|
Medical Patent Abstract
An apparatus and method for medical imaging, particularly for mammography,
wherein a body organ, such as a breast, is exposed to X-rays and
the X-rays are collected after attenuation through the object. The
recorded attenuations are processed and displaying a result of this
processing in the form of a representation of an image of the object.
The processing of the recorded attenuations form includes automatic
classification of zones of the breast into pathological or non-pathological
classes. The automatic classification takes into account at least
one classification input into the apparatus in advance in association
with data that can be collected by the apparatus, and using this
prior classification as a reference in order to produce a classification
of the same type if there is similarity between the collected data
and the data associated with this reference classification.
Medical Patent Claims
What is claimed is:
1. An apparatus for medical imaging comprising: means for exposing
an object to radiation; means for collecting the radiation after
attenuation through the object; means for processing recorded attenuations;
and means for displaying a result of the processing in the form
of a representation on an image of the object; wherein the means
for processing the recorded attenuations forms means for automatic
classification of zones of the object into pathological or non-pathological
classes, the means for automatic classification taking account of
at least one classification input into the apparatus in advance
in association with data that can be collected by the apparatus,
and using this prior classification as a reference in order to produce
a classification of the same type if there is similarity between
the collected data and the data associated with this reference classification;
and wherein the means for automatic classification makes a classification
of pixels of an image of the object at the same location in a several
dimensional space, the dimensions of this space each corresponding
to the value of a grey level at a given instant of a contrast appearance
rate at the pixel considered.
2. The apparatus according to claim 1 wherein the several dimensional
space further comprises dimensions that each represent a parameter
among the maximum slope of a contrast variation, the value of a
maximum contrast reached, a hold duration of the contrast at its
maximum value, at the pixel considered.
3. The apparatus according to claim 2 wherein the several dimensional
space further comprises at least two dimensions that each represent
the measured signal for two different radiation energies at the
pixel considered.
4. The apparatus according to claim 1 wherein the several dimensional
space further comprises at least two dimensions that each represent
the measured signal for two different radiation energies at the
pixel considered.
5. The apparatus according to claim 4 wherein the two different
radiation energies are located on opposites sides of a sudden change
in an attenuation coefficient of a contrast medium.
6. The apparatus according to claim 1 wherein the means for automatic
classification takes account of at least one manual classification
of a current image zone of the object as reference classification;
and automatically establishes the same classification in case of
similarity with other data collected on the same object.
7. The apparatus according to claim 1 wherein the means for automatic
classification includes means for learning; and carrying out recording
a collection of reference information comprising several reference
classifications associated with data that can be collected by the
apparatus.
8. The apparatus according to claim 1 comprising: means for enabling
a user to confirm or contradict an automatic classification made
by the apparatus; and means for taking account of this confirmation
or contradiction in order to incorporate the results of this confirmation
or contradiction as a reference classification.
9. The apparatus according to claim 1 wherein the means for classification
identifies incoherent spatial variations in the classification,
and modifying the classification of some locations in case of such
incoherent spatial variations.
10. The apparatus according to claim 1 comprising means for accounting
of one item of data among the number of changes in the sign of the
variation in grey levels, the age of the object, weight or medical
data, when creating a classification.
11. The apparatus according to claim 1 wherein the means for automatic
classification records a vector corresponding to each pixel location
in the image of the object, each vector comprising the dimensions
of the several dimensional space that correspond to grey levels
of the pixel at different times that correspond to an appearance
and disappearance of a contrast medium.
12. A method for medical imaging comprising: exposing an object
to radiation; collecting the radiation after attenuation through
the object; processing recorded attenuations comprising automatic
classifying of zones of the object into pathological or non-pathological
classes, the automatic classification taking account of at least
one classification input in advance in association with data that
can be collected, and using this prior classification as a reference
in order to produce a classification of the same type if there is
similarity between the collected data and the data associated with
this reference classification; and displaying a result of the processing
in the form of a representation on an image of the object the process;
wherein the automatic classification makes a classification of pixels
of the image of the object at the same location in a several dimensional
space, the dimensions of this space each corresponding to the value
of a grey level at a given instant of a contrast appearance rate
at the pixel considered.
13. The method according to claim 12 wherein the several dimensional
space further comprises dimensions that each represent a parameter
among the maximum slope of a contrast variation, the value of a
maximum contrast reached, a hold duration of the contrast at its
maximum value, at the pixel considered.
14. The method according to claim 13 wherein the several dimensional
space further comprises at least two dimensions that each represent
the measured signal for two different radiation energies at the
pixel considered.
15. The method according to claim 12 wherein the several dimensional
space further comprises at least two dimensions that each represent
the measured signal for two different radiation energies at the
pixel considered.
16. The method according to claim 15 wherein the two different
radiation energies are located on opposites sides of a sudden change
in an attenuation coefficient of a contrast medium.
17. The method according to claim 12 wherein the automatic classification
takes account of at least one manual classification of a current
image zone of the object as reference classification; and automatically
establishes the same classification in case of similarity with other
data collected on the same object.
18. The method according to claim 12 wherein the automatic classification
includes learning; and carrying out recording a collection of reference
information comprising several reference classifications associated
with data that can be collected by the apparatus.
19. The method according to claim 12 comprising: enabling a user
to confirm or contradict an automatic classification; and taking
into account of this confirmation or contradiction in order to incorporate
the results of this confirmation or contradiction as a reference
classification.
20. The method according to claim 12 wherein the classification
identifies incoherent spatial variations in the classification,
and modifying the classification of some locations in case of such
incoherent spatial variations.
21. The method according to claim 12 comprising: accounting for
one item of data among the number of changes in the sign of the
variation in grey levels, the age of the object, weight or medical
data, when creating a classification.
22. A computer program product comprising a computer readable medium
having computer readable program code means embodied in the medium,
the computer readable program code means implementing the method
according to claim 12.
23. An article of manufacture for use with a computer system, the
article of manufacture comprising a computer readable medium having
computer readable program code means embodied in the medium, the
program code means implementing of the method according to claim
12.
24. A program storage device readable by a computer tangibly embodying
a program of instructions executable by the computer to perform
the method according to claim 12.
25. The method according to claim 12 wherein the automatic classification
records a vector corresponding to each pixel location in the image
of the object, each vector comprising the dimensions of the several
dimensional space that correspond to grey levels of the pixel at
different times that correspond to an appearance and disappearance
of a contrast medium.
Medical Patent Description
CROSS-REFERENCE TO RELATED APPLICATIONS
This application claims the benefit of a priority under 35 USC
119(a)-(d) to-French Patent Application 04 05524 filed May 21, 2004,
the entire contents of which are hereby incorporated by reference.
BACKGROUND OF THE INVENTION
An embodiment of the invention relates to a method and apparatus
for medical imaging and in particular to a method and apparatus
for the classification of pixels in medical imaging. An embodiment
of the invention relates to Contrast Medium-enhanced Mammography
(CMM) by X-rays and the injection of a contrast medium.
Mammography is medical imaging intended particularly for the detection
of tumors by examination of successive images taken to reveal the
variation with time of impregnation of the contrast medium and its
gradual disappearance. In mammography the contrast medium tends
to attenuate X-rays significantly more than a non-impregnated tissue,
and thus reveals particularly vascularised zones such as tumors.
But the variation of contrast within the breast itself provides
an important indication about whether or not tumors are present,
by the rate at which this contrast appears and disappears.
At present, contrast medium-enhanced mammography is practiced within
the context of MRI, a technique that comprises making molecules
composing the examined organ vibrate. Within this context, the variation
of contrast in the breast is displayed on the screen in the form
of a sequence of images that the practitioner interprets based on
experience, as revealing or not revealing the presence of a tumor.
Marx et al., "Contrast-enhanced digital mammography (CEDM):
phantom experiment and first clinical results", Proc. SPIE--International
Soc. for Optical Engineering, vol. 4682, pp. 174-181, 2002, proposes
to produce maps representing the distribution of some parameters
in the breast. These parameters are measurements illustrating some
kinetic aspects of contrast variation obtained from a sequence of
X-ray images.
However, the diagnosis work to be done by the practitioner is still
considerable.
BRIEF DESCRIPTION OF THE INVENTION
According to an embodiment of the invention, an apparatus comprises
means for exposing an object, such as a body organ, e.g., the breast,
to a source of radiation, such as an X-ray beam; means for collecting
the radiation after attenuation through the object; means for processing
recorded attenuations and means for displaying the result of this
processing in the form of a representation on an image of the object.
The means for processing the recorded attenuations form means for
automatic classification of zones of the object into pathological
classes The means for automatic classification is suitable for taking
account of at least one classification input into the apparatus
in advance in association with data that can be collected by the
apparatus, and using this prior classification as a reference in
order to produce a classification of the same type if there is similarity
between the collected data and the data associated with this reference
classification.
An embodiment of the invention is a method for medical imaging,
particularly for mammography, comprising exposing an object, such
as a body organ, e.g., a breast, to radiation; collecting the radiation
after attenuation through the object; processing recorded attenuations
and displaying the result of this processing in the form of a representation
on an image of the breast, the processing of the recorded attenuations
comprises automatically classifying zones of the object into pathological
classes, the automatic classification taking into account at least
one prior classification input in association with data that can
is collected, and using this prior classification as a reference
in order to produce a classification of the same type if there is
similarity between the collected data and the data associated with
this reference classification.
BRIEF DESCRIPTION OF THE DRAWINGS
Other characteristics, purposes and advantages of the invention
will become clear after reading the detailed description given below
with reference to the attached figures among which:
FIG. 1 shows a time axis representing different instants at which
images are taken during impregnation/deimpregnation of a contrast
medium;
FIG. 2 shows the variation of a grey level measured during impregnation/deimpregnation
of a breast by the contrast medium;
FIG. 3 is a diagram showing a distribution of points in a several
dimensional space used for identification of a classification of
a local zone of the breast; and
FIG. 4 is a time axis illustrating the use of two photos with two
different energies in an embodiment of the invention.
DETAILED DESCRIPTION OF THE INVENTION
An embodiment of the invention is to improve the way in which the
practitioner is assisted in using X-rays for contrast medium enhanced
mammography, in diagnosing the presence of a particular pathology,
and particularly for identification of the presence of malignant
tumors
In this description, the term "grey level" will denote
a value representing as possible the attenuation recorded in the
presence of the contrast medium. In practice, these values are obtained
after application of a logarithm to the attenuation actually recorded,
since in the known manner the attenuation induced by the presence
of a contrast medium, typically a product containing iodine, is
exponential to the local concentration of the product. The logarithm
thus applied outputs a value approximately proportional to the attenuation
due to the product containing iodine after passing through the breast,
in other words the thickness actually impregnated by the product
containing iodine.
In a first variant, each point on the image (or pixel) of the examined
breast is associated with a vector in n dimensions, in which each
dimension corresponds to a different observation instant of this
same pixel. In other words, this vector associated with each pixel
represents the rate at which the contrast appears in this particular
pixel.
Thus, for each point located at the same location in each successive
image during impregnation/disappearance of the contract medium,
there is a vector X.sub.i,j associated with this point for which
each of the components G.sub.n(i,j) correspond to the grey level
recorded at each successive instant. N is the number of successive
sequences, and i,j are the coordinates of the same pixel in each
successive image in the sequence of images.
The result is thus a vector X.sub.i,j defined as follows:
.function..times..times..function. ##EQU00001##
Therefore the coefficients of this vector are distributed from
G.sub.1 to G.sub.n(i,j) and are representative of grey levels obtained
in instants t.sub.1 to t.sub.n.
The first variant uses these vectors X.sub.i,j to identify a similarity
between them and vectors representing a typical variation in contrast
with time in the presence of a specific pathology. More generally,
the objective is to sort the different vectors corresponding to
different points into classes that could reveal the existence of
some pathologies.
In one embodiment, these various vectors are classified into four
categories. A first category comprises vectors that could reveal
the presence of a malignant tumor at the pixel i,j considered. A
second category comprises the vectors that could indicate the presence
of a benign tumor at the pixel i,j considered. A third category
comprises vectors that could indicate the presence of healthy tissue
(parenchyma) at the pixel i,j considered. A fourth category comprises
vectors that could indicate the presence of a blood vessel at the
pixel i,j considered.
In another embodiment, with the purpose of detecting tumors, the
first and second classification categories (malignant tumors and
benign tumors) may be coincident.
The various categories may also be distributed into vessels, tumors,
and normal tissue, for detection purposes.
The following processing can be applied in order to determine which
of these categories is applicable.
Each vector may be considered to belong to an n dimensional space,
in which each dimension represents a given instant. The position
of the point according to this dimension then represents the value
of the grey level observed at the instant corresponding to this
dimension. This type of space is shown in FIG. 3, in two dimensions
to simplify the illustration. Therefore, these two dimensions correspond
to two images at two different instants. A vector X.sub.i,j will
be located on a median at 45.degree. if the values of the grey levels
are the same at instants t.sub.1 and t.sub.2.
In this case, the instant t.sub.1 was an image instant at which
no contrast medium had yet been impregnated in the breast, and it
can be understood that the point would belong to, the oblique line
at 45.degree. if the contrast medium were not present at instant
t.sub.2 either. These points located on the oblique may also be
located on zones of the breast in which vascularization is observed
to be negligible or non-existent. The points thus positioned are
classified as "parenchyma" in FIG. 3.
Thus, FIG. 2 shows the variation in the grey level as a function
of time, depending on whether the point at which this grey level
is observed forms part of a common tissue (parenchyma), a vessel,
a malignant tumor or a benign tumor:
On the other hand, a vector X.sub.i,j will be positioned further
above this oblique when the impregnation at time t.sub.2 is greater.
Two oblique bands 10 and 20 are shown, one band 10 close to the
median passing through the origin, and the other 20 further towards
the top. Therefore, the low band 10 represents a location in the
breast in which the impregnation at time t.sub.2 is relatively low.
Therefore, the highest band 20 represents locations within the breast
at which impregnation are already very high at time t.sub.2.
It is considered that points with low impregnation at time t.sub.2
(low band 10) correspond to the presence of a tumor, while points
with high impregnation at time t.sub.2 (high band 20) correspond
to the presence of a vessel at the point considered.
It should be noted now that it is known that malignant lesions/tumors
cause a very fast increase in the contrast, followed by a constant
period, and then fast disappearance of the contrast. It should be
noted also that benign lesions/tumors are marked by a gradual increase
in the contrast. It should be noted also that vessels are obviously
affected by fast contrast variations. Other tissues are less sensitive
to contrast variations.
When considering a number n of successive images, the same processing
is performed but this time in a space with n dimensions. The zones
corresponding to the different classification categories are then
zones in this space with n dimensions.
In one embodiment, the vectors thus localized on particular classification
zones are preferably vectors obtained after preprocessing. One desirable
preprocessing comprises subtracting using an initial vector corresponding
to an image taken without the presence of a contrast medium (this
initial image is called the mask). For example, another type of
preprocessing may comprise noise elimination filtering.
The classification may also be made on normalized data to compare
image sequences acquired under different conditions. Data may be
normalized to compensate for radiation conditions at different energies.
Data may also be normalized to compensate for a variable breast
thickness.
Additional components may also be provided in the vector, such
as the number of sign changes in recorded grey levels during the
image sequence, or such as the patient's age, weight or any other
data related to the patient's medical history. This data is also
integrated in the n dimensional space, each time in the form of
an additional dimension subsequently used for determining classifications.
In one variant, the vector X.sub.i,j also includes the coordinates
of the pixel considered in space. This embodiment can avoid incoherent
classification variations such as sudden classification changes
in nearby pixels.
In another embodiment, the dimensions of the classification space
do not necessarily correspond to a sequence of measurement instants.
Each dimension is dedicated to positioning in this space of a value
of a kinetic parameter calculated on the contrast variation. Thus,
one of the dimensions can be dedicated to the maximum recorded value
of the slope while determining the contrast at the pixel considered.
Another dimension can represent the maximum value of the contrast
recorded at the same pixel considered. Another dimension can represent
the hold duration of the maximum contrast at the same pixel considered.
In this variant, the m parameters thus represented in the n dimensional
space can easily be compared with data from previous sequences of
images, including when these images were taken at different times,
in other words at different number of times t.sub.1 . . . t.sub.n
or with a variable distribution in time.
Thus, FIG. 1 shows two image sequences (corresponding to the upper
triangles and the lower triangles respectively) that can be compared
more easily because these kinetic parameters have been produced,
although the images were not taken at the same instants.
According to a second embodiment of the invention, the space in
which the vectors X.sub.i,j are shown is a two dimensional space,
in which these two dimensions correspond to different radiation
energies used at different times or at the same time. In this variant,
the two instants are preferably very close to each other, in other
words in practice as close as possible.
This embodiment provoked a contrast difference between these two
images, due either to a different reaction of the same dose of the
contrast medium facing two different radiation energies. For example,
one of the radiations is located at about 25 to 35 keV, while the
other is about 40 to 49 keV. Thus advantage is taken that a contrast
medium, typically a product containing iodine, has a capacity to
attenuate X-rays that varies as a function of the energy in the
rays passing through it.
It is known that the attenuation coefficient p varies as a function
of the energy of the X-rays according to a variation law by which
the value of .mu. suddenly changes at a precisely determined energy,
this sudden change currently being called the K-edge. Thus, when
the two energies are located on the opposite sides of this K-edge,
the difference in contrast is particular high between the two acquisitions.
Consequently, at pixels in a position corresponding to a strong
presence of a substance containing iodine, the contrast will be
sensitive to the variation of energy between the two images. On
the other hand, zones without this impregnation will only have a
small reactivity to the energy variation.
These two acquisitions, preferably very close, are more generally
made at an optimum instant for observing such contrasts and their
differences, after the injection of the contrast medium. Thus in
this approach, the kinetic acquisition is replaced by a double energy
acquisition, the two images being acquired at different radiation
spectra (and therefore at different energies). One of the spectra
advantageously corresponds to a normal energy level for a conventional
mammographic examination, the other spectrum for example being a
spectrum typically used in the context of an enhanced contrast method.
The contrast for pixels with a low impregnation will be similar
at times t.sub.1, and t.sub.2, and will produce vectors X.sub.i,j
close to the oblique at 45.degree. passing through the origin. Pixels
i, j with strong impregnation will correspond to vectors X.sub.i,j
well above the oblique.
The variable height position of the vectors X.sub.i,j makes it
possible to classify them in different zones depending on the classification
category mentioned above to which they belong, if any. Consequently,
images taken at instants t.sub.1 and t.sub.2 within the kinetic
of the impregnation/deimpregnation reaction are particularly revealing
of the different categories.
In the above, we described the application of two radiations at
different energies chosen to be on each side of the sudden change
in the attenuation coefficient. However, this approach is also possible
even if the two energies are not on opposite sides of the K-edge.
Thus, a contrast difference can also be used when it is due to the
continuous variation of the attenuation coefficient as a function
of the radiation energy, in other words when the two energies chosen
are located in the typical part of the variation of the attenuation
coefficient, and not on opposite sides of the K-edge.
Double energy acquisitions may be carried out many times while
the contrast is increasing/reducing, and be analyzed in a space
with 2n dimensions like the spaces mentioned above. Recommendations
for spatial consistency, the use of data applicable to the patient,
pre-processing of vectors, normalization of data, use of kinetic
parameters derived from the variation in contrast differences, may
also be applicable in this "double energy" variant.
We will now describe the operation of a means for processing capable
of making the classification in one of the spaces with two dimensions
or n dimensions described above. This means for processing are means
capable of acquiring reference data used subsequently for automatic
production of the classification. To achieve this, this means (apart
from conventional data processing equipment) could implement a network
of neurons or a machine with support vectors. This means will use
initial information input into the system as a reference result.
This information is defined as reference information preferably
contains vectors X.sub.i,j like those defined above that can be
used in the classification space with n or with m dimensions. Therefore,
mean for classification are intended to be able to input vectors
that can be used according to any one of the disclosed embodiments,
and take account in the device of the fact that these input vectors
correspond to a pixel belonging to one of the classification categories.
A first operating mode comprises learning or training in the means
for processing by inputting a collection of test data with predefined
and associated classifications, in a preliminary phase. Thus, in
a first embodiment, there are distinct implementation steps for
the apparatus and method performed at different times. One step
comprises acquisition of learning data. Another step is how to use
the apparatus, in other words, application of learning acquired
on specific acquisitions.
In the variant in which the means for processing use vectors comprising
successive grey levels, the training vectors will comprise a series
of successive grey levels at the pixels considered. Each of these
vectors is associated with the data according to which the corresponding
pixel belongs to one of the classification categories, in a predefined
manner.
A vector encountered afterwards will be categorized as belonging
to the same class as one of the reference vectors if it is similar
to this reference vector, for example, at a distance less than a
predetermined threshold in the n dimensional space.
The same approach will be applied in the case in which the vector
comprises kinetic parameters derived from successive grey levels,
in other words, parameters such as the slope or the maximum grey
level.
This learning is also applicable in the case of vectors representing
double energy images. The reference vectors (in this case learning
vectors) include the results of two contrast readings with different
energy for examinations carried out later and the classification
results assigned by a visual diagnosis made on these readings by
a practitioner or by a laboratory analysis.
According to one variant, means for automatically establishing
a classification of zones encountered are provided, while remaining
under the guidance of the practitioner. In this approach, the means
for processing displays the sequence of images produced. The practitioner
examines the sequence of images and identifies at least one zone
representative of each class, by experience. The means for processing
uses these manual identifications to compare the remainder of the
image with the zones thus classified. If the image sequence reveals
other zones that appear similar to those identified by the practitioner,
then the method and apparatus classifies these zones in the same
categories that the practitioner selected for the zones used as
reference.
This similarity is identified in the same way as described above,
using vectors associated with pixels identified by the practitioner
as reference data. The method and apparatus displays the zones considered
as being similar and submits this result to the practitioner. In
this case, the reference data defining the classes are at least
partly defined directly by the practitioner.
In another operating mode, the method and apparatus combined the
two approaches mentioned above. In this case, the means for processing
makes automatic classification starting from a learning done earlier.
The result is displayed on the screen in the form of a map identifying
the different zones corresponding to the different classes. In a
further step, the user confirms or contradicts the classification
made on these different zones. The means for processing takes into
account this confirmation or contradiction made by the practitioner.
The method and apparatus integrates data learned earlier and the
information comprising data reclassified by the practitioner, when
a new automatic classification is necessary.
The processing may then be repeated on the same sequence starting
from the learning thus updated. In other words, the means for learning
means is reactivated after a first automatic classification to include
additional learning data like those introduced by the practitioner
in the form of confirmations or contradictions of the first result.
The various means described above, for which a classification will
be automatically output, may for example be used under the control
of software capable of carrying out the various processing steps
when it is implemented on an appropriate processor.
Obviously, the various arrangements or processing described above,
and others comprising improvements thereof, can be combined differently
in each of the disclosed embodiments to achieve the same result.
One skilled in the art may make or propose various modifications
to the structure/way and/or function and/or results and/or steps
of the disclosed embodiments and equivalents thereof without departing
from the scope and extant of the invention.
|