 |
Description  |
|
|
BACKGROUND OF THE INVENTION
Instrumentation has recently evolved which can measure the extremely weak,
time varying magnetic fields outside the human body. These magnetic fields
have been shown to be generated by electrical sources within the human
body.
Magnetic field information is often combined with measurements of the
temporal and spatial variations in the electrical potentials on the body
surface for additional information about the internal condition of the
body. Because the magnetic and electric measurements generate a tremendous
amount of data in a short period of time, it is desirable to reduce the
data and present it in a form which is more familiar and useful to medical
diagnosticians. To accomplish this, it is generally believed that a
presentation of the sources which generate the measured fields and
potentials is desirable. Therefore a procedure which computes the
locations and conditions of the sources from the measured data is
required.
The problem with computing source locations and conditions from the
magnetic field and electric potential measurements, is that the measured
information is insufficient to uniquely determine the sources. In other
words, many different source distributions can give rise to a specific set
of measurements. The difficulty created by this ambiguity is referred to
as "the inverse problem". Therefore it is necessary to incorporate as much
a-priori knowledge as possible into the calculation.
In addition to an accurate description of the source distribution, it is
necessary to be able to show the spatial relationship between the current
distribution and anatomical structures in the subject. While techniques
for "data fusion" have been developed, there remain many problems with
merging complex 3D data sets.
SUMMARY OF INVENTION
In accordance with the present invention a method is provided for locating
a current source distribution in a living creature. Sensors, suitable for
detecting a magnetic field emanating from the living creature, are placed
in various locations outside of the living creature. The living creature
is modeled as a collection of adjoining cubic or rectangular regions.
Certain of these regions or "voxels" are selected (or segmented) manually
using interactive graphical techniques or automatically by choosing a
range of parameter values from the magnetic resonance (MR) data. Once the
segmentation is complete, a "minimum norm" calculation is performed using
the sensor readings and the segmented voxel location information. A
current source distribution is determined as a result of these
calculations and is displayed along with the anatomical data using volume
visualization techniques.
BRIEF DESCRIPTION OF THE DRAWING
While the novel features of the invention are set forth with particularity
in the appended claims, the invention, both as to organization and
content, will be better understood and appreciated, along with other
objects and features thereof, from the following detailed description
taken in conjunction with the drawings, in which:
The sole figure, is a flowchart representation of the present inventive
method.
DETAILED DESCRIPTION OF THE INVENTION
The present inventive method for computing source distributions uses the
minimum norm technique in a modified way so that instead of computing a
response matrix for the entire volume of a patient, the response matrix is
computed only for locations where sources are expected. In order to do
this, the present method takes as input a database of 3-dimensional (3D)
anatomical data such as generated using the method described in commonly
assigned U.S. Pat. No. 4,729,098 which is incorporated herein by
reference. The polygonal mesh generated thereby can be simplified to make
the present invention more efficiently practiced using a method such as
that described in commonly assigned co-pending application Ser. No.
07/815,772, which is incorporated herein by reference.
The sole FIGURE is a flowchart representation of the present invention. The
process begins at function block 110 where an MR or CT scan is performed
on a subject, resulting in 3D anatomical data file 112 of the subject in
the form of N voxels. At 114, a segmentation of the 3D data is done to
reduce data file 112 to data file 116 containing M voxels selected from
the N voxels of file 112. At 118, S sensor locations are specified and
measurements of magnetic field are taken. At 120, a sequence of
single-dipole, forward calculations are performed by placing a dipole at
each of the M tissue voxels and for each of 3 orthogonal directions.
Therefore a total of 3M distinct sets of forward calculations are used.
For each single dipole, the calculated responses at each sensor forms an
S-dimensional vector. By grouping the vectors, a 3MxS dimensional matrix
122 is created. Next at 124, a singular value decomposition is performed
on this matrix. The eigenvectors whose eigenvalues are above a preselected
threshold are retained as "basis vectors". The remaining eigenvectors are
discarded. The basis vector(s) 126 can be interpreted as a weighted sum of
sources. Therefore each basis vector represents a distributed
electromagnetic source. Because the basis vectors are orthogonal and span
the measurement space, they can be used to uniquely represent any
measurement. At 128, the measurements are expressed over time in terms of
basis vectors 126, resulting in a distributed sum of sources which evolves
over time. Finally, the computed current distribution is displayed with
the anatomical data at 130.
As an example of the present invention, for evoked sources in the brain, it
is commonly assumed that the signal arises somewhere in the cerebral
cortex. Therefore, in this case, only sources in the cortex need to be
considered. This portion of the brain can be grossly identified from the
difference in proton density between the brain and the neighboring cerebro
spinal fluid. Voxels near the cortex can be selected manually using
interactive graphical techniques or they can be selected automatically by
choosing a certain range of parameter values from the MR data.
The inventive method begins by mathematically dividing the body volume into
a collection of adjoining cubic or rectangular regions. Each region or
"voxel" is represented by the numeric co-ordinates of its center or one of
its corners. In this way the body volume can be represented as a finite
list of coordinate numbers. An electrical current distribution in the body
volume is represented through another list associated to the coordinate
list, where each set of numbers in the associated list describes the
average current intensity and direction within each voxel. If the number
of voxels is denoted by N, then 3N numbers are needed to completely
specify the current distribution, since 3 numbers can uniquely specify the
current within each voxel. Stated another way, any current distribution
can be completely represented as a vector of length 3N.
To complete the model setup, the positions of the sensors (to detect the
magnetic fields and electrical potentials) are fixed with respect to the
body. Typically, there are dozens of sensors. The number of sensors will
be denoted by S. Without loss of generality, it can be assumed that each
sensor measures a scalar component of the field (or potential) at a
specific spatial location. Therefore, the measurement at a specific
instant in time can be represented by a vector of length S.
The "response function" of the system can be represented by a matrix, such
that if the vector specifying the current distribution is multiplied by
the matrix, a new vector of length S is formed which gives the numerical
value measured by each detector at a given instant of time:
(response vector)=(current vector).times.[response matrix] (1)
As the current changes in time, the response vector changes in a
corresponding way.
The current distribution is a "row" vector and the measured response is a
"column" vector. In this case, the response matrix is a 3N.times.S matrix.
The response matrix can be assumed to be static or at least change much
more slowly than the currents. Clearly, the physical information, such as
the detector characteristics and the source and sensor geometry, is
embedded in the response matrix. A description of how the response matrix
is obtained will be provided below.
Since the response vector can be computed from the current vector, equation
(1) suggests that the current vector can be computed from the measured
response by multiplying both sides of the equation by the inverse of the
response matrix. Unfortunately, the response matrix has no inverse, partly
because the sensor cannot measure linearly independent information within
practical ranges of accuracy, and partly because the matrix does not have
the same number of rows and columns, which is a necessary condition. The
best that is possible, is that a "singular value decomposition" (SVD) can
be performed on the response matrix. (A singular value decomposition is a
standard mathematical procedure which is described in many linear algebra
textbooks.) This will provide a matrix which can be used to uniquely
decompose a measured response into a distributed current source, so that:
(respo | | |
