Professional Documents
Culture Documents
d
s
s
1
+ s
2
d
a
.
The rst exponential term takes into account the effect of
melanin present in the epidermal layer. As light traverses twice
through the melanin layer a factor of 2 is introduced in this term.
The second exponential term takes into account the effect of ex-
tra vascular bilirubin. The third term in the expression is used
to estimate the back-scattered light from the dermal layer of the
skin tissue. The parameters in the rst exponential term: m,
m
,
l
epi
are melanin volume fraction (skin tone), absorption coef-
cient of single melanosome, and epidermal layer thickness. The
second exponential term parameters: c
eb
,
eb
, l
eb
are the extra
vascular bilirubin concentration, molar extinction coefcient of
bilirubin, and the path length which the light travels (i.e., the
thickness of the extra vascular bilirubin layer). The parameters
d
s
and
d
a
are the scattering and absorption coefcients of der-
mal layer. The parameters s1 and s2 take into account the effects
probe geometry.
From systems analysis perspective, an inverse model of the
system can be used to extract parameter from measured data. In
this work, the parameters of interest are the physiologic factors
that contribute to the optical properties of the tissue. A nonlin-
ear least squares solver has been adopted as an inverse model
for this work. Sequential quadratic program (SQP) is one of the
most popular and robust algorithms for nonlinear continuous
optimization. In this research, we have used the NLSSOL (non-
linear least squares solver) routine available in the optimization
software TOMLAB interfaced with MATLAB [6]. The primary
goal of this solver is to minimize the error function (differ-
ence between measured and simulated spectra) by updating the
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 58, NO. 3, MARCH 2011 779
Fig. 3. The pre-, post-spectra data of pig skin samples (pig2 spectral database)
soaked in bilirubin solutions.
parameters of the analytical algorithm to nd a best match to
measured spectrum. Preliminary studies of the signal process-
ing algorithm on simulated spectra using the forward analytical
model have been shown to estimate the bilirubin concentration
with an average error of less than 15% [7].
IV. JAUNDICE ANIMAL MODEL
Due to its similarity with human skinin aspects, such as lay-
ered architecture and optical properties, porcine skin tissue can
be used as an animal model for skin reectance studies. To
model a jaundice condition with normal skin, we require that all
skin tissue absorbance parameters (e.g., melanin concentration,
hemoglobin concentration, epidermis thickness, and scattering
parameters) be held constant except the bilirubin concentration.
To accomplish this reectance spectrum of skin tissue on the
pig forehead was captured before the skin sample was biopsied.
Bilirubin solutions of different concentrations are then prepared
and added to tissue culture plates. The biopsied skin samples
were placed in the tissue culture dishes and left oating in the
bilirubin solutions. The tissue culture plates are sealed with
paralm and placed on a gentle shaker platform for 4 h.
This process replicates the in vivo condition that leads to
jaundice. After the 4 h soaking period, the tissue samples are
removed from the tissue culture plates and placed back on
their respective pig foreheads to capture a post soak spectral
scan. Post soak spectral scans of different tissue samples resem-
ble the different jaundice levels of skin tissue. Tissue samples
which were exposed to low, medium, and high concentrations of
bilirubin, showed the anticipated increase in the bound bilirubin
concentration, as shown in Fig. 3.
Finally, three spectral databases of pre and post spectral scans
have been collected on three different pigs to study the per-
formance of the algorithm. The positive correlation of R
2
=
0.95,0.96,0.96 between the model estimated and soaked biliru-
bin concentration in the three spectral databases forms the basis
of justication that the signal processing model is able to track
the bilirubin concentration changes in jaundice skin samples.
The next step in the process is validating the model is by
comparing the bilirubin concentration values estimated from
the model to the bilirubin concentration calculated from the
assay process of the skin samples. The assay is performed by
rst freezing and then pulverizing a small portion of the soaked
TABLE I
MODEL ESTIMATED VALUES AND THE ASSAY CALCULATED VALUES FOR THE
SKIN SAMPLES (29, 13. 3)
TABLE II
SHOWS THE COMPARISON BETWEEN VALUES QUANTIFIED FROM THE SIGNAL
PROCESSING ALGORITHM TO THE VALUES CALCULATED FROM THE BLOOD
SAMPLE USING BIOCHEMISTRY
tissue. Once pulverized, the skin tissue can be homogenized
into a solution, which can then be assayed for quantication of
bilirubin. The assay calculated values listed in Table I show a
good correlation with the model estimated values. The deviation
in the absolute values is associated with the different volume
perspectives for the two methods (tissue volume versus solution
volume).
As the jaundice skin has a yellowish appearance, and based
on the jaundice physiology, the intensity of the yellowness of
the skin increases as the more bilirubin is present in the skin.
The pig skin tissue samples soaked in higher concentrations of
bilirubin solution, had showed a strong yellowish signal. The
linear increase in yellow color intensity of the post-soaked skin
samples strongly correlated to the amount of bilirubin present in
the skin sample. The interpretation of the yellow color intensity
along with the spectral processing algorithm would provide a
real-time quantication of the jaundice progression.
V. HUMAN VOLUNTEER SPECTRAL DATA
As a step further toward building a robust signal process-
ing algorithm, four volunteer subjects were recruited to cap-
ture diffuse reectance spectral scans of human skin tissue on
the ventral side of forearm. After the spectral scans were col-
lected, a small amount of blood was drawn to perform bio-
chemistry assays on the blood samples to quantify bilirubin,
hemoglobin, and hematocrit. The output parameters, such as
hemoglobin, bilirubin, and hematocrit numbers generated from
the signal processing algorithm, are compared with the numbers
780 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 58, NO. 3, MARCH 2011
Fig. 4. (a) Reectance spectral prole of forehead and nger site of light
pigmented. (Caucasian skin). (b) Reectance spectral prole of forehead and
nger site of lightly dark pigmented individual (Asian skin).
calculated from biochemistry analysis conducted on the blood
drawn from the volunteers is shown in Table II.
The performance of the model on initial sample size has given
a clear understanding of the system behavior, with a high posi-
tive correlation in the prediction of hemoglobin and hematocrit
value.
The spectral data collected on different skin sites of an indi-
vidual and different colored individuals have been analyzed, as
shown in Fig. 4(a) and (b).
The performance of the algorithm is justied by the higher
numbers for the blood-volume fraction and oxygen saturation
on the nger site compared to the forehead. The spectral data
collected on the nger site of a dark skinned individual has
hemoglobin signal bands that are clearly visible, primarily due
to the absence of melanin on the nger tissue site (vascular side).
The spectral data collected on the forehead has the melanin con-
tent from the epidermal layer masking the hemoglobin bands.
As most of the spectral information used to quantify the biliru-
bin is collected from the surface to a depth of around 12 mm,
which is usually the epidermis and dermal layer. With the prin-
ciple of blanching the skin the light can go further deep and
collect the spectral information from the subcutaneous fat layer
too.
The in vivo human data results collected off of the different
individuals was used as a starting step to train the algorithm
for varying optical properties. The poor correlation observed in
the bilirubin device measurement and standard assay could be
strongly improved by collecting more spectral data on healthy
and jaundice individuals. As more spectral data is fed into the
feedback mode of the signal processing algorithm inputs from
the standard assay can form the basis to correct for the missing
parameters in the model. This preliminary human data and the
pig skin jaundice model studies has established the fact that
the device has passed the concept and development phase and
could be headed into the clinical trial phase. The point of care
device with the preliminary results through the signal processing
algorithm does give us the proof that multiple parameters, such
as hemoglobin, hematocrit, and oxygen saturation in addition to
bilirubin, could be quantied using a single device.
VI. CONCLUSION
In this work, a noninvasive point-of-care device for the quan-
tication of bilirubin is developed and evaluated. The device
works by analyzing diffuse reectance spectra collected from
optical reectance from skin tissue. The system is comprised of
hardware components and a signal processing algorithm. The
signal processing algorithm allows the device to quantify the
different chromophores present in the skin tissue. Initial assess-
ment of the device has been completed using a porcine model
for jaundice that involves soaking biopsied pig skin in varying
concentrations of bilirubin solution. Analysis of these skin sam-
ples shows that the device is able to track the changes associated
with the absorption of unbound bilirubin in skin tissue.
REFERENCES
[1] S. N. el-Beshbishi, K. E. Shattuck, A. A. Mohammad, and J. R. Petersen,
Hyperbilirubinemia and transcutaneous bilirubinometry, Clin. Chem.,
vol. 55, no. 7, pp. 12801287, Jul. 2009.
[2] C. E. Ahlfors, Bilirubin-albumin binding and free bilirubin, J. Perina-
tol., vol. 21, pp. S40S42, Dec. 2001.
[3] G. Zonios, Lev T. Perelman, V. Backman, R. Manoharan, M. Fitzmaurice,
J. Van Dam, and M. S. Feld, Diffuse reectance spectroscopy of human
adenomatous colon polyps in vivo, Appl. Opt., vol. 38, pp. 66286637,
1999.
[4] L. F. A. Douven and G. W. Lucassen, Retrieval of optical properties of
skin from measurement and modelling the diffuse reectance, in Proc.
SPIE, 2000, vol. 3914, pp. 312323.
[5] G. Zonios and A. Dimou, Modeling diffuse reectance fromsemi-innite
turbid media: Application to the study of skin optical properties, Opt.
Exp., vol. 14, pp. 86618674, 2006.
[6] [Online]. Available: http://tomopt.com/tomlab/products/sol/solvers/
NLSSOL.php (Accessed: Jan. 22, 2010).
[7] S. K. Alla, J. F. Clark, and F. R. Beyette, Signal processing system to
extract serum bilirubin concentration from diffuse reectance spectrum of
human skin, in Proc. IEEE Ann. Int. Conf. Eng. Med. Biol. Soc. (EMBC),
Sep. 2009, pp. 12901293.
[8] T. Dimitrova, L. Georgieva, C. Pattichis, and M. Neofytou, Qualitative
visual image analysis of bruise age determination: A Survey, in Proc.
IEEE Ann. Int. Conf. Eng. Med. Biol. Soc. (EMBS), 2006, pp. 48404843.