U.S. patent application number 10/638656 was filed with the patent office on 2004-02-19 for multivariate analysis of green to ultraviolet spectra of cell and tissue samples.
Invention is credited to Freeman, Jenny, Mansfield, James.
Application Number | 20040034292 10/638656 |
Document ID | / |
Family ID | 22672465 |
Filed Date | 2004-02-19 |
United States Patent
Application |
20040034292 |
Kind Code |
A1 |
Mansfield, James ; et
al. |
February 19, 2004 |
Multivariate analysis of green to ultraviolet spectra of cell and
tissue samples
Abstract
This invention relates to methods for processing in vivo skin
auto-fluorescence spectra for determining blood glucose levels. The
invention also relates to methods of classifying cells or tissue
samples or quantifying a component of a cell or tissue using a
multivariate classification or quantification model.
Inventors: |
Mansfield, James; (Boston,
MA) ; Freeman, Jenny; (Weston, MA) |
Correspondence
Address: |
MINTZ, LEVIN, COHN, FERRIS, GLOVSKY
AND POPEO, P.C.
ONE FINANCIAL CENTER
BOSTON
MA
02111
US
|
Family ID: |
22672465 |
Appl. No.: |
10/638656 |
Filed: |
August 11, 2003 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
10638656 |
Aug 11, 2003 |
|
|
|
09785531 |
Feb 20, 2001 |
|
|
|
60183356 |
Feb 18, 2000 |
|
|
|
Current U.S.
Class: |
600/316 |
Current CPC
Class: |
A61B 5/14532 20130101;
A61B 5/0059 20130101; A61B 5/0071 20130101; G01J 3/28 20130101;
A61B 5/441 20130101; A61B 5/445 20130101; G01N 21/6486 20130101;
A61B 5/7264 20130101; G01N 2201/129 20130101; A61B 5/1455
20130101 |
Class at
Publication: |
600/316 |
International
Class: |
A61B 005/00 |
Claims
1. A method for processing in vivo skin auto fluorescence spectra
emitted by a skin surface of a patient to determine a blood glucose
level of the patient comprising the steps of: collecting auto
fluorescence spectra emitted from the skin surface of the patient;
and correcting the collected spectra using multivariate analysis to
account for skin surface variables.
2. The method of claim 1 wherein the multivariate analysis
comprises one or more quantification, classification, or data
processing techniques selected from the group consisting of:
partial least squares, principal component regression, linear
regression, multiple linear regression, stepwise linear regression,
ridge regression, radial basis functions, linear discriminant
analysis, cluster analysis, neural network analysis, smoothing
filters, laplacian operators, maximum likelihood estimators,
maximum entropy, first and second derivatives, peak enhancement,
Fourier self-deconvolution, principal components, and varimax
rotations.
3. An instrument for determining a correct glucose level of a
patient by measuring in vivo autofluorescence of the patient's skin
comprising: means for irradiating the skin with a plurality of
excitation wavelengths; means for collecting a plurality of emitted
wavelengths; and means for analyzing the collected wavelengths to
determine a preliminary blood glucose level, said means for
analyzing comprising means for correcting the preliminary blood
glucose level to account for variations in skin, said means for
correcting comprising using one or more multivariate analytical
methodologies to determine the correct glucose level of the
patient.
Description
RELATED APPLICATION
[0001] The present invention claims priority to U.S. Provisional
Patent Application No. 60/183,356, filed Feb. 18, 2000, and titled
"Multivariate Analysis of Green to Ultraviolet Spectra of Cell and
Tissue Samples."
BACKGROUND OF THE INVENTION
[0002] 1. Field of the Invention
[0003] This invention relates to analysis methodology and
multivariate classification of diagnostic spectra, and, in
particular, to methods for processing in vivo skin
auto-fluorescence spectra for determining blood glucose levels. The
invention also relates to methods of classifying cells or tissue
samples or quantifying a component of a cell or tissue using a
multivariate classification or quantification model.
[0004] 2. Description of the Background
[0005] Near-IR spectra taken from agricultural samples, such as
grains, oil, seeds and feeds, etc., have been used to quantitate
various bulk constituents, e.g., total protein, water content, or
fat content. See, P. Williams et al., "Agricultural Applications of
Near-IR Spectroscopy and PLS Processing," Canadian Grain
Commission.
[0006] Multivariate statistical methods have long been used in the
analysis of biomedical samples by infrared and near infrared,
generally under the name "chemometrics." See, U.S. Pat. No.
5,596,992 to Haaland et al., titled "Multivariate Classification of
Infrared Spectra of Cell and Tissue Samples," and U.S. Pat. No.
5,857,462 to Thomas et al., titled "Systematic Wavelength Selection
For Improved Multivariate Spectral Analysis."
[0007] The use of multivariate methods for the analysis of ex vivo
tissue samples is well established. For spectra taken in vivo, some
work has been done. Linear discriminant analysis has been used to
classify visible/near-IR spectra of human finger joints into early
and late rheumatoid arthritis classes. Multivariate methods have
been used to classify fluorescence spectra taken in vivo from
cervixes according to the presence or absence of cervical cancer or
pre-cancerous tissues.
[0008] In general, the field of chemometrics is well established,
and the use of multivariate statistical methods for the analysis of
complex spectra is common. These methods are used in pharmaceutical
analysis, industrial applications, and, more recently, biomedical
spectral analysis.
SUMMARY OF THE INVENTION
[0009] Recently, it has been discovered that glucose levels can be
determined in vivo by measuring fluorescence spectra emitted from
the skin surface following excitation with one or more wavelengths.
See, U.S. patent application Ser. No. 09/287,486, titled,
"Non-Invasive Tissue Glucose Level Monitoring," filed Apr. 6, 1999,
and incorporated in its entirety herein by reference. Particularly,
peak ratios, correlation analysis, and linear regression analysis
have been used to analyze skin autofluorescence spectra for the
purpose of determining the blood glucose concentration. Partial
least squares ("PLS") analysis of near-IR spectra is the basis of
all infrared efforts towards non-invasive glucose monitoring.
[0010] Analysis of collected spectra is complicated by the fact
that it can be difficult to distinguish changes or variations in
the spectra due to skin variables, such as skin inhomogeneity, UV
damage, age, erythema, and the like. The present invention
addresses this problem by providing a method of processing in vivo
skin auto fluorescence spectra to account for these types of
variables.
[0011] Accordingly, one embodiment of the invention is directed to
a method for processing in vivo skin auto fluorescence spectra
emitted by a skin surface of a patient to determine a blood glucose
level of the patient. The method comprises the steps of collecting
auto fluorescent spectra emitted from the skin surface of the
patient, and correcting the collected spectra using multivariate
analysis techniques to account for variables among skin
surfaces.
[0012] Another embodiment is directed to an instrument for
determining a correct glucose level of a patient by measuring in
vivo auto-fluorescence of the patient's skin comprising: means for
irradiating the skin with a plurality of excitation wavelengths;
means for collecting a plurality of emitted wavelengths; and means
for analyzing the collected wavelengths to determine a preliminary
blood glucose level. The means for analyzing comprises a means for
correcting the preliminary blood glucose level to account for
variations in skin using one or more multivariate analytical
techniques to determine the correct glucose level of the
patient.
[0013] In addition, the present invention also relates to methods
of classifying cells or tissue samples, or quantifying their
components, using multivariate analysis of the measured intensities
of a plurality of wavelengths of emitted radiation.
[0014] Other embodiments and advantages of the invention are set
forth in part in the description which follows, and in part, will
be obvious from this description, or may be learned from the
practice of the invention.
DESCRIPTION OF THE INVENTION
[0015] As embodied and broadly described herein, the present
invention is directed to the processing of in vivo skin
auto-fluorescence spectra for the purposes of determining blood
glucose levels. In-vivo fluorescence spectra have been shown to
correlate with blood glucose levels. See, Id. Although large
changes in skin fluorescence spectra due to changes in blood
glucose levels have been observed, it can sometimes be difficult to
separate the variations in the spectra caused by changes in blood
glucose from other spectral changes due to factors such as skin
inhomogeneity, age effects, UV damage, erythema, etc.
[0016] For large subject populations, it is desirable to be able to
determine an algorithm for converting skin fluorescence spectra
into glucose values which works on a large percentage of the
population as opposed to a single individual. Thus, there is a need
for an analysis method that takes into account more spectral
information than that which is found at a single wavelength or
two.
[0017] By analyzing large numbers of spectra from a wide range of
individuals, a useful instrument for the non-invasive monitoring of
glucose using fluorescence excitation spectroscopy may be developed
which accommodates differences in skin. By using multivariate
statistical approaches, a quantitation algorithm useful across many
individuals may be created. Many multivariate techniques are useful
in this regard. Useful analytical methodologies include, but are
not limited to: quantification methodologies, such as, partial
least squares, principal component regression ("PCR"), linear
regression, multiple linear regression, stepwise linear regression,
ridge regression, radial basis functions, and the like;
classification methodologies, such as, linear discriminant analysis
("LDA"), cluster analysis (e.g., k-means, C-means, etc., both fuzzy
and hard), neural network ("NN") analysis; and data processing
methodologies, such as, 1-D or 2-D smoothing filters (based on
median-filtering, mean filtering, discrete cosine, wavelet, or
Fourier transform), Laplacian operators, maximum likelihood
estimators, maximum entropy methods, first and second derivatives
(both in 1-D and 2-D implementations), peak enhancement methods
(such as Fourier self-deconvolution), principal components analysis
as a pre-processing step, and varimax rotations for PLS and PC
methods.
[0018] Other methodologies and data processing methods may further
include sorting data according to their glucose values, followed by
the application of one or more data filtering/smoothing algorithms,
within an individual in a small dataset or within each individual
for larger, multiple-person datasets. Sorting by glucose or any
other relevant analyte has at least two desirable effects: (1) it
groups data with similar glucose values together, so that the
subsequent application of filtering techniques will reduce "noise"
not attributable to glucose, and (2) it reduces temporal
correlation inherent in preserving a dataset as a time series, and
thereby reduces spurious correlation effects.
[0019] In addition or alternately, spectral transformation
algorithms may be applied to each person's data prior to smoothing
or sorting. These transfer functions will enable calibrations made
on spectra from one individual to be more easily transferable to
spectra from another individual or individuals by minimizing the
spectral differences between them. Such algorithms may be as simple
as the ratio of the means of the spectra of two individuals, or
some complex algorithm which takes into account the responsivity
characteristics of each spectrometer.
[0020] Methods of the invention may also include pre-classification
of spectra into categories of glucose levels prior to
quantification. This can be done with any of the supervised
classification methods listed above, e.g., LDA, PCR, NN, and the
like. Sequential binary division of spectra may also be applied,
e.g., above/below 150, then, if below 150, above/below 100, if
above 150, then above/below 200, etc.
[0021] Furthermore, non-linear model fitting techniques can be used
to incorporate prior models of absorption and emission spectra of
known fluorophores, and subsequently use the parameters of the best
fit model as part of the multivariate analysis.
[0022] In addition, methods of the invention may also use
wavelength-selection algorithms to reduce the number of spectral
data points prior to classification or quantitation. Examples of
these methods include genetic algorithm methodologies, step-wise
linear regression and comprehensive combinatorial linear
discriminant analysis, and the like.
[0023] Accordingly, one embodiment of the invention is directed to
a method for processing in vivo skin auto fluorescence spectra
emitted by a skin surface of a patient to determine a blood glucose
level of the patient comprising the steps of collecting auto
fluorescence spectra emitted from the skin surface of the patient
and correcting the collected spectra using multivariate analysis
methods to account for variables among skin surfaces. The
multivariate analysis method may comprise one or more
quantification, classification or data processing methods selected
from the group consisting of partial least squares, principal
component regression, linear regression, multiple linear
regression, stepwise linear regression, ridge regression, linear
discriminant analysis, cluster analysis (k-means, C-means, etc.,
both fuzzy and hard), neural network analysis, smoothing filters
(based on median filtering, mean filtering, discrete cosine,
wavelet and Fourier transform smoothing all in both 1-D and 2-D
methods), laplacian operators, maximum likelihood estimators,
maximum entropy methods, first and second derivatives (both in 1-D
and 2-D implementations), peak enhancement methods such as Fourier
self-deconvolution, principal components analysis as a
pre-processing step, and varimax rotations for PLS and PC
methods.
[0024] Another embodiment is directed to an instrument for
determining a correct glucose level of a patient by measuring in
vivo auto-fluorescence of the patient's skin comprising: means for
irradiating the skin with a plurality of excitation wavelengths;
means for collecting a plurality of emitted wavelengths; and means
for analyzing the collected wavelengths to determine a preliminary
blood glucose level. The means for analyzing comprising means for
correcting the preliminary blood glucose level to account for
variations in skin. The means for correcting comprising using one
or more multivariate analytical methodologies to determine the
correct glucose level of the patient.
[0025] Quantification Models
[0026] The present invention is useful for quantifying components
in a cell or tissue, and may be used, for example, to quantify
tissue glucose levels in vivo. Accordingly, one embodiment of the
invention is directed to a method of quantifying a component of a
cell or tissue sample comprising the steps of: generating a single
excitation wavelength or plurality of different excitation
wavelengths of green to ultraviolet light; irradiating the sample
with the light and measuring the intensity of the stimulated
emission of the sample at a minimum of three different wavelengths
of lower energy than the excitation light or at a plurality of
wavelengths of lower energy than the excitation light; and
quantifying one or more components of the cell or tissue from the
measured intensities by using a multivariate quantification model.
The green to ultraviolet light may be in the green to violet range
of wavelengths, or alternately, it may be in the violet to
near-ultraviolet range of wavelengths.
[0027] The component quantified may be glucose or another desired
component. Irradiating may be done in vivo or in vitro.
[0028] In a preferred embodiment, the step of quantifying the
component of the sample includes at least one spectral data
pre-processing step. In one such embodiment, the pre-processing
step includes at least one of the steps of selecting wavelengths,
correcting for a linear baseline, and normalizing a spectral region
surrounding the different wavelengths, used for classification of
one spectral band in that spectral region. Alternately, the
pre-processing step includes at least one of the steps of
normalizing for total area of the spectrum, filtering or smoothing
the data, or pre-sorting by analyte.
[0029] Multivariate quantification may be done by a partial least
squares technique, by a principal component regression technique,
or by one of multiple linear regression, stepwise linear regression
or ridge regression.
[0030] In a preferred embodiment of this method, the step of
quantifying the component of the sample is performed by a
multivariate algorithm using the measured intensity information and
at least one multivariate quantification model which is a function
of conventionally determined cell or tissue component quantities
from a set of reference samples and a set of spectral intensities
as a function of wavelength obtained from irradiating the set of
reference samples with green to ultraviolet light and monitoring
the stimulated emission.
[0031] Another embodiment of the invention is directed to a method
of quantifying a component of a cell or tissue sample comprising:
generating a single excitation wavelength or plurality of different
excitation wavelengths of mid-ultraviolet light; irradiating the
sample with said light and measuring the intensity of the
stimulated emission of the sample at a minimum of three different
wavelengths of lower energy than the excitation light or at a
plurality of wavelengths of lower energy than the excitation light;
generating at least one multivariate quantification model, said
model quantifying the different components of the sample as a
function of the intensity characteristics at the measured
wavelengths in relation to a reference quantitation result;
calculating the quantity of the component from the measured
intensities by using multivariate quantitation of the intensities
at the at least three different wavelengths based on the
quantitation model; and quantifying the component from the measured
intensities by using said multivariate quantification model.
[0032] As with the previous embodiment, the sample component may be
quantified in vitro or in vivo. Components which may be analyzed
include glucose.
[0033] Preferably, the step of quantifying the component of the
samples includes at least one spectral data pre-processing step.
The pre-processing step preferably includes at least one of the
steps of selecting wavelengths, correcting for a linear baseline,
and normalizing a spectral region surrounding the different
wavelengths, used for classification of one spectral band in that
spectral region. Alternately, the pre-processing step includes at
least one of the steps of normalizing for total area of the
spectrum, filtering or smoothing the data, or pre-sorting the data
by analyte. Multivariate quantification may be done by a partial
least squares technique, by a principal component regression
technique, or by one of multiple linear regression, stepwise linear
regression or ridge regression.
[0034] In a preferred embodiment, the step of quantifying the
component of the sample is performed by a multivariate algorithm
using the measured intensity information and at least one
multivariate quantification model which is a function of
conventionally determined cell or tissue component quantities from
a set of reference samples and a set of spectral intensities as a
function of wavelength obtained from irradiating the set of
reference samples with green to ultraviolet light and monitoring
the stimulated emission.
[0035] The present invention is also directed to a system for
quantifying one or more components of a cell or tissue sample
comprising: means for generating a single excitation wavelength or
a plurality of different excitation wavelengths of green to
ultraviolet light, means for directing at least a portion of the
green to ultraviolet light into the sample; means for collecting at
least a portion of the stimulated emission light after the
excitation light has interacted with the sample; means for
measuring an intensity of the collected stimulated emission light
at least three different wavelengths; means, coupled to the
measuring means, for storing the measured intensities as a function
of the wavelength; means for storing at least one multivariate
quantification model which contains data indicative of a correct
quantification of components of known cell or tissue samples; and
processor means coupled to the means for storing the measured
intensities and the means for storing the model, the processor
means serving as means for calculating the quantity of the
components of the cell or tissue sample by use of the multivariate
quantification model and the measured intensities.
[0036] In one embodiment of the system, the means to direct the
light and the means to collect the light comprise an endoscope.
Alternately, the means to direct the light and the means to collect
the light may comprise a fiber optic bundle. The system may further
include means to determine outliers.
[0037] Classification Models
[0038] The present invention may also be used to classify cells or
tissue samples. For example, one such embodiment is directed to a
method of classifying a cell or tissue sample comprising the steps
of: generating a single excitation wavelength or plurality of
different excitation wavelengths of green to ultraviolet light;
irradiating the sample with said light and measuring the intensity
of the stimulated emission of the sample at a minimum of three
different wavelengths of lower energy than the excitation light or
at a plurality of wavelengths of lower energy than the excitation
light; and classifying the sample as one of two or more cell or
tissue types from the measured intensities by using a multivariate
classification model.
[0039] The green to ultraviolet light may be in the green to violet
range of wavelengths, or alternately, in the violet to
near-ultraviolet range of wavelengths.
[0040] The sample may be classified as normal or abnormal.
Irradiating may be done in vivo or in vitro.
[0041] Preferably, the step of classifying the samples includes at
least one spectral data pre-processing step. For example, the
pre-processing step may include at least one of the steps of
selecting wavelengths, correcting for a linear baseline, and
normalizing a spectral region surrounding the different
wavelengths, used for classification of one spectral band in that
spectral region. Alternately, the pre-processing step may include
at least one of the steps of normalizing for total area of the
spectrum, filtering or smoothing the data, or pre-sorting the data
by analyte.
[0042] Multivariate classification may be done by a linear
discriminant analysis technique. Preferably, the linear
discriminant analysis is preceded by a principal component
analyzing step limiting the number of discriminant variables.
[0043] In a preferred embodiment of the method, the step of
classifying the sample is performed by a multivariate algorithm
using the measured intensity information and at least one
multivariate classification model which is a function of
conventionally determined cell or tissue sample classes from a set
of reference samples and a set of spectral intensities as a
function of wavelength obtained from irradiating the set of
reference samples with green to ultraviolet light and monitoring
the stimulated emission.
[0044] Another embodiment of the invention is directed to a method
of classifying a cell or tissue sample comprising: generating a
single excitation wavelength or plurality of different excitation
wavelengths of mid-ultraviolet light; irradiating the sample with
said light and measuring the intensity of the stimulated emission
of the sample at a minimum of three different wavelengths of lower
energy than the excitation light or at a plurality of wavelengths
of lower energy than the excitation light; generating at least one
multivariate classification model, said model classifying the
sample as a function of the intensity characteristics at the
measured wavelengths in relation to a reference classification;
calculating the classification of the sample from the measured
intensities by using multivariate classification of the intensities
at the at least three different wavelengths based on the
classification model; and classifying the sample as one of two or
more cell or tissue types from the measured intensities by using
said multivariate classification model.
[0045] Classifying may be done in vitro or in vivo. The sample may
be classified as normal or abnormal. Preferably, the step of
classifying of the samples includes at least one spectral data
pre-processing step. For example, the pre-processing step may
include at least one of the steps of selecting wavelengths,
correcting for a linear baseline, and normalizing a spectral region
surrounding the different wavelengths, used for classification of
one spectral band in that spectral region. Alternately, the
pre-processing step may include at least one of the steps of
normalizing for total area of the spectrum, filtering or smoothing
the data, or pre-sorting the data by analyte.
[0046] In one embodiment, multivariate classification is done by a
linear discriminant analysis technique. In this embodiment, the
linear discriminant analysis is preferably preceded by a principal
component analyzing step limiting the number of discriminant
variables.
[0047] In a preferred embodiment of this method, the step of
classifying the sample is performed by a multivariate algorithm
using the measured intensity information and at least one
multivariate classification model which is a function of
conventionally determined cell or tissue sample classes from a set
of reference samples and a set of spectral intensities as a
function of wavelength obtained from irradiating the set of
reference samples with green to ultraviolet light and monitoring
the stimulated emission.
[0048] Another embodiment is directed to a system for classifying
cell or tissue samples comprising: means for generating a single
excitation wavelength or a plurality of different excitation
wavelengths of green to ultraviolet light; means for directing at
least a portion of the green to ultraviolet light into the samples;
means for collecting at least a portion of the stimulated emission
light after the excitation light has interacted with the samples;
means for measuring an intensity of the collected stimulated
emission light at least three different wavelengths; means, coupled
to the measuring means, for storing the measured intensities as a
function of the wavelength; means for storing at least one
multivariate classification model which contains data indicative of
a correct classification of known cell or tissue samples; and
processor means coupled to the means for storing the measured
intensities and the means for storing the model, the processor
means serving as means for calculating the classification of the
cell or tissue samples as one of two or more cells or tissues by
use of the multivariate classification model and the measured
intensities.
[0049] In one embodiment, the means to direct the light and the
means to collect the light comprise an endoscope. Alternately, the
means to direct the light and the means to collect the light
comprises a fiber optic bundle. The system may further include
means to determine outliers.
[0050] In the above embodiments, the means for generating
excitation radiation may be any type of excitation source,
preferably, xenon arc lamps (plus appropriate filters and/or
monochromators); a plurality of laser diodes or LEDs; mercury
lamps; halogen lamps; tungsten filament lamps; or any combination
thereof. Further, appropriate filters and/or monochromators can be
added.
[0051] In addition to using a fiber optic bundle or endoscope,
suitable means for directing or collecting radiation may comprise
any of the following: liquid light guides; system of optical
components (mirrors, lenses, etc.); individual fiber optic cables;
plastic optical components; quartz optical components; or any
combination thereof.
[0052] In the above embodiments, suitable means for measuring an
intensity of the radiation may be selected form the group
consisting of photodiodes; photodiode arrays; avalance photodiodes;
LEDs; laser diodes; charge couple device (CCD) detectors (arrays or
individually); silicon detectors; or any combination thereof.
Suitable storing means may be computers (hardware and software);
EPROMs; programmed firmware; and the like. Further, suitable
processing means may be any type of existing digital processing
devices.
[0053] Other embodiments and uses of the invention will be apparent
to those skilled in the art from consideration of the specification
and practice of the invention disclosed herein. All references
cited herein, including all U.S. and foreign patents and patent
applications, are specifically and entirely hereby incorporated
herein by reference, including, but not limited to, U.S. patent
application Ser. No. 09/287,486, titled "Non-Invasive Tissue
Glucose Level Monitoring," filed Apr. 6, 1999. U.S. patent
application titled "Reduction of Inter-Subject Variation Via
Transfer Standardization," U.S. patent application titled
"Generation of Spatially-Averaged Excitation-Emission Map in
Heterogeneous Tissue," and U.S. patent application titled
"Non-Invasive Tissue Glucose Level Monitoring," all filed
contemporaneously herewith, are entirely and specifically
incorporated by reference. It is intended that the specification
and examples be considered exemplary only, with the true scope and
spirit of the invention indicated by the following claims.
* * * * *