Spectroscopic Analysis for biological samples :
towards in situ sample analysis of body fluids
Gilwon Yoon
September 27, 2006
Seoul National University of Technology
Spectra of water, Hb(RBC), albumin, glucosefrom visible to NIR (water compensated)
400 600 800 1000 1200 1400 1600 1800 2000 2200 2400
0.00
0.05
0.10
0.15
0.5 mm pathlength, temperature control, 37
albumin, 8 g/dl
hemoglobin, 16.9 g/dl
glucose, 5g/dl
abso
rban
ce
wavelength (nm)
0
1
2
3
abso
rban
ce, w
ater
water
Absorption spectrum of water
Involved Key Technologies
Spectroscopicdetection
target component
Interferingsubstances
inhomogeneous medium
Visible/IR Light source
Light interaction with tissue
High S/N electronic detection
Chemometrics
Clinical testStatisticalanalysis
Prediction ofconcentration
Spectroscopic Analysis – Statistical Methods
Influence of measurement setup : Transmission or reflection measurement
Influence of red blood cell (hemoglobin) in partial least squares regression (PLSR) analysis
Independent Component Analysis (ICA) – a method without calibration process
I. Influence of measurement setup :Transmission or reflection measurement
(a) (b)
Light source
Mono-chromator
slit
DetectorSample
Detector
Comparison between reflectance and transmittance
Jeon, Hwang, Hahn, Yoon (2006), 11:1:014022, Journal of Biomedical Optics
wavelength region [nm]
SEC[mg/dl]
SEP[mg/dl]
diffuse reflectance (10mm thick sample)
1100-2500 27.38 275.44
1100-1850 15.91 437.54
1850-2500 30.57 192.00
diffuse transmittance(1mm thick sample)
1100-1800 2064-2338
3.22 24.69
1100-1800 2.88 26.77
2064-2338 4.50 43.51
diffuse transmittance(2mm thick sample)
1100-1830 2050-2392
26.46 39.07
1100-1830 40.47 83.58
Diffuse reflectance between 1100 and 1850 nm (a) SECV with respect to the optimal number of factor, (b) Loading vector of calibration model, (c) Regression vector of calibration model, (d) Prediction of glucose illustrated with the intralipid concentrations of sample solutions.
0 2 4 6 8 100
50
100
150
200
250
300
350
400
1000 1200 1400 1600 1800 2000-0.20
-0.15
-0.10
-0.05
0.00
0.05
0.10
0.15
1000 1200 1400 1600 1800 2000
-60000
-40000
-20000
0
20000
40000
0 200 400 600 800 1000
0
200
400
600
800
1000
1200
1400
1600
s4.08
s4.08
s4.08s4.08
s4.08
s4.08
s4.08
s4.16
s4.16
s4.16
s4.16
s4.16
s4.16
s4.16
s4
s4
s4s4
s4
s4
s4
(b)S
EC
V [
mg
/dl]
factor
(a)
load
ing
vec
tor
[a.u
.]
wavelength [nm]
factor1 factor2 factor3
(c)
reg
ress
ion
vec
tor
[a.u
.]
wavelength [nm]
(d)
SEP= 437.54 mg/dlCV= 98.8%
ref prediction
pre
dic
ted
glu
cose
[m
g/d
l]
reference glucose [mg/dl]
Diffuse transmittance with 1 mm thick samples (a) SECV with respect to the optimal number of factor, (b) Loading vector of calibration model, (c) Regression vector of calibration model, and (d) Prediction of glucose concentrations.
0 2 4 6 8 100
50
100
150
200
250
300
350
400
1000 1200 1400 1600 2200 2400-0.20
-0.15
-0.10
-0.05
0.00
0.05
0.10
0.15
0.20
0.25
1000 1200 1400 1600 1800 2200 2400-10000
-8000
-6000
-4000
-2000
0
2000
4000
6000
8000
0 200 400 600 800 1000
0
200
400
600
800
1000
1200
s4.16
s4
s4
s4.16
s4
s4
s4.16s4
s4.08s4
s4.16
s4.08
s4.08
s4.16
s4.08
s4.08
s4.16
s4.08
s4.08
s4.16
s4
(b)
Prediction1mm path lengthSEP= 24.69 mg/dl
SE
CV
[m
g/d
l]
factor
(a)
load
ing
vec
tor
[a.u
.]
wavelength [nm]
factor1 factor2 factor3
(c)
reg
ress
ion
vec
tor
[a.u
.]
wavelength [nm]
(d)
pre
dic
ted
glu
cose
[m
g/d
l]
reference glucose [mg/dl]
ref prediction
II. Influence of red blood cell (hemoglobin) in partial least squares regression (PLSR) analysis
Absorption becomes much stronger towards longer wavelengths Dominance of hemoglobin Interferences among the substances in blood or extracellular fluid Effect of preprocessing methods
Biological Samples in the near infrared (1000 – 2500 nm)
a) Whole blood spectra of
98 samples and saline
spectrum, b) Whole blood
spectra are correlated
with hemoglobin and
glucose concentrations at
each wavelength and
computed correlations
coefficients are shown.
1200 1400 1600 1800 22000.0
0.2
0.4
0.6
0.8
1.01200 1400 1600 1800 2200
0.0
0.3
0.6
0.9
1.2
1.5
1.8
2.1
2.4
b
Glucose
Hemoglobin
corr
elat
ion
co
effi
cien
t (r
)
wavelength (nm)
whole blood
Saline
1888 2044
a
abso
rban
ce (
a.u
.)
What is a maximally achievable accuracy ?
The standard error of glucose prediction was 25.5 mg/dl and the coefficient of variation in prediction was 11.2%.
Kim and Yoon (2006), 11: 041128, Journal of Biomedical Optics
Calibration set
Prediction setmean valueof glucose
SEPa (rPreb) VCPre
c [%]
Hbcal Hbpre 228 25.5 (0.9764) 11.2
Hbhigh
Hbmid 213 23.1 (0.9817) 10.8
Hblow 221 48.7 (0.9279) 22.0
Hbmid
Hbhigh 207 39.3 (0.9465) 19.0
Hblow 221 46.9 (0.9328) 21.2
Hblow
Hbhigh 207 74.2 (0.8672) 35.8
Hbmid 213 33.8 (0.9603) 15.9
III. Independent Component Analysis (ICA) –method without calibration process
Identification of pure, or individual, absorption spectra of constituent components from the mixture spectra without a priori knowledge of the mixture.
This method was tested with a two-component system consisting of aqueous solution of both glucose and sucrose, which exhibit distinct but closely overlapped spectra.
ICA combined with principal component analysis was able to identify a spectrum for each component, the correct number of components, and the concentrations of the components in the mixture. This method does not need calibration process.
Hahn and Yoon (2006), in print, 45:32, November, Applied Optics
Pure, or individual, water-subtracted absorption profiles of Glucose (G) and Sucrose (S)
960 980 1000 1020 1040 1060 1080 1100 1120
Ab
so
rba
nc
e
S
G
Wavenumber (cm-1)
25 measured mid-IR spectra for the mixtures of glucose and sucrose. Water absorption was subtracted to enhance the
absorption profile of each component.
960 980 1000 1020 1040 1060 1080 1100 11200.0
0.2
0.4
0.6
0.8
1.0
Abs
orba
nce
Wavenumber(cm-1)
Extracted pure-component spectra from measured IR spectra of 25. ‘Pure’ and ‘ICA’ represent pure-component absorption spectrum and the ICA-method extracted absorption spectrum respectively.
960 980 1000 1020 1040 1060 1080 1100 11200.2
0.4
0.6
0.8
1.0
1.2
960 980 1000 1020 1040 1060 1080 1100 1120
0.4
0.5
0.6
0.7
0.8
ICA
Pure
Glucose
Wavenumber(cm-1)
Abs
orba
nce
Abs
orba
nce
ICA
Pure
Sucrose
Wavenumber(cm-1)
Scatter plot for the reference concentrations and ICs from measured mid-IR spectra
1 2 3 4 50
1
2
3
4
1 2 3 4 5
-11
-10
-9
-8
IC1(
a.u
.)
IC2(
a.u
.)
Sucrose (a.u.)
Glucose(a.u.)
Measurement geometry or setup – loading factor analysis ca
n provide actual contribution of wavelength in prediction
Dominant absorber such as RBC(hemoglobin) and water in n
ear infrared effect substantially. A proper care is needed.
A new method that does not require no concentration inform
ation and calibration process is introduced.
Summary in Spectroscopic analysis