fingerprint by studying the Raman spectrum of the tissue. To be able to properly do so, it is needed to further state these changes and how they can be related to the Raman spectrum of the tissue.
2.1 Raman spectroscopy
The Raman effect was discovered by the Indian physicist C.V. Raman in 1928. Photons from the incident light strikes the sample, as a result, most of the photons are scattered elastically i.e.
Rayleigh scattered with the same frequency and energy as illustrated in Figure 1. However in some cases, the molecule either gives or loses energy to the photon and the photon thereby experience a shift in it’s frequency and is inelastically scattered with Stokes or Anti-Stokes Raman scattering.
Figure 1: Sketch of possible types of photon scattering that can occur as a result of a photon interacting with a molecule. An inciding photon is absorbed by the molecule and is either scattered with the same wavevelgth i.e. Rayleigh scattered or with an wavelength shift proportional to the vibration of the molecule i.e. Anti-Stokes- or Stokes Raman scattered.
Since the Raman shift occurs from interaction with the molecules in the sample, the energy shift of the photon corresponds to the transition between initial and final state in the scattering molecule.
A certain Raman shift can thereby be related to specific transitions between different vibrational- and rotational states within the molecule and thus creates a spectrum with characteristic Raman lines for each particular molecular species. The Raman effect is however very weak since only around 1 out of 10’000’000 photons are Raman scattered. The intensity of a certain Raman peak is proportional to the number of scattering molecules and because of the fingerprint like spectra, the Raman effect can be a powerful tool in quantitative and qualitative analysis. [6] In a Raman spectrum, the intensity of the Raman scattered radiation is plotted as a function of the wavelength shift from the incident radiation such as for exapmle an exciting laser beam presented in inverse centimeters, [cm 1]. [7] The wavelength shift of Raman spectra are typically divided in two Regions, the fingerprint region spanning wavelength shifts below 1500 [cm 1] and the high
wavenumber region above 1500 [cm 1]. [8] An example of a Raman spectrum can be seen in Figure 2.
Figure 2: Example of a Raman spectrum where the horizontal axis represents the wavelength shift in [cm 1] and the vertical axis represents the intensity at each wavelength shift presented in arbitrary units (a.u.).
2.2 Anatomy of the human brain and brain cancer
The brain is a complex organ that works as the command center of the body. Along with the nervous system, the brain controlls our feelings, thoughts, senses, the majority of our actions as well as takes in information and helps us to apprehend the world aroud us. [9] A sketch of the human brain and it’s parts can be seen in Figure 3.
The nervous system consists of two parts, the central nervous system (CNS), that consists of the brain and the spinal cord, and the peripheral nervous system (PNS). The PNS works as a network throughout the body that collects information from the nerves and other supporting cells and communicates back to the CNS. The whole brain and spinal cord is covered by a connective tissue, the meninges, which can be divided further in three types of meningeal tissue: the arch- noid, dura mater and pia mater.[10] The CNS of the brain can be studied further in parts as the brain stem, the cerebral hemispheres and the cerebellum. The brain stem takes care of the body’s autonomic and reflexive processes such as heart rate and breathing, but it also has an important role in transferring information between the PNS and the brain. Next to the brain stem lies the cerebellum which is also known as "the little brain". The cerebellum coordinates movements and controls the balance. Above the brain stem and the cerebellum lies the two cerebral hemispheres or the cerebral cortex, which is often referred to as "the brain", which in it’s turn takes care of in- formation processing, memory, learning, sensory perception. It is also in the cerebral hemispheres that all decision-making takes place. [9]
The two hemispheres are nearly symmetrical, however, most of the brain’s language control and analytical tasks are taken care of in the left side while the right side takes care of more creative
Figure 3: Illustration of the human brain and it’s parts: the Cerebral hemispheres, the Cerebellum and the Brain stem connected to the spinal cord. All parts of the brain are covered by the Cerebral cortex as well as the Meninges - the brain’s connective tissue and are protected within the skull.
tasks. However, both hemispheres are primarily built up by the same type of cells called neurons.
Neurons are a special type of cells that are quite similar to the rest of the cells in the body, but they have extensions called axons and dendrites that allow the neurons to communicate with each other through synapses. A sketch of a neuron cell can be seen in Figure 4. The different parts of the neurons in it’s turn build up two types of brain matter that are called gray and white brain matter.
The cell bodies and dendrites of the neurons forms the gray matter while the axons, covered in a fatty white insulation called myelin, forms the white matter. [9]
Figure 4: Illustration of a neuron cell showing the cell body with an Axon covered in Myelin ending in the Axon terminal. The Axon terminal connects to another cell and makes it possible for the Neuron cell to communicate by sending signals through the Axon.
In a healthy brain, the parts of the CNS works together without problems and allows all brain functions to work properly. This means that if for example a tumor starts to grow in the CNS, the brain functions can be affected and cause disruptions in for example a persons speech,
movements or thought processes. [11] Typically, brain tumors are divided in two main types:
Primary and secondary depending on where they originate. Primary brain tumors are those tumors that originate in the brain. They are further divided in low-grade tumors that typically are slow growing and high-grade tumors that are more likely to grow rapidly. Primary brain tumors can be divided further in Gliomas and Non-glioma tumors. Gliomas are one of the most common types of brain tumors. Gliomas are tumors that most likely start growing from Glial cells that are one sort of supportive cells in the brain, Gliomas are graded depending on how agressive or fast growing they are likely to be and divided in types: Astrocytoma, Oligodendroglioma, Ependymoma and Brain stem glioma. Non-glioma tumors originate from other cells in the brain than Glial cells and are divided into different types such as Meningioma and Medulloblastuoma to mention a few.
Secondary brain tumors or brain metastases are tumors that starts growing in another part of the body and then spread to the brain. Secondary brain tumors are the most common type among adults. [11]
2.3 Biochemical composition of malignant and benignant brain tissue
As mentioned, brain tissue contains a large amount of both proteins and lipids, with a higher lipid concentration in white brain matter (⇠ 15%) than in grey brain matter (⇠ 4%). Due to the change in concentration and composition of both lipids and proteins that occur in necrosis (the transition from normal brain tissue to neoplastic brain tissue), proteins and lipids may function as biological markers for Raman spectroscopy. [3]
Proteins are built of so called amino acids which are linked together by peptides or amide bonds in different ways to form proteins. The structure of a protein is characterized by the combina- tion of amino acids which is called the primary structure, as well as the arrangement in the three dimensional space for a local part of the protein, also called the secondary structure. The most common types of secondary structures are ↵ helix and sheet. [12] Proteins can be linked together with four different types of amide bonds of which two are Raman active: the amide I bond from C=O stretching and the amide III bond from N-H and C-N stretching. In addition, the skeletal vibrations of a protein, C-C stretching vibrations or skeletal modes, are also Raman active. These bonds for different structures can be studied in a Raman spectrum in the fingerprint region with peaks corresponding to Table 1. [13]
Table 1: Location of Raman peaks in [cm 1] corresponding to amide I (C=O), amide III (C-N,C-H) bonds and skeletal vibrations (C-C) for different protein secondary structures.
Amide I Amide III Skeletal
Secondary structure raman shift [cm 1] raman shift [cm 1] raman shift [cm 1]
↵ helix 1654-1662 1258-1304 935-945
sheet 1665-1680 1227-1247 1002
Unordered 1654-1685 1235-1270 1100-1110
There are also a number of lipids that are Raman active and due to their content of long non- polar acyl chains, many lipids generate a strong Raman signal which makes lipids suitable markers in pathology. The strong Raman signal from lipids is related to their presence of hydrocarbon chains with peaks that can be observed both in the fingerprint- and the high wavenumber region.
The bands from CH2scissoring, CH3twisting and from C-C stretching vibrations can be observed in the fingerprint region and the band from C-H stretching can be observed in the high wavnumber region as seen in Table 2. [14]
The most common lipids in brain tissue are cholosterol and phospholipids. Previous studies show that necrosis in human brain tissue can be associated with an overall reduction in total lipids at the same time as the concentration of some lipids increase, for example the phospholipid phos- phatidylcholine (PC). It has also been shown that the concentration of cholesterol esther increase up to 100 times in gliomas compared to benign tissue. [3]
Table 2: Location of Raman peaks in [cm 1] corresponding to CH2scissoring, CH3twisting, C-C and C-H stretching in lipids.
Raman shift [cm 1] Raman shift [cm 1] Hydrocarbon chain fingerprint region high wavenumber region
CH2 scissoring 1400-1500 -
CH3 twisting 1250-1300 -
C-C stretching 1050-1200 -
C-H stretching - 2800- 3100
2.4 Photosensitation of brain tissue with 5 Aminolevolulinic acid
Photosensitizers are substances that either directly or indirectly induce fluorescence in the stained tissue either by applying the substance topologically, intravenously or orally. The most common photosensitizers are bacterichlorin, chlorin and Porphyrin. 5 Aminolevoluinic acid (5-ALA) is an indirect photosensitizer that triggers production of Protoporphyrin IX (PpIX) in the cells. 5-ALA is used as a photosensitizer in neurosurgery due to the the optimized uptake in brain tumor cells. [15]
By illuminating the photosentisized tissue with blue light (390-410 nm) and observing the tis- sue, the healthy tissue appears blue as a result of the exciting light and the tumorous tissue appears pink or red as a result from the PpIX fluorescence. [15]
Photosensitation with 5-ALA has been succesful on several types of brain tumors with gliobas- toma (agressive grade IV tumor) as the main target. However, there has been some reports of low grade tumors and metastases where staining with 5-ALA have not given rise to fluorescence in the tumorous tissue. [15]
Raman bands of Ala peptides such as Ala5 (5-ALA) are presented in Table 3. Peptides are short strings of Amino acids and therefore, Ala peptides usually show high intensities of Raman bands typically associated with proteins such as from Amide I, II and II. [16]
Table 3: Raman bands of Ala peptides in the fingerprint region. [16]
Raman shift [cm 1] Assignment
1244 Amide III3
1302 Amide III2
1365 C↵-H symmetric bending vibration 1388 C↵-H symmetric bending vibration 1550 Amide II, CN stretching, NH bending 1657 Amide I, C=O stretching
2.5 Collection of Raman spectroscopic data
In order to analyze the light containing the back scattered Raman signal from the tissue sample needs to be collected. If the Raman signal was induced with an exciting laser beam, the laser light needs to be filtered out with some kind of optical filter in order to resolve the Raman signal from the light. The filter needs to be chosen to match the wavelength of the exciting laser beam.
To collect the Raman signal, the backscattered light is typically focused into an optical fiber that transports the signal to a spectrometer where the light is dispersed using gratings and then pro- jected onto a detector consisting of a 1 or 2 dimensional array of pixels. In that way, the Raman signal is collected so that the detector registers the signal resolved in wavelength shifts. The first pixel on the detector will collect photons with the smallest wavelength shift and the next pixel will collect the photons from the next spectral position i.e. with the next wavelength shift and so on.
The last pixel will collect photons with the greatest wavelength shift in the spectral window. In that way, the photon intensity for each wavelength shift in the signal is collected in a spectrum.
[17] A sketch of a general Raman setup can be seen in Figure 5.
Figure 5: Sketch of a Raman setup, where the laser beam is illustrated with a green line and visible light as a dashed line. The light scattered from the sample is focused into an optical fibre and led into a spectrometer with a detectror.
A charged coupled device (CCD) is a very commonly used detector in digital imaging. The CCD’s surface is built up by pixels upon a semiconducting surface, the CCD then converts photons from incident light to photoelectrons on the detector surface which are collected after each exposure of the detector surface in the readout process as illustraded in Figure 6. [18] When an incident photon approaches the detector surface, the photon is absorbed in a pixel and an electron is released in the semiconductor creating an electron-hole pair. The electrons are then transferred down each column to an output row in the CCD by shifting each pixel creating a row of charges into an output amplifier and afterwards, the charge is converted to a voltage which can be measured and recorded. This process is repeated until the information in all rows of the CCD are collected which makes the CCD ready to use for a new detection. [19]
2.6 Noise in Raman spectroscopic data
The noise in Raman spectroscopic data are typically stated to originate from five independent sources resulting in a total variance 2.
2= x2+ 2b+ 2d+ 2f+ r2 (1)
Where 2xis the variance of the measured photon signal, x, and is therefore a natural limit of the accuracy of the Raman spectrum. b2is the variance of background fluctuations generated by fluctuations in the laser beam or in the temperature of the detector as well as of a fluorescence signal. d2 is the variance of the dark current in the CCD detector, 2f the variance of the 1/f noise and r2, the variance in readout noise that occurs when the charge due to the photon stream is transformed into a voltage signal that is sampled to a digital signal. The main source of noise in this process is the inherent noise from the amplification of the signal generated in the detector.
Usually, the influence from this noise component is much lower than from different sources in the system. Analysis of Raman spectroscopic data is also limited by the collected signal from the
Figure 6: Illustration of Charge Coupled Device (CCD) during a readout process. The pixel data are transferred down to the readout row one column at a time and is amplified and converted do a DC-signal before it can be processed to an image.
flourescent background. Without consideration of the flourescence, the analysis is limited to only the strongest Raman bands since the Raman signal is many times weaker in intensity then the fluorescence. It is important to consider the interference of fluorescence especially when studying biological specimen that contain many chemical compounds with intense intristic fluorescence sig- nals.[20] Raman spectroscopic data also suffers from influence from cosmic rays i.e. particles of high energy that interacts with the detector. Cosmic rays are visible in Raman spectra as narrow peaks of high intensity that appears randomly at different locations on the detector surface and since these peaks interfers with the Raman signal, it is desired to remove them. [21]
The intensity of various noise sources can vary and therefore they influence the acquired Raman spectrum in different ways. Therefore, many different methods can be used to reduce their influ- ence either by pre processing of Raman spectroscopic data or modifications of the experimental the setup.
2.6.1 Noise reduction and preprocessing
The simplest method to reduce background noise in Raman spectra is to use a longer aquisition time to improve the SNR (Signal to Noise Ratio). By doing so, the number of photons from the signal is increased at the same time as the collected background noise is constant. However, regarding noise of type 1/f, an increased acquisition time can in some cases lead to the opposite result. Typically, the main source of noise in Raman spectroscopic data is the dark current in the CCD detector. To reduce the influence of noise due to dark current, the CCD detector is usually cooled, however the control unit of the cooling device can give rise to 1/f like fluctuations.
[20]To reduce influence from other noise sources, the spectroscopic data is usually pre processed with some signal processing methods. One method frequently used for filtering is the Savitzky and Golay algorithm that smooths the spectrum to reduce noise influence, which also can lead to some modification of the spectroscopic peaks. [20]
3 Methodology
To perform this study, a Raman system was built and evaluated in order to perform in situ mea- surements on tissue from brain tumors. The spectroscopic data was then analyzed and compared to previous research in order to diagnose the tissue and determine whether it was benign or malign and in that case also the grade of malignancy. In order to do so, two methods conducted by dif- ferent teams have been studied closer in this thesis to compare the results achieved using different methods to diagnose tissue. In addition to that, the setup was first evaluated on tissue samples from conventional meat i.e. minced beef, pork and chicken.
3.1 Description of approach 1
In a study called Novel strategies of Raman imaging for brain tumor research by Imiela Anna et al.
[22], they showed that it is possible to distinguish low- from high grade brain tumors by studying the tumor metabolism. By understanding the molecular mechanisms that cause the metabolic changes with neoplasm, Raman spectroscopy could be used as a real time method to study tumor metabolism and diagnose tumors.
Imiela Anna et al. showed that a combination of studying the overall ratio of lipids to pro- teins, the lipid alterations in malignant medulloblastoma (primary central nervous system (CNS) tumor of grade IV [23]) and low grade gliomas (tumors that begin in glial cells [23]), the content of nucleic acids i.e. the DNA to RNA turnover in tumors, as well as the configuration of -sheet and ↵-helix protein structures, can be used to distinguish necrotic from healthy tissue and also the grade of the tumors malignancy from WHO grade I-IV. In this study, 16 samples from tumorus (WHO grade I,II and IV) as well as one sample from non-tumorous fresh human brain tissue was investigated. To locate and analyse the tumorous tissue, both MRI- and Raman imaging were used. The spectroscopic data was obtained using a confocal Raman microscope with an excita- tion wavelength of 532 nm from a Nd:YAG laser focused with an objective of 40x magnification. [22]
In Table 4, a comparison between Raman bands observed in the study when measuring the Raman spectrum from normal brain tissue is presented together with average locations of that Raman band as well as their tentative assignments. From Table 4, it is possible to emphasize that the peaks at 2845 and 2854 [cm 1] contains information about the vibrational features of lipids and the peak around 2930-2940 [cm 1] are assigned to proteins. Therefore, the ratio between the intensity of these peaks, i.e. II29302945, can be used to calculate the overall ratio between lipids and proteins in the tissue sample. By studying this ratio for all tissue samples, Anna et al. could see that that in general, the ratio was significantly higher in all grades of brain tumors compared to normal brain tissue with a lipid to protein ratio of 1.456 ± 0.016. [22]
Anna et al. also showed that high grade tumors can be distinguished by their protein structure due to a decrease in ↵-helix conformation and an increase of -sheet conformation compared to normal tissue. This can be observed in the spectral region at several locations presented in Ta- ble 4. Anna et al. show results of a decreased intensity of the peak around 1658 cm 1 which is consistent with a decrease of amide I ↵-helix. They also show that the intensity of the peak from amide II around 1584 cm 1increases. Similarily, the intensity of the peak around 1230-1250 cm 1, corresponding to amide III -sheet, was enhanced in high grade tumors compared to normal tissue in the same time as the intensity of the peak around 1270-1280 cm 1, corresponding to amide III ↵-helix, was decreased. With these observations, Anna et al. demonstrated that high grade tumors can be related to a modification of ↵-helix to -sheet structures of the proteins.
In most of the studied high grade tumors, Anna et al. could also observe a significant increase in intensity of the peak around 751 cm 1 corresponding to nucleic acids, which is consistent with an increased content of nucleic acids in the high grade tumor compared to normal CNS tissue.
This correlates with the previous observation of the increased intensity of the peak around 1584 cm 1 corresponding to amide II and nucleic acids. These findings shows the DNA/RNA turnover in the tumor and are therefore also a way to distinguish tumorous from non-tumorous tissue.
Table 4: Observed locations of Raman bands in [cm 1] compared to average locations in normal brain tissue and their assignments. [22]
Observed raman shift Average raman shift Assignment
[cm 1] [cm 1]
721 729 Phospholipid
751 Nucleic acids, Trp (Trypthophan)
883 880 Tyr, Lipids/Carbohydrates/Collagen C-C-N+,
C-O-C ring, C-C
958 935 Hydroxyproline/Collagen backbone CH=CH bending
1004 1004 Phenylalanine symmetric (sym.) ring breathing of protein
1064 1068 Lipids/Collagen C-C stretching
1091 1096 Phospholipids, O-P-O sym. stretching,
P=O sym. from nucleic acids/cell membrane phospholipids
1080-1158 1158 Proteins (C-C,C-N stretching),
P=O sym. from nucleic acids and phospholipids
1189 1199 C-C6H5 Phenylalanine, Trp
1238 1240 Phospholipid, O-P-O antisym. stretching, Amide III -sheet 1248 1220-1285 Nucleic acids (Try, Ala)/Proteins(Amide III -sheet
or random coil), Lipid, phospholipid =C-H bending 1304 1304 Lipids, phospholipids, C-H2 twisting, collagen
protein (amide III), DNA
1437-1444 1444 Fatty acids, triglycerides, CH2 or CH3 deformations
1453 1461 Proteins C-H wag, CH2 or CH3deformations, phospholipids, CH2scissoring
1558 1556 Proteins, amide II/amide II -sheet
1584 1586 Amide II, aromatic amino acids in proteins, nucleic acids 1658 1655 Unsaturated fatty acids, triglycerides (C=C) stretching,
amide I ↵-helix
1732 1743 (C=O) stretching, triglycerides
2845/2854 2854 Fatty acids, triglycerides, CH2 sym. stretching
2888 2888 Lipids, C-H2 antisym. stretching
2931/2940 2935 Proteins/lipids, CH3sym. stretching
3009 3008 Lipids =C-H stretching
3067 3060 Nucleic acids/proteins C-H aromatic
When studying the lipid content in the high grade tumors, Anna et al. found a significantly lower intensity of the peaks around 1437-1444 and 1453 cm 1 in most of the studied high grade tumors, corresponding to the content of CH2 and CH3 deformations among others, compared to normal CNS tissue. In correlation to that, Anna et al. also could observe a decreased intensity for the peak around 2845-2854 cm 1 in the high wavenumber region which indicates the same behaviour. The decreased intensity can be related to a decrease in saturated CH2bonds in lipids that occur as a result of the alteration in lipid metabolism in a tumor compared to normal CNS tissue.
For the low grade tumors that were studied, Anna et al. saw that the spectra from a low grade tumor was similar to the ones from normal CNS tissue. The spectra were partly overlapping. They could however see some differences that could be possible to use to distinguish low grade tumors from normal CNS tissue. For example, the found signs of enhaced levels of trythopan with a peak around 1338 cm 1. However, for low grade tumors, Anna et al. could not observe any signs of
↵-helix to -sheet modification of protein structures as they could for high grade tumors.
In summary, to follow the methodology conducted by Anna et al., the following biological markers will be examined in the spectroscopic data collected in this study to diagnose the tissue:
• The intensity ratio of the peaks at 2930 and 2854 [cm 1]as II29302854 corresponding to the lipid to protein ratio in the sample in comparison to the ratio in normal CNS tissue of 1.456 ± 0.016 will be studied. A significantly higher ratio is an indication of a high grade tumor.
• A decreased intensity of the peak around 1658 [cm 1], from amide I ↵-helix, in addition to increased intensities of the peaks around 1584 and 1230-1250 [cm 1], from amide II and Amide III -sheet, indicates a high grade tumor since these alterations correspond to the modification of ↵-helix to -sheet structure in proteins.
• Increased intensities of the peaks around 751 and 1584 [cm 1]from nucleic acids correspond- ing to DNA/RNA turnover in the tissue indicates a high grade tumor.
• The intensity of peaks around 1437-1444, 1453 [cm 1]and 2845-2854 [cm 1]from CH2and CH3bonds in lipids. A decreased intensity i.e. a decrease in saturated CH2bonds compared to normal CNS tissue can be an indication of a high grade tumor.
• Enhanced intensity of the peak around 1338 [cm 1]from trythopan at the same time as no signs of ↵-helix to -sheet modification of protein structures can be observed could be an indication of a low grade tumor.
3.2 Description of approach 2
As the second approach to diagnose brain tissue, a study called Human brain cancer studied by resonance Raman spectroscopy conducted by Yan Zhou et al. [4] was be used. In this study, they showed that it is possible to distinguish cancerous from normal brain tissue with an optical biopsy approach. This approach is based on studying the molecular fingerprints from tissues with Raman spectroscopy and intensity ratios in the collected spectra since alterations can be related to patho- logic changes in cells and tissue that occur with cancer. In particular, alterations in lipid content, ratio between vibrations from methyl ( CH3) and methylene ( CH2 )as well as altered levels of cytochrome c due to changes that occur in cancerous compared to normal or benign brain tissue was investigated further in this study.
Zhou et al. studied six types of brain tissue in vitro with a confocal micro-Raman system and an excitation wavelength of 532 nm over a spectral region of 500 to 4000 cm 1. With this setup, Zhou et el. investigated the use of RR (Resonance Raman) spectroscopy to distinguish malignant from normal brain tissue. The six different brain tissue samples came from tumors diagnosed with immunohistochemistry and histopathology after operation. Of the six types of tissue, Zhou et al.
showed promising results from three types of tissue: malignant, and benign meningioma as well as normal meningeal tissue. This indicates that RR spectroscopy can be used as an in vitro techique Zhou et al. performed two separate studies, each on three types of brain tissue. The first study was performed on tissue from three different types of tumors: glioblastoma multiforma of grade IV, benign acoustic neuroma and benign pituitary adenoma. In the RR spectra from gliobastoma multiforma showed enhanced peaks at 752, 1004, 1172, 1337, 1358, 1547, 1587, 1606, 2896 and 2938 cm 1. Of which, the peaks at 1547, 1587, 1606 and 1004 are characteristic Raman bands from amide II and collagens of type I and IV.
When studying cancerous, normal and benign meningeal brain tissue, Zhou et al. were able to find enhanced peaks at 750, 1004, 1156, 1358, 1548, 1587, 1605 and 1639 cm 1 in the fingerprint region as well as between 2800-3000 cm 1 in the high waveumber region. [4] The biochemical assignment of these peaks are presented in Table 5.
Zhou et al. could also observe alterations in the relative intensity between the peak at 1587 cm 1 from Cytochrome c and mithocondria, and the peak at 1605 cm 1 from Phenalanine and Tyrosine. [4] The intensity ratio as II15871605 was calculated to 0.97 for the cancerous tissue and 0.89 and 0.88 for Benign and normal tissue respectively. This alteration can be explained by the cell alterations that occur with cancer. In cancerous tissue, Cytochrome c might be released due to
Table 5: Locations of Raman bands in [cm 1] and their assignments found in the study Human brain cancer studied by Resonance Raman spectroscopy performed by Zhou et al. [4]
Raman shift [cm 1] Assignment Remark
754 CH2 rock, sym. breathing Trp, mithocondria, Cyt c (Cytochrome c)
1004 Sym. CC aromatic ring breathing Phenylalanine, Collagen I and IV 1088 CC stretch, CC skeletal stretch trans, Protein, phospholipid,
P O2 symmetric glycogen Collagen IV and I
1156 C=C stretch -carotene
1301 Amide III, (N-H) bending, ↵-helix, Collagen IV and I (C-N) stretch, (CH3) bending
1358 CH3-(C=O) Trp, mithocondria, NADH
(Nicotinamide adenine dinucleotide + H) 1548 Amide II, in plane (N-H) bending, Trp, Cyt c, NADH
(C-N) stretch
1587 C-C stretching, C-H bending Trp, mithocondria, NADH 1605 CO stretching, C=C bending Phenylalanine, Tyrosine
1639 Amide I ↵-helix Proteins
2850 (CH2)stretch Poly methylene chain, fat
2891 (CH2Fermi resonance) stretch Poly methylene chain 2934 (CH3, Fermi resonance) stretch Proteins, fat
the mutation of the mithocondrial membrane that is caused by cancer cells.
In the high wavenumber region, the intensity ratio of the strong Raman band at 2935 cm 1from Methyl (-CH3) symmetric stretching in comparison to the the peak at 2880 cm 1from Methylene (CH2) also showed alterations when comparing the spectra from cancerous, benign and normal tissue. The intensity ratio from methyl/methylene was observed to decrease with cancerous tissue and was calculated as calculated as II29352880 to 1.25, 1.47 and 1.79 for cancerous, benign and normal tissue respectively. This phenomena can be explained with a change in molecular conformation order due to the abnormal and fast cell growth in cancer cells. Cancerous cells grows uncontrollably and invades normal cells. Due to the fast growth of cancer cells, a tumor takes up more space in the brain than normal tissue and can therefore interfere with the brain’s functions. Healthy cells have a more stable metabolism and biochemistry than cancerous cells and therefore, normal cells have a higher molecular conformation which could cause alterations as for example the variations in the CH3to CH2ratio. Zhou et al. showed that the higher order coefficient in normal meningeal tissue may suggest the use of a statistical method with a "disorder or order molecular conformation coefficient" to separate malignant from normal meningeal tissue. [4] Also the width of the board profile for the broad peak between 2800-3000 cm 1was found to increase in cancerous tissue and was calculated to 1, 0.98 and 0.95 for cancerous, benign and normal tissue respectively.
In the malignant meningeal tissue, Zhou et. al. could also observe enhanced peaks from amides as well as collagens of type I and IV in the RR spectra at 1088 and 1302 cm 1respectively. In spec- tra from all three types of meningeal tissue, a strong enhanced peak from Amide II at 1547 cm 1 could be observed. This gives an indication that the excitation wavelength of 532 nm resonates with the Amide II mode, this band can therefore be used as a indication of the RR frequency. [4]
In order to follow the study conducted by Zhou et al. the following biological markers will be studied in order to distinguish normal from cancerous and benign tissue.:
• The ratio of Cytochrome c compared to Tyrosine and Phenalanine as the intensity ratio II15871605, which should be higher in cancerous compared to benign and normal tissue.
• The intensity of the Raman bands at 1088 and 1301 cm 1 from proteins (amides) and Col- lagen I, IV which should be enhanced in spectra from cancerous tissue.
• The ratio of CH3 to CH2as the intensity ratio II29352880 which should be significantly smaller in cancerous tissue than in normal and benign. The intensity ratio should also be smaller for benign tissue compared to normal due to the cell alterations that occur in a tumor.
3.3 Raman bands for tumor diagnostics
The two studies by Anna et al. [22] and Zhou et al. [4] were conducted with slightly different Raman setups and settings, a summary of the setups used in the two studies can be seen in Table 6 together with the number of samples used in both studies.
Table 6: Summary of Raman setups and brain tissue samples used in Approach 1 and 2.
Approach 1 Approach 2
Laser wavelength 532 nm 532 nm
Laser power 10 mW 0.9 mW
Exposure time 0.5 s 60 s
Number of samples 17 6
Sample size 16 µm thick 10x5x2 cm
Condition of tissue samples Thawed from frozen Thawed from frozen
Based on the two approaches, the intensity at a number of Raman bands should be studied in order to diagnose brain tumors as proposed in the two approaches. A summary of the Raman bands are presented in Table 7 along with their assignment.
Table 7: Summary of Raman bands to study for brain tumor diagnostics according to Zhou et al.
[4] and Anna et al. [22].
Approach 1 Approach 2
Raman bands [cm 1] Assignment Raman bands [cm 1] Assignment 751 Nucleic acids, Trp
1248 Nucleic acids, Proteins, 1088 Lipids
1301 Amide III, Collagen IV, I
1338 Trp
1444 Fatty acids, Triglycerides 1453 Proteins, Phospholipids 1584 Amide II, Aromatic amino
acids, Nucleic acids
1587 Trp, Mithocondria, NADH 1605 Phenylalanine, Tyrosine 1658 Unsaturated fatty acids,
Triglycerides, Amide I 2854 Fatty acids, Triglycerides
2880 Poly methylene chain, Lipids 2930 Proteins, Lipids
2935 Proteins, Lipids In addition, a number of ratios between peak intensites where to be analysed that are presented in Table 8 along with their assignment.
3.4 Materials and experimental setup
3.4.1 Tissue samples
The biological tissue samples of human brain tissue were brought directly from the operation ta- ble to the Raman measurement station where they where studied in room temperature in fresh
Table 8: Summary of ratios between peak intensities to study for brain tumor diagnostics according to Zhou et al. [4] and Anna et al. [22].
Approach 1 Approach 2 Intensity ratio Intensity ratio
I1587/I1605
I2930/I2854
I2935/I2880
condition. All samples where obtained during surgical operations at the University hospital in Linköping, Sweden. The samples were brougth back to the opration room again as soon as the measurements where finalized. The patients gave informed written content in participating in the study Dnr 2015/138-32. A total of 6 samples from 5 different tumors of unknown malignancy and varying sizes were studied. Two different samples where obtained from the same tumor stained with 5 Aminolevulinic Acid (5-ALA), with a standard dose of 5-ALA for fluorescence guided surgery of 20 mg/kg body weight, and one of the two samples was illuminated with blue light prior to the measurements. A full list of the brain tissue samples that were studied can be seen in Table 9 together with the preliminary assessment of the tumor.
Table 9: Summary of the investigated samples, their preliminary assessment and the number of measurement points for each sample as well as the number of outlier measurement points for each sample.
Number of Number of outlier Preliminary
Sample measurement points measurement points tumor assessment Note
1 5 1 High malignant 5-ALA stained
5-ALA stained
2 6 0 High malignant and illuminated
3 9 2 Low malignant Metastasis
4 7 1 Low malignant
5 8 0 Low malignant Pituitary
6 6 1 Low malignant Pituitary
For each tissue sample, spectra in both the high- and low- wavenumber region were collected at 2-4 points respectively. In total, spectra were collected from 5-9 measurement points for each tissue sample during an exposure time of 5x50 s at each point with an exciting laser beam of 532 nm set to a laser power of 12,5 mW.
The tissue samples from conventional meat used for evaluation of the Raman setup were ob- tained from a grocery store and can be seen in Figure 7. The samples where kept chilled in a refrigerator and where examined fresh and untreated of room temperature. For each tissue sam- ple, spectra was collected from one measurement point with an exposure time of 5x30 s with an exciting laser beam of 532 nm and a laser power measured to 12,5 mW before entering the system.
Figure 7: Photos of tissue samples from conventional meat on top of the microscope objective.
The sample to the left is entrecot illuminated from below with white light and the sample to the left is pork belly illuminated from below with the laser beam.
3.4.2 Materials
The materials used in the experimental setup is listed below. Along with this, a number of miscellaneous equipment such as clamps, bolts and rods where used to build the setup as well as a breadboard used as base. The laser used as an excitation beam was a green laser of wavelength 532 nm with a power of 12,5 mW measured at the exit of the laser beam. Both the spectrometer and Guppy camera are equipped with a CCD detector to collect image data. The spectroscopic data was preprocessed using the numeric tool MatLab R2017a (Mathworks).
• Green laser: 532 nm Y d : NV O4
• Guppy camera for imaging, Mako U503B (Allied Vision, Germany) Detector size: 2592x1944 pixels
• Spectrometer: Schamrock 303i (Andor Technology, Ireland)
• Microscope with:
Objective: 40x (Olympus, Japan)
Manual xy-stage (Merzhäuser Gmbh, Germany)
• LED-ring for white light illumantion of sample
• Lenses,mirrors and optical fibre for focusing and guiding light
• Edge filters (Semrock, USA) 3.4.3 Experimental setup
The experimental setup was built at the John Fields Laboratory at Luleå University of Technology in Luleå. A picture of the setup can be seen in Figure 9 and a sketch of the setup can be seen in Figure 8.
Figure 8: Sketch of the experimental setup used to perform the scectroscopic studies. The dashed line represents the optical path for the visible light and the green, solid line represents the laser beam path.
The laser beam was guided through the microscope objective by mirrors and focused on the tissue sample by the objective. The Raman scattered light was then led back through the objective into the optical fibre through an edge filter in order to filter out any light from the laser beam.
The Raman signal was then led through the optical fibre into the spectrometer and collected on a CCD detector, and the resulting spectra was studied in a computer. Using the same setup with the removable mirror blocking the laser beam, a picture of the tissue sample was obtained by illuminating the sample with a light source from below. The Removable mirror then guided the resulting image onto the Guppy cameras and the image was viewed on a computer screen.
Figure 9: Photo of the experimental setup used to perform the spcectroscopic studies.
3.5 Calibration of measurement system
In order to calibrate the spectrometer, measurements were performed on polysterene (Petri dish).
Polysterene has a distinct Raman band at 1000.4 [cm 1]. By collecting a Raman spectra of Polystyrene and correcting the offset in the spectrometer so that the observed peak is located as close to the accurate value as possible, the observed peaks will be located as close to their true value as possible. The sample of Polystyrene used for calibration of the spectrometer was a conventional Petri dish shown in Figure 10 and the resulting Raman spectrum can be seen in figure 11.
3.6 Evaluation of measurement system
The measurement system was evaluated by taking Raman spectra of conventional meat from the grocery store. Both beef (Entrecot) and pork (Pork belly) were investigated due to the high content of proteins and fatty acids. [24]
3.7 Data analysis
The spectroscopic data was pre-processed in MatLab R2017a using conventional methods to com- pensate for cosmic rays, differences in gain over the CCD detector, electric noise and other statistic alterations in the signal. The data was pre-processed to be able to isolate the Raman signal from the spectra collected i the measurements. The noise was removed by Savitzky and Golay filtering and the flourescent background was removed with a baseline subtraction. Cosmic rays was also removed from the spectroscopic data. The MatLab function used for pre processing can be seen in Appendix B.
Figure 10: Photo of the Petri dish made of Polystyrene on top of the microscope objective. The Petri dish was used for calibration of the Raman spectrometer.
3.7.1 CCD image
The full image from the CCD was obtained in order to reconstruct a spectrum with as little influence as possible from parts of the CCD that were not properly illuminated. In order to do so, the spectra from each channel was observed and only the ones that included a signal were used in the reconstruction.
3.7.2 Cosmic ray removal
To compensate for cosmic rays i.e. photons of very high intensity in the spectroscopic data, a two dimensional, second difference detection of cosmic rays with inpainting by a weighted local average of size 5x5 pixels (ignoring the cosmic ray) was used, inspired by Shulze and Turner. [21]
3.7.3 Standardization of data
The spectroscopic data in the CCD-image was standardized by subtracting the mean value from each observation y, i.e. row of pixels in the image, and dividing each observation by its standard deviation to compensate for differences in gain and illumination throughout the CCD detector.
The standardized vector for one observation y was obtained as ystand as stated in Equation (2) where ¯y is the mean value of each observation y and sy the standard deviation of that observation.
ystand= y y¯
sy (2)
3.7.4 Background subtraction
When using Raman spectroscopy for analysis of biological tissue, one challenge is to distinguish the Raman signal from the fluorescence that also is obtained when illuminating the tissue. Since
Figure 11: Analysed spectrum from Polystyrene used for calibration of the CCD detector collected during 10 s with a laser power of 0,5 mW.
the magnitude of the fluorescence often is many times higher in intensity compared to the Raman signal, it is of importance to minimize the influence of the fluorescence to be able to reveal the Raman signal. In this study, an iterative method of polynomial curve-fitting was used, as described by Lieber and Mahadevan-Jensen. [25] When applying a variant of this method, a polynomial of degree 3 was fitted to the baseline of the spectrum by comparing the polynomial to the spectrum and iterating 175 times until a baseline was found. The baseline was then subtraced from the spectrum to compensate for the fluorescent background.
3.7.5 Noise reduction
To reduce noise in the Raman spectroscopic data a Savitzky and Golay filter (MatLab R2017a with a framelength of 9 and a polynomial of degree 3) was used on the final spectrum at each sample. The final spectrum was constructed from the mean value of 2-4 measurement points for each sample, leaving out the measurement points that were considered as outliers.
3.7.6 Variance between measurement points
To get a view of how the data between measurement points in each sample differed from each other, the variance between each measurement point was calculated for each sample in MatLab R2017a. The variance is a measure of how far from the mean a data point is and gives therefore an image of how spread the data points in a set are. [26] If the variance is low, it means that the data points are close to each other. And on the opposite, if the variance is high, it means that the data points in the set are far from each other. From each sample, a matrix was created with the data collected from each measurement point as a row in the matrix. By computing the variance in each column in the matrix, the variance between measurement points for each Raman shift were obtained. To get a resulting spectra as representative as possible, it is therefore desired to have a low variance between the measurement points in each sample since that means that the spectra from the points are more similar to each other.
4 Results
4.1 Evaluation measurements
From the evaluation measurements on conventional meat i.e. pork belly and entrecot, Raman spectra were analysed in the fingerprint region, see Figure 12.
Figure 12: Analysed spectra from porkbelly to the left and entrecot to the right presented in the fingerprint region. The spectra was collected with an exposure time of 30 s and an laser power of 12.5 mW.
From the evaluation measurements it was possible to see numerous expected Raman bands in pork and beef due to the large content of proteins and lipids. For example, in pork belly it was possible to see Raman bands from C-C stretching, CH3 twisting and CH2 scissoring in lipids at 1061, 1299 and 1439 [cm 1], respectively. For entrecot, the same Raman bands were observed at 1063, 1295 and 1439 [cm 1], respectively. It was also possible to see Raman bands related to proteins such as from Amide I ↵ helix that was observed in both pork belly and entrecot at 1653 [cm 1]. Based on the presence of these Raman bands and their strong intensities, it was concluded that the Raman system built in this study was possible to be used also to measure on brain tissue that is known to consist a high content of both proteins and lipids, of which several Raman bands are possible to use for brain tumor diagnostics.
4.2 Measurements on brain tissue
Raman spectra were analysed from all tissue samples and the observed peak intensity for Raman bands used in brain tumor diagnosis was noted for each tissue sample in Table 10 and 11 for peaks in the fingerprint region and high wave number region respectively. The number of meausrement points for each sample is stated in the methods section in Table 9. The location for the Raman bands associated with brain tissue are marked with black, dashed lines in each spectrum. For the samples stained with 5-ALA, the location for the Raman bands associated with 5-ALA are marked with red, dashed lines in each spectrum. In this section, only the final two spectra are presented for each sample. Spectroscopic data from all measurement points in each sample can be seen in Appendix A.
The analysed spectra from sample 1 that was photosetisized with 5-ALA can be seen in Figure 13. In both the fingerprint region and the high wavenumber region, there was a lot of interference from the background as can be seen in the spectra. The detected Raman signal was low in com- parison with the signal from the background and made it difficult to distinguish the Raman lines.
It was however possible to see some Raman bands associated with brain tissue as well as some bands associated with 5-ALA.
Figure 13: Analysed spectrum from sample 1 presented in the fingerprint region on the top and the high wavenumber region below. Average location for Raman bands associated with brain tissue are marked in black and for 5-ALA in red.
The analysed spectra from sample 2, that was photosensitized with 5-ALA and illuminated with blue light can be seen in Figure 14. When the tissue sample had been illuminated with blue light, the intensity of the background signal was much lower and it was possible to see distinct Raman peaks associated with brain tissue. The high wave number region was however still dominated by signal from the background and it was hard to distinguish any Raman bands associated with brain tissue.
Figure 14: Analysed spectrum from sample 2 presented in the fingerprint region on the top and the high wavenumber region below. The average location for Raman bands associated with brain tissue are marked in black and for 5-ALA in red.
The analysed spectra from sample 3 can be seen in Figure 15. It was possible to reveal some distinct Raman bands associated with brain tissue in both the fingerprint- and the high wavenum- ber region. However, the spectrum in the fingerprint region appears similar to spectra associated with reduced Neuroglobin (NGB) which can be associated with cancerous tissue [27].
Figure 15: Analysed spectrum from sample 3. The average location for Raman bands associated with brain tissue are marked in black.
The analysed spectra from sample 4 can be seen in Figure 16. The signal from the background was high, but it was still possible to distinguish Raman bands associated with brain tissue in both the fingerprint- and the high wavenumber region.
Figure 16: Analysed spectrum from sample 4. The average location for Raman bands associated with brain tissue are marked in black.
The analysed spectra from sample 5 can be seen in Figure 17. It was possible to distinguish Raman bands associated with brain tissue in both the fingerprint- and the high wavenumber region.
Figure 17: Analysed spectrum from sample 5. The average location for Raman bands associated with brain tissue are marked in black.
The analysed spectra from sample 6 can be seen in Figure 18. In the fingerprint region, the
spectrum was dominated with signal from the background which made it difficult to see Raman bands associated with brain tissue. In the high wavenumber region, the signal from the background was a bit less dominant which made it possible to see some peaks associated with brain tissue.
Figure 18: Analysed spectrum from sample 6. The average location for Raman bands associated with brain tissue are marked in black.
A difference spectra between sample 2 and 1 was created by subtracting the analysed spectrum of tissue sample 1 from the analysed spectra from tissue sample 2 in Figure 19. The spectrum illus- trates the difference between brain tissue that was stained with 5-ALA and brain tissue that was stained with 5-ALA and illuminated with blue light before the measurements. In the fingerprint region, it was possible to see some distinct peaks around 1600 [cm 1]and 1300 [cm 1]associated with 5-ALA. It was also possible to see Raman bands associated with brain tissue at 749, 1331, 1580 and 1598 [cm 1] from Nucleic acids, Trp, Amide II and Phenalaline, Tyrosine respectively.
[4], [22] The rest of the difference spectrum is seen as a broad "hill" and not as distinct Raman lines.
