JPH04319334A - Biomagnetism imaging system - Google Patents

Abstract

PURPOSE:To arrange that the surmise of an electric current source position in a living body may be conducted in a real time and displayed by using an SQUID for a magnetic sensor. CONSTITUTION:At a multi-channel living body magnetic imaging device in which the distribution of a faint magnetic field generated from a living body is measured by using a superconductive quantum interference element (SQUID) that is a magnetic sensor, and a living body inside active electric current location is surmised from the measured data, and its distribution is conducted with imaging, a living body electric current source position is surmized by using a neural network 2. At this time, the analogue voltage or digital signal of an SQUID magnetic flux meter which is proportional to a received magnetic field strength of each pick up coil is used as the input of the neural network 2, and as the output of the network 2, analogue voltage proportional to mutually vertical 3 directional positions from a reference point where the electric current source is and to the 3 directional components of electric current intensity, or electric current density at two-dimensional and three-dimensional dispersion points, is outputted and displayed at a display.

Description

[Detailed description of the invention]

0001

[Industrial Application Field] The present invention relates to a superconducting quantum interference device (
In a biomagnetic imaging device using SQUID) as a magnetic sensor, a neural network is used to estimate the position of a current source in the living body.
The present invention relates to a biomagnetic imaging device that images the results.

With the recent development of superconducting device technology,
Magnetic measurement devices using superconducting quantum interference devices (SQUIDs) are being applied as medical diagnostic devices. In order to determine the position that contributes to heart disease or brain disease from the measured magnetic field, it is necessary to solve an inverse problem for the position of the current source, that is, to determine the position of the current source where the magnetic field is generated from the measured magnetic field. It is necessary to estimate the position by solving an estimation problem.

In this case, in order to put it to practical use as a medical diagnostic device, it is necessary to estimate the position of the current source in real time and display it on a display.

0004

2. Description of the Related Art FIG. 8 is a diagram illustrating a conventional biomagnetic imaging apparatus. In conventional biomagnetic imaging devices, ■ In order to estimate the current source in the heart, a homogeneous infinite conductor is used as the shape data, and in the case of the brain, a homogeneous or multilayer concentric conductor sphere is used, and the operation shown in Fig. 8 is performed. As shown in the flow, a virtual current dipole (current element) was used as a current source, the magnetic field generated from it was calculated, and the current dipole with the smallest difference from the actual measured magnetic field was selected as the current source position.

[0005]

[Problems to be Solved by the Invention] However, in such a method of comparing the difference between the calculated magnetic field and the measured magnetic field, the calculation does not converge within a finite number of times, so it takes time to accurately determine the current source.

[0006] Therefore, it is difficult to realize a real-time device that determines the position of the current dipole while making measurements and displays its movement. In view of the above conventional drawbacks, the present invention aims to provide a biomagnetic imaging device that displays the movement of a current source in real time by using a neural network as a method for estimating a current source.

[0007]

[Means for Solving the Problems] FIG. 1 is an explanatory diagram of the principle of the present invention, and FIG. 1(a) shows the principle of estimating the position of a current dipole using a neural network 2. b) shows the principle of the neuron model. The above problems are solved by a biomagnetic imaging device configured as follows.

(1) A superconducting quantum interference device (SQUID) 14, which is a magnetic sensor, is used to measure the distribution of a weak magnetic field generated from a living body 16, and from the measurement data, the position of active current inside the living body is determined. This is a multi-channel biomagnetic imaging device that estimates and images the distribution of a signal (VB) corresponding to the magnetic field of the bioelectric current source.
1, VB2, .

(2) In the biomagnetic imaging device described in item (1) above, a superconducting quantum interference device (SQUID) is used as an input of the neural network 2 and outputs in proportion to the magnetic field strength received by each pickup coil 13. ) Magnetometer 1 analog voltage (VB1, V
B2, ~), and as the output of the neural network 2, three mutually perpendicular signals from a reference point with a current source are
An analog voltage (V
x, Vy, Vz, VQx, VQy, VQz).

(3) In the biomagnetic imaging device described in item (1) above, as a learning method for the neural network 2, a pseudo current serving as a current source is placed in a phantom 19 having a boundary resembling the shape of a living body 16. The dipole 17 is arranged and the weighting coefficient (Wji) of each neuron element 8 of the neural network 2 is determined based on the magnetic field generated therefrom.

(4) In the biomagnetic imaging apparatus described in (1) and (3) above, the phantom 19
The configuration is made such that the shape of is variable. (5) In the biomagnetic imaging device according to item (1) above, the input of the neural network 2 is a superconducting quantum interference device (SQUID) magnetometer 1a that outputs a number of pulses proportional to the magnetic field strength of the living body.
The configuration is configured so that the pulse output is converted into a counted digital value.

(6) In the biomagnetic imaging device according to item (1) above, each of the pickup coils 13 is used as an input of the neural network 2.
A superconducting quantum interference device (SQ) that is proportional to the magnetic field strength received by
UID) Magnetometer 1 analog voltage (VB1, VB2,
~) is used to set the value of current density Q (xn, yn, zn, or xn, yn) at three-dimensional or two-dimensional discrete points as the output of the neural network 2.

[0013]

[Operation] FIGS. 1 and 2 are diagrams explaining the principle of the biomagnetic imaging apparatus of the present invention. As shown in FIG. 1(a), in the present invention, the neuron element shown in FIG. 2(b),
That is, an element that multiplies input Yi by weighting coefficient Wji (weighting coefficient setting element) 5. Adder that adds each multiplied value and outputs Xj = ΣWji*Yi
A neural network 2 is formed by combining a plurality of neuron elements 8 in multiple stages that produce a nonlinear output (based on a nonlinear characteristic circuit 7) as shown in the figure depending on the values of 6 and Xj. By having a computer learn to produce a predetermined output, we construct a processing system that can respond to multiple input patterns.

[0014] Then, as shown in Fig. 1(a), the living body 1
6. Magnetic fields B1, B2, B3, . ．． ．． ,Bn
A multi-channel superconducting quantum interference device (SQUID)
) Voltage V measured by magnetometer 1 and proportional to input magnetic flux
B1, VB2, VB3, . ．． ．． ．． , VBn are input to the neural network 2.

[0015] Neural network 2 is a superconducting quantum interference device (SQUID) magnetometer based on the pick-up coil arrangement (r), current source position (r'), and magnetic field strength (
B) is expressed as the well-known Biot-Savart law B=μQ (r-r') /4π|r-r'|3 where, μ: Magnetic permeability in vacuum Q: Current intensity of current dipole (ampere)・Meter)r
: Coordinates of the pickup coil r': Coordinates of the current dipole. However, the above Q, r, r' are mutually perpendicular three-directional components of the analog voltage corresponding to the coordinate (r') of the current dipole calculated from the vector value. (Vx,
Vy, Vz) and the mutually perpendicular three-directional components (VQx, VQy, VQz) of the analog voltage corresponding to the magnetic field strength B generated by the current dipole, by learning them as teacher patterns. source position (x, y, z)
Information regarding magnetic field strength (Qx, Qy, Qz) can be obtained as electrical signals (Vx, Vy, Vz, VQx, VQy, VQz), and from the electrical signals, the signal processing circuit 3
By creating a scanning signal and a brightness signal for the display 4 and displaying them on the display 4, it becomes possible to display the position of the current source and the magnetic field strength. By continuously performing the above operations every time, it becomes possible to display a moving image of the current source.

[0016]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Examples of the present invention will be described below in detail with reference to the drawings. The above-mentioned FIGS. 1 and 2 are diagrams explaining the principle of the present invention, and FIGS. 3 and 4 are diagrams showing an embodiment of the present invention, and FIG.
Fig. 4(b) shows the case where the weighting coefficient (Wji) of the neural network is calculated by a computer.
5 shows a case where the weighting coefficient (Wji) of a neural network is calculated by a computer using a phantom having the same boundary conditions as the human body, and FIGS. Figures 5(a1) and (b1) show the case where the current density at three-dimensional discrete points is set as the output of the neural network, and Figures 6(a2) and (b2)
) shows the case where the current density at two-dimensional discrete points is set as the output of the neural network, and FIG. 7 shows the case where digital values are input as the input of the neural network.

In the present invention, a superconducting quantum interference device (SQUID) 14, which is a magnetic sensor, is used to measure the distribution of a weak magnetic field generated from a living body, and from the measurement data, the active current inside the living body is determined. In a multichannel biomagnetic imaging device that estimates and images its distribution, the position of a bioelectric current source is estimated using a neural network 2. At this time, each pickup coil 13 is used as an input to the neural network 2.
A superconducting quantum interference device (S
QUID) Using the analog voltage or digital signal of the magnetometer 1, the output of the neural network 2 is the position (x, y, z) in three mutually perpendicular directions from a reference point of the current source, and the current intensity. 3-directional component (Qx
, Qy, Qz)
Vz, VQx, VQy, VQz), or current density Q(xn, yn, zn,
Alternatively, a means for outputting (xn, yn) and displaying it on a display is a necessary means for carrying out the present invention. still,
The same reference numerals indicate the same objects throughout the figures.

FIG. 3(a) shows the case of estimating the position of the current source in the heart. A magnetic field generated from the subject 16 is detected by a pickup coil 13 and guided to a superconducting quantum interference device (SQUID) 14, which is a highly sensitive magnetic flux sensor.

10 is the superconducting quantum interference device (SQUI)
D) shows a control circuit for operating 14 and converting it into a voltage proportional to the magnetic field strength. 12 is a pickup coil 13 and a superconducting quantum interference device (SQUI)
D) It is a low temperature heat insulating container (Dewar) for keeping 14 at a low temperature. 15 is a superconducting quantum interference device (SQUID)
There are 14 holders.

A voltage proportional to the magnetic field strength entering each pickup coil 13 is applied to a neural network.
2, the position of the current dipole and the signals proportional to the current intensity Vx, Vy, Vz and VQx, VQy, V
It is converted to Qz.

The learning method for the neural network 2 is based on the aforementioned Biot-Savart law, a superconducting quantum interference device (SQUID), a magnetometer (as shown in the figure, a pickup coil 13, a superconducting Quantum interference device (SQUID) 14, and control circuit 10
1 spatial parameters such as the diameter S of the pickup coil 13, the differential order, the position of the coil arrangement, and the superconducting quantum interference device (SQUID).
Learning is performed by calculating the theoretical magnetic field strength B from the sensitivity of the magnetometer 1 using a computer, and using this as a teaching signal to calculate the weighting coefficient (Wji) of each neuron element of the neural network 2 configured on the computer. is possible.

[0022] Normally, in order to detect a weak magnetic field generated from the living body, the pickup coil 13 is, for example,
Two magnets are placed close to each other in opposite directions to cancel earth's magnetic field unrelated to the biomagnetic field. That is, if the magnetic fields detected by the respective pickup coils 13 are B0 and B1 (however, the magnetic field detected by the pickup coil 13 placed closer to the living body is B0), then the magnetic field B of the magnetic field generated from the living body is is determined by B=B0-B1=dB/dr. At this time, the differential order becomes first order.

[0023] However, B0 and B1 detected by each pickup coil are as follows: B0, B1=∫μQ× (r-
r') /4π|rr'|3ds.

When the sensitivity constant of the superconducting quantum interference device (SQUID) magnetometer 1 is k, the electric signal VBi output by the magnetic field generated in the living body is determined by VBi=kB.

Therefore, the position r of the current dipole in the living body
', and by assuming the position r of the pickup coil 13, the magnetic field that can be detected by the pickup coil 13 can be calculated, and the superconducting quantum interference device (SQUI)
D) Since the output voltage VBi of the magnetometer 1 can be determined on a computer, by using this as a teacher signal, the weight (Wji) of each neuron element 8 of the neural network 2 built on the computer can be determined. can be calculated.

In order to display the strength and positional movement of the current dipole obtained as described above on the display 4, the positional information Vx, Vy, Vz is input to the scan control circuit 9 of the display 4, and the current dipole is Display dipole strength information VQx, VQy, VQz 4
A brightness control circuit 10 generates a signal with an intensity proportional to the current dipole intensity to adjust the brightness on the display 4.

By doing so, it becomes possible to display temporal changes in the current dipole on the display 4 in real time. Also, MRI like 11,
Alternatively, by providing a device that reads data from X-ray CT and displays it overlay on the display 4, it becomes possible to simultaneously display anatomical information and the propagation path of the current path in the nervous system.

Although FIG. 3(a) shows a case in which the magnetic field from the chest wall of the heart is measured, similar position estimation is also possible in the brain magnetic field. The above-mentioned first-order or second-order differential pickup coils 13 are arranged in a grid pattern, but in the case of a brain magnetic field, the pickup coils 13 are arranged so as to surround the head.

Next, as shown in FIG. 4(b), the human body 16
For a phantom 19 having the same boundary conditions as,
Filled with an aqueous solution 20 having a resistivity equal to the resistivity inside the human body, the magnetic field generated from the pseudo current dipole 17 by the current source 21 is measured by a superconducting quantum interference device (SQUID) magnetometer 1, and its output and the pseudo current dipole are measured.
17 directly from the neural network 2
By performing learning to determine the weighting coefficient Wji, it becomes possible to estimate the position under conditions closer to those of the human body. Here, 23 is a device for driving the pseudo current dipole 17, and at the same time transmits position information to a computer 24 for learning.

Furthermore, by forming the phantom 19 with a structure or material that can change its shape, it becomes possible for the neural network 2 to learn more in accordance with each individual subject.

In FIG. 3(a), the dipole position (x, y, z) is output from neural network 2.
The current strength (Qx, Qy, Qz) is used as output to visualize the movement of the current dipole.
1), (b1), by learning to find the current density Q (xn, yn, zn) at a certain grid point, we visualize changes in the three-dimensional current dipole density. It is also possible.

Specifically, as shown in FIG. 5(b1), a three-dimensional pseudo current dipole {black dots at the intersections
For convenience, the relative positional relationship (r',
r), find the magnetic field Bi of each pickup coil (i) 13, Bi = ΣBj, and use this as a teacher signal to apply the neural network.
The weighting coefficient of neuron element 8 configuring 2 (
Wji) is calculated.

Similarly, as shown in FIGS. 6(a2) and (b2), by learning only the lattice points on a particular plane, the state of current activity on a certain plane can be selectively visualized. It is also possible to make any plane observable by making each weighting constant variable.

[0034] In this method, the number of output points can be reduced by one dimension by making it flat, so it is possible to visualize higher density with fewer neurocircuits, and it is also easier to compare with tomographic images such as MRI. There are some advantages.

In addition, in FIG. 3(a), analog voltages (VB1, VB
2, ~), but as shown in Figure 7, a pulse output superconducting quantum interference device (SQUID) magnetometer 1a
using the counter 25 to convert the value into a digital value, and directly transmit each bit to the neural network 2
It may also be input to the neuron element 8a of a. In this case, the response function of the first stage neuron element 8a is set so as to give a digital response of 0 or 1. Also, one superconducting quantum interference device (SQUID) 14
Correspondingly, a multi-bit neuron element 8 output by the counter 25 is required. Also, the teacher signal at this time is as shown in the figure, for example, in FIGS. 5 and 6.
The current density Q1 (x1, y1, z1) to Qn (xn, yn, zn) at the three-dimensional or two-dimensional discrete points shown in
), or the positions (x, y, z) of the current source shown in Fig. 3(a) in three mutually perpendicular directions from a certain reference point, and the three-directional components of the current density ( Qx,
Analog voltages (Vx, Vy, V
z, VQx, VQy, VQz).

As described above, the present invention uses a superconducting quantum interference device (SQUID) 14, which is a magnetic sensor, to measure the distribution of a weak magnetic field generated from a living body, and from the measurement data, it is possible to determine the activity inside the living body. In a multi-channel biomagnetic imaging device that estimates the position of current and images its distribution, the position of a biocurrent source is estimated using a neural network 2. At this time, each pickup coil is used as an input for neural network 2.
13 A superconducting quantum interference device (SQUID) that is proportional to the magnetic field strength received by the magnetic flux meter 1. Using the analog voltage or digital signal of 1, the neural network 2 outputs signals in three mutually perpendicular directions from a reference point with a current source. position (x, y, z), and the three-directional components of current intensity (Q
analog voltage (Vx, Vy
, Vz, VQx, VQy, VQz), or three-dimensional, 2
Current density Q(xn, yn, zn,
Or xn, yn) is output and displayed on the display.

[0037]

[Effects of the Invention] As explained above, the biomagnetic imaging device of the present invention uses a superconducting quantum interference device (SQUID), which is a magnetic sensor, to measure the distribution of a weak magnetic field generated from a living body. In a multi-channel biomagnetic imaging device that estimates the position of active current inside a living body from measurement data and images its distribution, the position of a biocurrent source is estimated using a neural network 2. At this time, as an input to neural network 2,
SQ proportional to the magnetic field strength received by each pickup coil
Using the analog voltage or digital signal of the UID flux meter, the output of the neural network 2 is an analog voltage proportional to the position in three mutually perpendicular directions from a certain reference point of the current source and the three-directional components of the current intensity. Or
Since the current density at three-dimensional and two-dimensional discrete points is output and displayed on the display, it is possible to estimate the position of the current activity source in real time using the magnetic field generated from the living body. , it becomes possible to display the movement of the current source in real time, which is useful for estimating the location of the affected area of diseases that suddenly occur such as arrhythmia and epilepsy, and will greatly contribute to improving the performance of such biomagnetic imaging devices. It has a good effect.

[Brief explanation of drawings]

[Figure 1] Diagram explaining the principle of the present invention (Part 1)

[Figure 2] Diagram explaining the principle of the present invention (Part 2)

[Fig. 3] Diagram showing one embodiment of the present invention (Part 1)

[Fig. 4] Diagram showing one embodiment of the present invention (Part 2)

[Fig. 5] Diagram showing another embodiment of the present invention (Part 1)

FIG. 6 is a diagram showing another embodiment of the present invention (part 2).
)

[Fig. 7] Diagram showing another embodiment of the present invention (Part 3)

[
Figure 8: Diagram explaining a conventional biomagnetic imaging device

[Explanation of symbols]

1 Superconducting quantum interference device (SQUID) Magnetometer 2 Neural network 3 Signal processing circuit
4 Display 5 Weighting coefficient setting element
6 Adder 7 Nonlinear characteristic circuit 8, 8a Neuron element circuit
9 Scanning control circuit 10 Brightness control circuit
11 Tomographic image data reading device 12 Dewar
13 Pickup coil 14 Superconducting quantum interference device (SQUID) 16
Living body or human body 17 Pseudo current dipole
19 Human body phantom

Claims (6)

[Claims] Claim 1: A superconducting quantum interference device (S
QUID) (14) was used to measure the distribution of the weak magnetic field generated from the living body (16), and from the measurement data,
This is a multi-channel biomagnetic imaging device that estimates the position of active current inside a living body and images its distribution, and includes signals (VB1, VB1,
VB2, ~) is input to the neural network (2), the teacher signal is calculated and given on a computer, and the position of the biological current source is estimated using the neural network (2). Biomagnetic imaging device. 2. The biomagnetic imaging device, comprising:
As an input to the neural network (2), a superconducting quantum interference device (SQUID) magnetometer (which outputs an output in proportion to the magnetic field strength received by each pickup coil (13)) is used.
1) Using the analog voltages (VB1, VB2, ~), as the output of the neural network (2),
Positions in three mutually perpendicular directions from a reference point with a current source (
x, y, z) and the three-directional components of current intensity (Qx,
Analog voltage (Vx, Vy, V
z, VQx, VQy, VQz). The biomagnetic imaging device according to claim 1. 3. The biomagnetic imaging device, comprising:
As a learning method for the neural network (2),
A phantom (
19), there is a pseudo current dipole (17) that serves as a current source.
The biomagnetic imaging according to claim 1, characterized in that the weighting coefficient (Wji) of each neuron element (8) of the neural network (2) is determined based on the magnetic field generated therefrom. Device. 4. The biomagnetic imaging device, further comprising:
The biomagnetic imaging device according to claim 1 or 3, characterized in that the shape of the phantom (19) is variable. 5. The biomagnetic imaging device, further comprising:
The input of the above neural network (2) is
16) The biomagnetism according to claim 1, characterized in that the pulse output from the superconducting quantum interference device (SQUID) magnetometer (1a) which outputs a number of pulses proportional to the magnetic field strength is converted into a digital value. Imaging equipment. 6. The biomagnetic imaging device, comprising:
As an input to the neural network (2), a superconducting quantum interference device (SQUID) magnetometer (1) proportional to the magnetic field strength received by each of the pickup coils (13).
Using the analog voltages (VB1, VB2, ~), the current density Q (xn, y
2. The biomagnetic imaging apparatus according to claim 1, wherein the biomagnetic imaging device sets a value of (n, zn, or xn, yn).