6/8/2009

Beckman Institute
at The University of Illinois

Vibrational Spectroscopy
and Imaging

Rohit Bhargava

Spectroscopy generalization

Compare the light that went in with the light that comes out

Known radiation goes in Collect exiting radiation

Interacts with sample

A general definition of spectroscopy is the study of the

interaction between radiation and an analyte as a
function of wavelength.

Spectroscopy (Radiation)

1010-108 cm-1: Re-arrangement of elementary particles (gamma ray)

108-106 cm-1: Transitions of inner electrons (X-ray)
106-104 cm-1: Transitions of valence electrons (UV-vis spectroscopy)
104-102 cm-1: Transitions between vibrational levels (IR and Raman)
102-100 cm-1: Transitions between rotational levels (Microwave, IR, Raman, THz)
100-10-2 cm-1: Transitions between electron spin levels (magnetic – ESR)
10-2-10-4 cm-1: Transitions between electron nuclear spin levels (magnetic – NMR)
Radio
Medium
Short

Long
UV

Cosmic Gamma X IR Micro UHF

Ultra violet Vis Infrared

Near Mid Far

1 400 750 2,500 16,000 1,000,000 nm

1
 6/8/2009

Spectroscopy of liquid water

UV Spectrum Visible Spectrum NIR Spectrum IR Spectrum

0.07 0.07 14000
Absorption (1/cm)

120
2.5
0.06 0.06 12000
2
100
1.5
0.05 0.05 10000
1
80 0.5
0.04 0.04 8000
0

60 800 1000 1200

0.03 0.03 6000

0.02 40
0.02 4000

0.01 0.01 20
2000

0 0 0
200 250 300 350 400 400 500 600 700 1000 1500 2000 250 0
4000 6000 8000 10000 12000 14000 16000

Wavelength (nm)

Ultra violet Vis Infrared

Near Mid Far

1 400 750 2,500 16,000 1,000,000 nm

Why is water blue?

Gas Assignment liquid •Gas and liquid have different

positions
3651 v1 3400
3755 v3 •Combinations and overtones are
5332 v2+v3 5150 weaker
7251 v1+v3 6900
8807 v1+v2+v3 8400 • Significant example of absorption
10613 2v1+v3 10300 determining color
13831 3v1+v3 13160 (760 nm)
•What is the effect of cold water?
14319 v1+3v3 13510 (740 nm) Ice?

15348 3v1+v2+v3 15150 (660 nm)

15832 v1+v2+3v3 15150 (660 nm)

16822 3v3+2v2+v1 weak

So, what are these vibrational modes

Symmetrical Antisymmetrical Scissoring Rocking Wagging Twisting

stretching stretching

Simplest Model of a Diatomic Molecule: beads and spring harmonic oscillator

- Hooke’s Law Restoring force is proportional to displacement (K~105 dynes

- Newton’s law: force if mass x acceleration
1
0 ~ 1012-1014 Hz
2

2
 6/8/2009

Spectroscopy : Molecular Basis

Raman Scattering
  :  0 2 0

        :  0 2 0

  : 0 2
    :  0 0   … ….
0
1
  0 0 2 0 0 0 2 0 2 0
2 0 0
:   

Note: - change in polarizability, excitation frequency and shift by vibrational frequency

- Two types of shifts to lower (Stokes) and higher (anti-Stokes): which population is favored?

Raman Spectroscopy: Basic Concept

Detector

785
785nm
Sample

Ultra violet Vis Infrared

Near Mid Far

1 400 750 2,500 16,000 1,000,000 nm

First observed by Sir C. Venkata Raman in 1928

using sunlight and photographic filters (won Nobel
price in physics in 1930). Nobelprize.org

Raman Spectroscopy: Basic Concept

Electronic
States

UV-VIS
Absorption

Chance of occurring is Wavelength Dependant

Resonance Raman Scattering?

http://en.wikipedia.org/wiki/File:Raman_energy_levels.jpg

3
 6/8/2009

Raman Spectroscopy: Basic Concept

Electronic
States

UV-VIS
Absorption

If light is not absorbed… the majority of the photons pass through the sample by
means of Rayleigh Scattering

http://en.wikipedia.org/wiki/File:Raman_energy_levels.jpg

Raman Spectroscopy: Basic Concept

Electronic
States

UV-VIS
Absorption

Some of the photons (~1/10,000,000) that are not absorbed will pass through the
sample and undergo Stokes Raman Scattering

http://en.wikipedia.org/wiki/File:Raman_energy_levels.jpg

Raman Spectroscopy: Basic Concept

Electronic
States

UV-VIS
Absorption

Even fewer photons (temperature dependant) that are not absorbed will pass
through the sample and undergo Anti-Stokes Raman Scattering

http://en.wikipedia.org/wiki/File:Raman_energy_levels.jpg

4
 6/8/2009

Raman Spectroscopy: Basic Concept

…
…
780nm
781nm
782nm
783nm
785 nm Molecules in 784nm

Photon the sample

786nm
787nm
788nm
789nm
790nm
791nm
792nm
793nm
794nm
795nm
Detector 796nm
797nm
798nm
785nm 799nm
Sample 800nm
801nm
802nm
803nm
Ultra violet Vis Infrared 804nm
805nm
806nm
Near Mid Far

…
…
1 400 750 2,500 16,000 1,000,000 nm Detector Array

Raman Spectroscopy: Basic Concept

…
…

780nm
781nm
782nm
783nm
Molecule in 784nm

the sample
786nm
787nm
788nm
789nm
790nm
791nm
792nm
793nm
93
794nm
795nm
Detector 796nm
797nm
798nm
785nm 799nm
Sample 800nm
801nm
802nm
803nm
Ultra violet Vis Infrared 804nm
805nm
806nm
Near Mid Far
…
…

1 400 750 2,500 16,000 1,000,000 nm Detector Array

Raman Spectroscopy: Basic Concept

…
…

780nm
Rayleigh Scatter 781nm
782nm
783nm
Molecules in 784nm

the sample
786nm
787nm
785 nm Light 788nm
Gets rejected 789nm
790nm
791nm
792nm
793nm
93
794nm
795nm
Detector 796nm
797nm
798nm
785nm 799nm
Sample 800nm
801nm
802nm
803nm
Ultra violet Vis Infrared 804nm
805nm
806nm
Near Mid Far
…
…

1 400 750 2,500 16,000 1,000,000 nm Detector Array

5
 6/8/2009

Raman Spectroscopy: Basic Concept

…
…
780nm
781nm
782nm
783nm
785 nm Molecules in 784nm

Photon the sample

786nm
787nm
788nm
789nm
790nm
791nm
792nm
793nm
794nm
795nm
Detector 796nm
797nm
798nm
785nm 799nm
Sample 800nm
801nm
802nm
803nm
Ultra violet Vis Infrared 804nm
805nm
806nm
Near Mid Far

…
…
1 400 750 2,500 16,000 1,000,000 nm Detector Array

Raman Spectroscopy: Basic Concept

…
…

780nm
781nm
782nm
783nm
Molecule in 784nm

the sample
786nm
787nm
788nm
789nm
790nm
791nm
792nm
793nm
93
794nm
795nm
Detector 796nm
797nm
798nm
785nm 799nm
Sample 800nm
801nm
802nm
803nm
Ultra violet Vis Infrared 804nm
805nm
806nm
Near Mid Far
…
…

1 400 750 2,500 16,000 1,000,000 nm Detector Array

Raman Spectroscopy: Basic Concept

…
…

780nm
Anti-stokes Raman 781nm
782nm
783nm
Molecules in 784nm

the sample
786nm
787nm
781 nm Light 788nm
Passed through 789nm
790nm
to the detector 791nm
792nm
793nm
93
794nm
795nm
Detector 796nm
797nm
798nm
785nm 799nm
Sample 800nm
801nm
802nm
803nm
Ultra violet Vis Infrared 804nm
805nm
806nm
Near Mid Far
…
…

1 400 750 2,500 16,000 1,000,000 nm Detector Array

6
 6/8/2009

Raman Spectroscopy: Basic Concept

…
…
780nm
781nm
782nm
783nm
785 nm Molecules in 784nm

Photon the sample

786nm
787nm
788nm
789nm
790nm
791nm
792nm
793nm
794nm
795nm
Detector 796nm
797nm
798nm
785nm 799nm
Sample 800nm
801nm
802nm
803nm
Ultra violet Vis Infrared 804nm
805nm
806nm
Near Mid Far

…
…
1 400 750 2,500 16,000 1,000,000 nm Detector Array

Raman Spectroscopy: Basic Concept

…
…

780nm
781nm
782nm
783nm
Molecule in 784nm

the sample
786nm
787nm
788nm
789nm
790nm
791nm
792nm
793nm
93
794nm
795nm
Detector 796nm
797nm
798nm
785nm 799nm
Sample 800nm
801nm
802nm
803nm
Ultra violet Vis Infrared 804nm
805nm
806nm
Near Mid Far
…
…

1 400 750 2,500 16,000 1,000,000 nm Detector Array

Raman Spectroscopy: Basic Concept

…
…

780nm
Stokes Raman 781nm
782nm
783nm
Molecules in 784nm

the sample
786nm
787nm
790 nm Light 788nm
Passed through 789nm
790nm
to the detector 791nm
792nm
793nm
93
794nm
795nm
Detector 796nm
797nm
798nm
785nm 799nm
Sample 800nm
801nm
802nm
803nm
Ultra violet Vis Infrared 804nm
805nm
806nm
Near Mid Far
…
…

1 400 750 2,500 16,000 1,000,000 nm Detector Array

7
 6/8/2009

Raman Spectroscopy: Basic Concept

786nm
787nm
788nm
Billions of 785nm photons 789nm
790nm
Molecules in 791nm
792nm
a soybean 793nm
794nm
795nm
796nm
797nm
798nm
799nm
800nm
801nm
802nm
803nm
Detector 804nm
805nm
806nm
785nm

…
…
Soybean
Detector Array
Ultra violet Vis Infrared
Near Mid Far

1 400 750 2,500 16,000 1,000,000 nm

Raman Spectroscopy: Basic Concept

786nm
787nm
788nm
9 789nm
790nm
791nm
792nm
ntensity (A.U.)

8
793nm
794nm
7 795nm
796nm
6 797nm
798nm
799nm
5 800nm
801
801nm
Raman In

802nm
4 803nm
804nm
805nm
3 806nm
Detector Array

…
…

… 2
Detector Array
803nm

801nm
800nm
799nm
798nm
797nm
796nm
795nm
794nm
793nm
792nm
791nm
790nm
789nm
788nm
787nm
786nm
806nm
805nm
804nm

802nm

… 1

0
400 600 800 1000 1200 1400 1600 1800

810nm Raman Shift (cm-1) 915nm

Raman Spectroscopy: Raman Shift

Excitation Frequency
9
Soybean 785nm = 0.0000785cm
1
8 = 12738.9 cm-1
0.0000785cm
7

6 Wavelength of 400 cm-1

Raman shift in nm
5 12738.9 cm-1 – 400 cm-1 = 12338.9 cm-1
1
4 = 0.0000810cm
12338.9 cm-1
3 810nm

2
***The Raman shift is independent of
1 excitation frequency… the spectral
band positions will be the same
0
400 600 800 1000 1200 1400 1600 1800
whether you excite with a 785nm
laser or a 532 nm laser
810nm Raman Shift (cm-1) 915nm

8
 6/8/2009

Raman Spectroscopy: Raman Shift

Wavelength Dependence of Raman Cross-Section

Important advantage of
Raman… wavelength
select ability to avoid
light absorption and
fluorescence… however
the higher the
wavelength the lower the
Raman efficiency
(Raman Cross Section).

3400 cm-1 O-H stretch of liquid water

Appl. Opt. 36, 2686-2688 (1997)

Raman Spectroscopy: Selection Rules

Raman Bands result from the

Vibrations of specific chemical
Bonds.
Soybean band assignments:
Wave number Molecule Assignment

800-900 C-C Backbone

1003 Phe Ring breathing
1010-1200 C-C Backbone
1265 RC=ONH2 Amide III
1450 CH2 Stretch
1660 RCH=CHR Cis
1670 RCH=CHR Trans
1750 RC=OOR Ester

Raman Spectroscopy: Examples

NIR Spectra

9 Raman Spectrum
n Intensity (A.U.)

4
Raman

IR Spectrum 3

0
400 600 800 1000 1200 1400 1600 1800

810nm Raman Shift (cm-1) 915nm

2500nm 11,000nm

9
 6/8/2009

Raman Spectroscopy: Instrumentation

CCD

Las
ser

Kaiser Optical

Raman Spectroscopy: Instrumentation

-This is a axial dispersive spectrograph

-Reflective spectrographs are also common

Spectrograph

-Interferometers are used for FT-Raman

Kaiser Optical

Raman Spectroscopy: Instrumentation

-Charged-coupled devices (CCDs) are the

most common detectors
(256x1024 pixel array)

-FT Raman is used for frequencies where

ctor

CCDs have low quantum efficiency

Detec
Height

Spectral Dimension

10
 6/8/2009

Raman Spectroscopy:
Advantages and Disadvantages

Advantages:

-Little sample preparation (Polishing and fixing to a slide is common)

-Not sensitive to water (Good for biological samples)

-High chemical specificity (Narrow spectral bands)

-Qualitative and Quantitative information

-Non-destructive (A measurement does not chemically or physically change the sample)

-Can take measurements on solids, liquids, or gases

-Measurements are taken without touching the sample (Remote sensing)

-Easily coupled with fiber-optics

Raman Spectroscopy:
Advantages and Disadvantages

Disadvantages

-Acquisition times tend to be longer than other techniques (real time

measurements have been demonstrated… but something like video rate
imaging is not yet a reality)

-Raman signal tends to be weak

-Raman signal is often mixed with a fluorescent background signal, which

can make signal processing difficult.

-High laser powers and burn delicate samples

Raman Microscopy

785 nm Laser
Microscope

Spectrograph & CCD

Dichroic mirror
1.2
Transmittance (%))

0.8

0.6

0.4

0.2

0
1034
1068
1103
1138
1172
1207
307
342
376
411
446
480
515
549
584
619
653
688
722
757
792
826
861
895
930
965
999

Wavelengh (nm)

Wavelength (nm)

11
 6/8/2009

Confocal Raman Spectroscopy

Confocal Raman Microscopy

Tabaskblat et al., Appl. Spec 1992

Hyperspectral Raman Mapping

Grabber
Frame

PhAT Probe

Microscope

Spectrograph
& CCD

785 nm Laser

Fiber
~~~~ 1

Launcher

50

Collection fibers are transposed into

a line when to enter the spectrograph
Illumination Collection

Hyperspectral Raman Mapping

White Light Illuminated Region

Bone

50μm
PMMA

Raman map
49
Raman Intensity (A.U.)

1
33
Raman Intensity (A.U.)

14

9
34

23

10
13
21
3

41

29
8

17
16
39
35

18

26

47
32
7
24
44

30
5

36
27

42
31
11

12
22

28
46
4
45

2
37
20

6
25
19
40
38
48

50
43

15

Raman Shift (cm-1) Raman Shift (cm-1)

12
 6/8/2009

Spatially Offset Raman Spectroscopy

OD

ID

Spectrograph
35 3
14
22 44 18 21 33
13

& CCD
20 11 5 39 49
16 34
37 12 24 1
48
7 8 23
38 45 31
36 26 41 9
43 2
40 27
4 32 17 10
15 25

Probe
785 nm Laser
46 42 47 29
50 19
6 28 30

Spacing
Fiber
Illumination

Converging
ring
g
Collection
launcher
lens
Fibers

Collection
Dichroic Fibers

mirror
Illumination
Axicon
collimator
Fiber

lens
Converging

lens(es)
Diverging

Sample/
Collection
Specimen

Spatially Offset Raman Spectroscopy

ensity (A.U.)

Transcutaneous
Exposed bone
Bone factor
Raman Inte
Water cooling

Anesthesia
Raman Shift (cm-1)

Schulmerich et. al., Appl. Spectrosc., 63(3) 286-295

Other Raman Techniques

-Surfaced enhanced Raman spectroscopy (SERS)

-Tip enhanced Raman spectroscopy (TERS)

-Resonance Raman spectroscopy

-Raman tomography

-Raman imaging

-Coherent anti-stokes Raman spectroscopy (CARS)

-Stimulated Raman spectroscopy

-Others I can’t think of at the moment and much more to come!

13
 6/8/2009

Infrared Spectroscopy

Theory of Infrared – Chemical bonds

• Infrared spectroscopy :
ƒ Principle: chemical bonds rotate or
vibrate at specific frequencies (Group
theory basis of IR Spectroscopy)
• Practical Aspects
ƒ The frequency at which resonance
occurs depends on the properties of the
chemical bond
ƒ Factors include shape of molecular
bond, energy levels and mass of atoms.
ƒ Chemical bonds can be divided into
those which are IR active and those
which are IR inactive

Molecular Basis of IR Absorption

Electronic
States

• Direct absorption of light UV-VIS

Absorption

• Beer’s law relates absorption to concentration

0
0    log C‐H
stretching C‐H
bending
Increasing 
absorption of IR 
radiation
C‐C
bending

• Selection rule Increasing wavelength

– Dipole moment (symmetrical mode: diatom)

Increasing wavenumber (energy, frequency)

14
 6/8/2009

Theory of Infrared – Chemical bonds

• To be an IR active mode, the motion must have a change in the
electric dipole moment of the bond
• Implications
ƒ The intensity of absorbance peaks is related to size of dipole moment.
ƒ Bonds with higher dipole moments tend to be covalent bonds with
highly different electronegativities e.g C
C=O
O
ƒ Symmetrical bonds are typically IR inactive e.g. C-C, N-N
ƒ The same chemical bond can also have multiple modes of vibration
e.g. phosphate has a symmetrical and antisymmetrical modes

Infrared Spectra
• Vibrational frequencies are directly resonant with optical
frequencies – vibrational transition is energy matching

Infrared Spectroscopy
• IR was discovered in 1801 by William Herschel who split the EM spectrum
using a prism – noted increase in temperature beyond the red part of visible
spectrum
• 1930’s began to be exploited for spectroscopy studies

• Two mains types of spectrometer – Dispersive and Fourier Transform

• Dispersive
Di i uses monochromatic
h i source off light
li h andd changes
h frequency
f over
time by moving grating, mirror or detector. Originally used prisms and later
grating.
• Largely replaced by Fourier Transform – Allows for all frequencies to be
measured at once: Fellgetts advantage – No slits: higher throughput

15
 6/8/2009

IR Spectroscopy: Theory
Retardation (cm)
-0.05 0.05 0.1 0.15 0.2 0.25
3
1  
1
 
2 Cb
1 2.2 2.5

2.0 W 2.0
4
Detector 2 and 2’
∞ 1.8
Focusing mirror I0 1.5
0.5 1 2   BD,0
∞ 1.6
1.0
3 3 1.4
∞
0.5
1 2 Sample 0 0.5 1 2   1.2
∞
0.0
∞ or 4000 3200 2400 1600 800 0
Fixed mirror 0.5 1 2   Wavenumber (cm-11)
∞ Cb:Centerburst, W:Wings

Beamsplitter ∞ 10000 Water

2’ 4 ,0 0.5 2  

Absorption Coeff (cm-1)

Broadband ∞ Tissue
Source δ= 0
∞
or 1000
, 0.5 2  
Moving mirror δ= δmax ∞
100
Computation (FT)
∞
, 2 , 2   10
4000 3500 3000 2500 2000 1500 1000
0
Wavenumber (cm-1)

FTIR Microscope

A popular
commercial
instrument
combining the
FTIR spectrometer
with a microscope

Point spectra or Mapping/Imaging?

A) Single Point Detector B) Linear Array Detector C) Focal Plane Array

(mapping) (imaging) (imaging)

Example improvements in speed

1 16 16,384
Single point 16 detector linear array 128x128 Focal Plane Array

In fact, Imaging is even quicker and allows for diffraction limit measurements

16
 6/8/2009

FT-IR spectroscopy Æ Imaging

1.6

0.4

100 μm

a.u.)
y 0.6

Absorbance (a
0.4

z 0.2

x 0.0
3200 2800 2400 2000 1600
Wavenumber (cm -1)
• Typical characteristics
– Wavelengths (2048 elements over 2.5 – 12.5 μm), x, y typically ~1024
• Computation is essential to recover data
– Manual examination is prohibitive
• Trade-offs: spatial coverage vs. resolution, spectral resolution vs.
signal-to-noise ratio, time vs. data quality/size vs. information

FTIR Chemical Imaging

10 5 A 1
CCD Visible Detector
10 4
MAPPING 3
10 3 2 4
Time (hrs)

10 2 7a
5
Aperture 10
Sample 6 7b
Stage 1
8
Aperture
10 -1 9
10 -2 10
10 -3 11
Mirror
10 -4
Rapid-Scan
Interferometer
Visible Light 1992 1996 2000 2004 2008
Source Year
Microscope

8 cm-1 , SNR of 1000:1, 1 Mpix, 6 x6 micron res., 1 cm x 1 cm

Blue – hardware, red – interferometry/software, green – emerging technology

1. E.N. Lewis et al. Anal. Chem. 67, 3377-3386 (1995) ; 2. Snively et al. Appl. Spectrosc. (1998); 3. Bhargava et al. Appl. Spectrosc. 53, 1313-
1322 (1999); 4. Snively et al. Opt. Lett., 24; 1841-1843 (1999); 5. Bhargava et al. Appl. Spectrosc. 54, 486-495 (2000); 6. Perkin-Elmer Inc.; 7a.
Varian Inc. ; 8. NIH-FBI camera; 9. Reddy et al (To be submitted); 10 & 11 – under development.

Settings and Parameters

A number of important considerations must be taken into
account prior to experiments;

• Sample preparation is critical!

• Signal to Noise – Number of co-additions
• Mirror speed
• Spectral resolution
• Spatial resolution
• Point spectra vs. Point Mapping vs. Imaging

17
 6/8/2009

Sample preparation
• Transmission or reflection

• IR transparent substrates – BaF2, CaF2

• IR reflective substrates – Gold, MirrIR Slides
• Reflection is low-cost, however reduction in spectral quality and
increase in scattering artifacts

Signal to noise
• It is important to get high quality noise-free data –
This can involve multiple strategies;
– Better sample preparation (more or less sample)
– More scans and averaging
– Reduce Michelson mirror speed
• Spectral resolution is important,
important this is the number of
points within a spectra, the fewer the faster the scans
but potential for loss of information
• Spatial resolution is also important, the higher the
spatial resolution, the lower the signal.
Take home message; Many factors must be adapted depending on what
you are interested in

Application 1: Forensic Sciences

 

Body oils

0.005

Absorbance at 2920 cm-1

Skin Flakes

0.000

1 mm

Absorbance at 1568 cm-1

18
 6/8/2009

An example of imaging…..

1.25

Absorbance (a.u.)
• Polymer crystallization…. 1.00
0.75
Crystalline
0.50
0.90 0.25 Difference
0.00
0.75
-0.25
0.60
Amorphous
-0.50
9s 21 s 24 s 27 s
0.45 900 1000 1100 1200 1300 1400 1500
Wavenumber (cm-1)
0.30

0.15

30 s 36 s 45 s 75 s 1.0
A
B
0.090 70 0.8 C

Absorbance (a.u.)
D
Average
0.075 60
0.6
0.060 50
0.4
0.045 40 D

C
0.030 30 0.2 B
A
0.015 20
0.0

Maximum Rate Maximum Rate Time (s) 10 100

Time (s)

Motivation

DIAGNOSTIC
• Cancer pathology
– Prostate cancer as a paradigm – 1 Screening
in 6 men
Is it suspicious?
– Biopsies: >1 million annually with
disease ~20% Biopsy
– Diagnoses: >200, 000 annually Is it cancer?
with lethal ~20%
– Grading
G di iis subjective,
bj ti variable,
i bl Diagnosis
leads to conflicting therapy routes What is the grade? Stage?
– Prognosis tools are not perfect – Will it metastasize?
97% undergo therapy
Therapy
– “Holy Grail” of oncologic pathology
Is adjuvant therapy needed?
– Primary evaluative standard for
research
Follow-up
• Manual recognition in Has cancer recurred?

stained tissue
RESEARCH

Cancer and Diagnostic Process

Biopsy

Epithelium Benign
Grade 3
Stroma
Grade 4

• No stains, no manual decisions

American Cancer Society, est. 2008

19
 6/8/2009

From Data to Knowledge

x=2048, y=2048, z=2048, t ~ ms to days

Information
y

z
x
￭ Model based design of experiments
￭ Hypothesis driven analysis – supervised data analysis
￭ Biologically inspired statistical pattern recognition of spectra
￭ Approach – Model, Train algorithm, Classify, Validate

Analytical Approaches
The types of analysis available to spectroscopists can be
divided into two groups;

All data has the same weighting and

no-prior information about what the
Un-Supervised
Un Supervised spectra
t corresponds d tto is
i kknown
-Useful for interest of discovery

Known information about data e.g.

classes is used for data analysis
Supervised -Useful for classification (more
amenable to clinical setting?)

Analytical Approaches
Which MVA approach?

Bayesian PLS
PCA

SIMCA HCA
FUZZY
LDA
KNN ANN

Important to determine which are best for analysis of point spectra and which for images

20
 6/8/2009

Integrative classification process

Prostate Histology
Acquired Data

Blood

Nerve
Epithelium

Ganglion

Stone
Endothelium
Fib. Stroma

Smooth Muscle
Pale Stroma

Lymphocytes
Model

Potential Metric Set

Algorithm
(Modified Bayesian)

Optimization

Optimal Metric Set

Calibration/Validation
Statistics Optimized Prediction
Sensitivity Analysis Algorithm
Gold Standard

Visualizing Classification Origin and Variance

patients
1 6 3 4 5 6 7 8 9 10 11 12 16 15 13 14 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38
1
2
3
4
5
6
metrics

7
8
9
10
11
12
13
14
15
16
17
18

Determining outliers – quality control

Variance and classification

Control and significant Variables

Spectral Diversity and Clustering

Examples of FTIR for Biomedical applications

Apoptosis
Fundamental biological Necrosis
Stem cell differentiation
processes Toxicology
Disease initiation and pprogression
g

Clinical Applications Disease diagnosis

Disease prognosis

21
 6/8/2009

FTIR-Biomedical Applications

Biochemical-cell fingerprint region

Integrative Process and Key Technologies

Biopsy

Custom Tissue Microarrays

* Kononen et al. Nat. Med. 4, 844-847 (1998).

Levin and Bhargava Annu. Rev. Phys. Chem. 56, 429-474 (2005).
Kong and Bhargava, Submitted (2009): Living tissue arrays
Samples

Hi-fi, fast FTIR Imaging

* Lewis et al. Anal. Chem. 67, 3377

3377-3381
3381 (1995).
Bhargava and Levin Anal. Chem. 73, 5157-5167 (2001)
Bhargava and Reddy Anal. Chem., Submitted (2008): Noise rejection
Imaging Data

Pattern Recognition Algorithms

Llora , Priya, Bhargava Nat. Computing In Press (2009).

Bhargava et al. Biochim Biophys Acta. 1758, 830-845 (2006).
Diagnosis Reddy and Bhargava (2009): Statistical Model

Statistical Validation
Fernandez et al. Nat. Biotechnol. 23, 469-474 (2005).
Prognosis Bhargava et al. Nat. Biotechnol. 25, 31-33 (2007).
Bhargava Anal. Bioanal. Chem. 389, 1155-1169 (2007).

22
 6/8/2009

Application 1: Prostate Histology

Validation of Prostate Histology: 970 samples, 10 million spectra

EPITHELIUM
FIBROUS STROMA
MIXED STROMA
SMOOTH MUSCLE
NERVE
GANGLION CELLS
BLOOD
LYMPHOCYTES
CORPORA AMYLACEA
ENDOTHELIUM

1.3 mm

Fernandez, Bhargava, Hewitt and Levin Nat. Biotechnol., 23, 469-474 (2005)

Prostate Cancer Diagnosis

Array – 80 Patients Array – Histology Pathology Design Pathology Result

• Overall pixel accuracy ~ 88.5% ; Heterogeneity in samples?

• 1 cancer sample classified as benign (71)
• 1 benign sample classified as cancerous (69)
• Sensitivity and specificity exceeding human capabilities
• Large validation studies underway

Whats in store?

• Label-free methods
– Talks on Monday focus on applications
• Lab tours
Theory
• More information
Applications
Instrumentation

Data Analysis
Sampling

23