WO2015142029A1

WO2015142029A1 - 3-dimensional cardiac outline reconstruction method

Info

Publication number: WO2015142029A1
Application number: PCT/KR2015/002569
Authority: WO
Inventors: 김기웅; 하태훈; 권혁찬; 이용호; 김진목
Original assignee: 한국표준과학연구원
Priority date: 2014-03-21
Filing date: 2015-03-17
Publication date: 2015-09-24
Also published as: KR101525485B1; CN106132288A; CN106132288B

Abstract

The present invention provides a cardiac outline reconstruction method and a magneto cardiogram measuring system. The cardiac outline reconstruction method comprises the steps of: obtaining a first cardiac outline from power of source current vector (source current power) obtained using an array-gain constraint minimum-norm with recursively updated gram matrix; AGMN-RUG) spatial filter; obtaining a second cardiac outline by applying a coherence mapping method; and configuring a third cardiac outline by coupling the first cardiac outline and the second cardiac outline with each other.

Description

3D heart contour reconstruction method

The present invention relates to a method for reconstructing a cardiac contour, and to a method for forming a cardiac contour in three dimensions by using multichannel data measured by a cardiac conduction device without the help of an anatomical imaging device such as an MRI device.

Magnetocardiography (MCG) measures heart disease by measuring the magnetic signals generated by the electrical activity of the myocardium through a superconducting quantum interference device (SQUID) sensor, an ultra-sensitive magnetic sensor. It is a device to diagnose non-invasive. Multi-channel data measured by a core map device can be used to localize lesions of heart disease. In particular, localizing the lesion of the heart disease in the three-dimensional cardiac model allows the doctor to intuitively show the information of the lesion.

However, in order to obtain the 3D heart model, anatomical image data measured by a patient's X-ray computed tomography (CT) device or magnetic resonance imaging (MRI) device is required. In order to obtain the anatomical image data, the patient may be exposed to unnecessary radiation exposure or a strong magnetic field. In addition, in order to obtain a heart model, a process of segmenting the heart with anatomical image data is required. The cardiac separation process may be time consuming, although the expert may work manually or use an automated cardiac separation algorithm.

U.S. Patent Publication No. 2008-0033312 by Kenji Nakai et al. Applies a Tikhonov regularization method to a minimum norm estimation method, one of non-adaptive spatial filtering methods, to provide current density. After estimating, we reported a method of generating three-dimensional heart contours based on this.

However, the non-adaptive spatial filtering method has a limitation in accurately generating a three-dimensional cardiac model because the estimation of a deep signal source is not accurate.

One technical problem to be solved of the present invention is to provide a three-dimensional cardiac visualization or mapping method.

The cardiac contour reconstruction method according to an embodiment of the present invention is a method for reconstructing a source current vector obtained using an array-gain constraint minimum-norm with recursively updated gram matrix (AGN-RUG) spatial filter. Obtaining a first contour of the heart from source current power; Obtaining a second contour of the heart by applying a coherence mapping method; And combining the first contour and the second contour to form a third contour of the heart.

In one embodiment of the present invention, obtaining a first contour of the heart comprises: measuring a magnetic signal generated in the myocardium using a cardiac diagram device; Extracting a cardiac magnetic signal of a P-wave and a T-wave interval representing atrial and ventricle from a multi-channel measured cardiac signal; Calculating a magnetic field (leadfield vector) using a horizontally layered volume conductor model approximating the unit source current of the heart and the position information of the core magnetic sensor; And normalizing the leadfield vector to compensate for the size of the signal source over distance.

In one embodiment of the present invention, obtaining the first contour of the heart comprises: initializing an arbitrary weight matrix to an identity matrix; Obtaining a gram matrix by combining the leadfield vector with an arbitrary weight matrix; Setting a normalization parameter in the obtained gram matrix in consideration of the signal-to-noise ratio of the measurement signal; Obtaining an array-gain constraint minimum-norm spatial filter by combining the normalized gram matrix and the normalized leadfield vector; Obtaining power of an estimated source current vector using the obtained array gain limit spatial filter and the measured atrial and ventricular signals; Checking whether the power of the obtained source current vector converges to a constant value; Updating the weighted matrix using the estimated source vector when the obtained power of the source current vector does not converge to a constant value; Normalizing the power of the obtained estimated source current vector; And reconstructing the first contour of the atrium and the ventricle by applying the first threshold value set as the signal-to-noise ratio to the power of the normalized source current vector.

In one embodiment of the present invention, the step of obtaining a second contour of the heart by applying a coherence mapping method includes: finding points at which the power of the source current vector is greatest in the atria and the ventricles, respectively; Reconstructing signal waveforms (estimated source current vectors) in all source regions by linearly combining the measured core signal and the minimum-size spatial filter of the regression update gram matrix array gain; Fast Fourier transforming the reconstructed signal waveform (estimated source current vector); Obtaining cross-spectrum and auto-spectrum for all source regions in the atria and ventricles using the fast Fourier transformed signal; Combining the obtained inter-spectrum with the self-spectrum to obtain coherence of the atrium and ventricular coherence; Normalizing the coherence of the obtained atrium and ventricles; And reconstructing the second contour of the atrium and the ventricles by applying a second threshold set to the normalized coherence signal-to-noise ratio.

The cardiac contour reconstruction method according to an embodiment of the present invention is a method for reconstructing a source current vector obtained using an array-gain constraint minimum-norm with recursively updated gram matrix (AGN-RUG) spatial filter. The heart is outlined from the source current power.

In one embodiment of the present invention, deriving the outline of the heart may include measuring a magnetic signal generated in the myocardium using a cardiac diagram device; Extracting a cardiac magnetic signal of a P-wave and a T-wave interval representing atrial and ventricle from a multi-channel measured cardiac signal; Calculating a magnetic field (leadfield vector) using a horizontally layered volume conductor model approximating the unit source current of the heart and the position information of the core magnetic sensor; And normalizing the leadfield vector to compensate for the size of the signal source over distance.

In one embodiment of the present invention, the method comprises: initializing an arbitrary weight matrix to a unitary matrix; Obtaining a gram matrix by combining the leadfield vector with an arbitrary weight matrix; Adding a normalization parameter to the obtained gram matrix in consideration of the signal-to-noise ratio of the measurement signal; Obtaining an array-gain constraint minimum-norm spatial filter by combining the normalized gram matrix and the normalized leadfield vector; Obtaining power of an estimated source current vector using the obtained array gain limit spatial filter and the measured atrial and ventricular signals; Checking whether the power of the obtained source current vector converges to a constant value; Updating the weighted matrix using the estimated source vector when the obtained power of the source current vector does not converge to a constant value; Normalizing the power of the obtained estimated source current vector; The method may further include reconstructing the atrium and the ventricles by applying the threshold of the normalized source current vector to a signal-to-noise ratio.

Cardiac contour reconstruction method according to an embodiment of the present invention comprises the steps of obtaining the point where the power of the source current vector is the largest in the atrium and ventricles, respectively; Reconstructing the signal waveforms (estimated source current vectors) in all source regions by linearly combining the measured core signal and the minimum size spatial filter of the regression update gram matrix array gain; Fast Fourier transforming the reconstructed signal waveform (estimated source current vector); Obtaining cross-spectrum and auto-spectrum for all source regions in the atria and ventricles using the fast Fourier transformed signal; Combining the obtained inter-spectrum with the self-spectrum to obtain coherence of the atrium and ventricular coherence; Normalizing the coherence of the obtained atrium and ventricles; And reconstructing the atrium and ventricles by applying a threshold set to the normalized coherence signal-to-noise ratio.

The recording medium which records the program according to an embodiment of the present invention can execute the above methods on a computer.

A core measurement system according to an embodiment of the present invention includes a core measurement sensor including a core sensor and a magnetic shield room and measuring a core signal; And a processor configured to reconstruct the contour of the heart by processing the core diagram signal. The processing unit uses the heart current from the power of the source current vector obtained using an array-gain constraint minimum-norm with recursively updated gram matrix (AGN-RUG) spatial filter. 1 Find the outline. The processor applies a coherence mapping method to obtain a second contour of the heart. The processing unit combines the first contour and the second contour to form a third contour of the heart.

In the cardiac contour three-dimensional reconstruction method according to an embodiment of the present invention, it is possible to reconstruct the heart model using only the MCG data of the patient without the use of CT or MRI apparatus required to obtain an anatomical image. Thus, patients can avoid the risk of exposure to unnecessary radiation or strong magnetic fields. In addition, if the imaging apparatus is not used, the economic burden on the patient can be reduced. In addition, since cardiac separation is not necessary to make a cardiac model, it can be used to diagnose a patient in real time using an MCG device.

1 is a flowchart illustrating a 3D heart contour reconstruction method according to an embodiment of the present invention.

2, 3, and 4 are flowcharts illustrating a three-dimensional heart contour reconstruction method according to an embodiment of the present invention.

5 is a view illustrating a core measurement apparatus according to an embodiment of the present invention.

FIG. 6 is a diagram illustrating cardiac waveform extraction from a multi-channel measured core map signal according to an embodiment of the present invention.

7 is a diagram illustrating a relationship between a source space and a sensor space required for calculating a lead-field vector of a core diagram according to an embodiment of the present invention.

8 illustrates coherence mapping of the atria in accordance with one embodiment of the present invention.

9 illustrates coherence mapping of the ventricles according to an embodiment of the present invention.

10 is a diagram illustrating a virtual source space and a virtual heart for numerical simulation according to an embodiment of the present invention.

11 is a diagram showing the results of numerical simulations according to an embodiment of the present invention.

12 is a diagram illustrating a cardiac phantom for a model experiment according to an embodiment of the present invention.

FIG. 13 is a diagram illustrating a numerical simulation result using the model heart of FIG. 12.

14 shows the ratio of normalization parameters to signal to noise ratios.

3-D cardiac visualization or mapping methods are useful for magnetocardiography (MCG) clinical applications. However, cardiac reconstruction requires additional image modalities. We propose a three-dimensional cardiac contour reconstruction method using only MCG measurement data without additional imaging techniques. Cardiac contours can be reconstructed by a combination of spatial filtering and coherence mapping methods.

The strength of cardiac activities can be represented by an array-gain constraint minimum-norm with recursively updated gram matrix (AGN-RUG) spatial filter.

In addition, the coherence between the maximum source point of the atrium and all the source points and the coherence between the maximum source point of the ventricles and all the source points were compared by coherence mapping.

Numerical simulations and phantom experiments demonstrated the effectiveness of three-dimensional cardiac contour reconstruction by comparing the original shape with the reconstructed composition.

Cardiac contour three-dimensional reconstruction method according to an embodiment of the present invention does not use an anatomical image obtained from a CT or MRI device, and generates a three-dimensional heart contour using only the cardiac diagram (MCG) data measured from the patient It is a way. Specifically, the cardiac first contour of the heart is obtained from the source current power obtained using an array-gain constraint minimum-norm with recursively updated gram matrix (AGN-RUG) spatial filter. Is saved. In addition, a second contour of the heart is obtained by applying a coherence mapping method. Finally, the first contour and the second contour are combined to form a third contour of the heart.

The magnetic core diagram (MCG) is a device that can non-invasively measure the magnetic field generated in the myocardium. Because of its excellent time and spatial resolution, cardiac diagrams are very useful in estimating reentrant excitation regions that cause myocardial ischemic regions or cardiac arrhythmia. . Recently, a study has been reported to visualize the recurrent excitatory region of the myocardium on the 3D cardiac surface using cardiac diagram. By visualizing cardiac electric activities on a three-dimensional cardiac model, many clinical applications are possible.

In general, however, there are some drawbacks to obtaining a three-dimensional heart model. First, complex segmentation processes are required from anatomical imaging devices such as X-ray computed tomography (CT) or magnetic resonance imaging (MRI). Recently, many research groups have proposed an automatic segmentation method that overcomes the disadvantages of the manual segmentation process. However, anatomical image acquisition is still essential. Second, when obtaining anatomical imaging information, the patient is exposed to X-rays or high magnetic fields, which can pose a potential risk. Finally, in order to obtain anatomical image information, the patient has to pay a high cost.

In order to overcome the shortcomings of the process of acquiring the 3D heart model, a method of generating a 3D heart model using only core measurement data has been proposed. The Nakai group of Japan estimates the current density by applying the Tikhonov regularization method to the minimum norm estimation method, which is one of non-adaptive spatial filtering methods. We have reported a method for generating three-dimensional cardiac contours. However, the non-adaptive spatial filtering method is not accurate in estimating the depth of the signal source. Therefore, the non-adaptive spatial filtering method has a limitation in generating a three-dimensional heart model accurately.

Compared to non-adaptive spatial filtering, adaptive spatial filtering shows high spatial resolution. Adaptive spatial filtering is resistant to noise interferences because it uses measurement geometry and covariance matrix. Spatial filtering is strong for weakly correlated signals, but exhibits large localization errors for strongly correlated signals. The electrical signals of the heart are strongly correlated. Therefore, localizing myocardial current sources with adaptive spatial filtering results in large errors. Therefore, both non-adaptive spatial and adaptive spatial filtering have limitations in reconstructing 3D cardiac models.

Recently, a minimum-size spatial filter with limited regression-gram matrix array gain with spatial resolution has been proposed. The spatial filter may derive a result corresponding to the adaptive spatial filter based on the non-adaptive spatial filter. Therefore, the spatial filter can be used to generate a cardiac model more accurately by avoiding problems caused by strongly correlated myocardial signals.

We also applied a coherence mapping method to generate the heart model more accurately. Electrical signals generated from the atria and ventricles show correlated waveforms in each region. Thus, by comparing the similarities of regenerated heart waveforms, a more accurate model of the atria and ventricles can be reconstructed.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the drawings, like reference numerals refer to like elements, and the size of each element in the drawings may be exaggerated for clarity and convenience of description.

1, 2, 3, and 4, the three-dimensional cardiac contour reconstruction method includes an array-gain constraint minimum-norm with recursively updated gram matrix (AGNN-RUG). Obtaining a first contour of the heart from a source current power obtained by using a spatial filter (S100); Obtaining a second contour of the heart by applying a coherence mapping method (S200); And combining the first contour and the second contour to form a third contour of the heart (S300).

Obtaining a first contour of the heart (S100) is processed as follows. The core diagram signal generated in the myocardium is measured using the core diagram apparatus (S110). The cardiac magnetic signals b _a (t) and b _v (t) of the P and T wave sections representing the atria and the ventricles are extracted from the cardiac signal measured by the multi-channel (S120). A magnetic field (leadfield vector L (r)) is calculated using the unitary source current of the heart and the horizontally layered volume conductor model approximating the heart and the position information of the core magnetic sensor (S130). In order to compensate for the size of the signal source according to the distance, the leadfield vector is normalized (S140).

A method for obtaining an array-gain constraint minimum-norm with recursively updated gram matrix (AGN-RUG) spatial filter is introduced. The arbitrary weight matrix H (r) is initialized to the unitary matrix I (S150). A gram matrix G is obtained by combining the lead field vector L (r) and the arbitrary weight matrix H (r) (S160). A regularization parameter (λ) is set in the obtained gram matrix G in consideration of the signal-to-noise ratio of the measurement signal (S170).

An array-gain constraint minimum-norm spatial filter is obtained by combining the normalized gram matrix G and the normalized leadfield vector (S180). The estimated source current vector and the estimated source current power are obtained using the obtained array gain limit size filter and the measured atrium and ventricular signals (S191). It is checked whether the obtained power of the estimated source current converges to a constant value (S192). If the obtained power of the estimated source current does not converge to a constant value, the weight matrix is updated using the estimated source current vector (S193). When the obtained power of the estimated source current converges to a constant value, the power of the obtained estimated source current vector is normalized (S194). The first contour of the atrium and the ventricles is reconstructed (195) by applying the normalized estimated source current power with a 1-1 threshold and a 1-2 threshold set to a signal-to-noise ratio.

The step of obtaining the second contour of the heart by applying the coherence mapping method is as follows.

By the method described above or another method, the estimated source current vector of the heart or the estimated source current of the heart is calculated. The points r _a and r _v having the largest power of the source current in the atrium and the ventricles are obtained, respectively (S210). A linear combination of the measured cardiac magnetic signal and the regression-update-gram matrix array-restricted minimum-size spatial filter provides a source current signal waveform (

) Is reconstructed (S220). According to a modified embodiment of the present invention, the source current signal waveform may use the value calculated in S191.

The reconstructed source current signal waveform is fast Fourier transformed (S230). Cross-spectrum (Γ _aq (f), Γ _vq (f)) and auto-spectrum (Γ _aa ) for all source regions in the atrium and ventricles using the fast Fourier transformed signal f), Γ _vv (f)) are obtained (S240). The atrial cross-spectrum is cross-correlation between the point where the source current has the greatest power in the atrium and another point. The ventricular cross-spectrum is a cross-correlation between the point where the source current power is largest in the ventricle and another point.

The atrial self-spectrum is auto-correlation of the point where the source current has the greatest power in the atrium. The ventricular self-spectrum is auto-correlation of the point where the power of the source current in the ventricle is greatest.

Atrial coherence and ventricular coherence (Coh _aq (f), Coh _vq (f)) are obtained by combining the obtained cross-spectrum and self-spectrum (S250). The coherence can be standardized. The second contour of the atrium and the ventricles is reconstructed by applying the normalized coherence to the 2-1 threshold and the 2-2 threshold set to the signal-to-noise ratio (S260).

Subsequently, the first contour and the second contour are combined with each other to form a third contour of the heart (S300).

Referring to FIG. 5, the core measurement apparatus 100 is installed in a magnetically shielded room 101. The core device 100 includes a 64 channel superconducting quantum interference device (SQUID). The pickup coil is a first order axial gradiometer with a baseline of 70 mm. The noise level of each SQUID sensor 112 is 5 fT _rms / Hz ^1/2 at 100 Hz and the sampling rate is 500 Hz. The SQUID sensor 112 measures a fine magnetic signal generated in the human body. The SQUID sensor 112 may operate in a cryogenic state below minus 250 degrees. Therefore, the SQUID sensor is installed inside the cooling means 114 to maintain the cryogenic temperature. The cooling means 114 may be disposed inside the dewar 116.

The driving circuit 124 drives the core drawing device. The amplifier and filter 118 is disposed inside the magnetic autism chamber. The power supply unit may supply power to the amplifier and the filter 118 and the like. The measured core figure signal is transmitted to the processing unit 122 and signal processed and analyzed (S110).

Referring to FIG. 6, in the measured cardiac signal, the P, Q, R, S, and T peaks appear sequentially. The section including the P peak is separated into P waves, the section including the Q, R, and S peaks is separated into QRS waves, and the section containing the T peak is separated into T waves. The P wave (b _a (t)) of the heart waveforms is a signal generated in the atrium, the QRS wave is a signal generated when the ventricles contract, and the T wave (b _v (t)) occurs when the ventricles relax Is a signal.

In the present invention, the start and end portions of the P wave are extracted to generate the atrial model, and the start and end portions of the T wave are extracted to generate the ventricular model (S120). The extracted P wave _(a b (t)) and T wave (b _v (t)) is given by:

[Equation 1]

Where n is the number of core sensors and t is the extraction interval of the heart waveform. The superscript T is the transpose of the matrix.

Referring to FIG. 7, the lead field vector is a magnetic field calculated in the sensor space when the unit current source and the conductor model are known. The source current is an equivalent current dipole (ECD) having a magnitude and a direction. The human body can be regarded as a conductor space z <0 having a constant electrical conductivity. The conductor model is a horizontally layered conductor model. The planar body model assumes that the plane below (z <0) is a conductor and the plane above (z> 0) is an insulator. The source current is in the conductor space. In addition, a source space V is selected from the conductor spaces. The heart is in the source space (V).

The source space may be a cuboid based on an xyz rectangular coordinate system. The source space may be composed of voxels divided by 10 mm units. The source space may be 150 mm in the x-axis direction, 170 mm in the y-axis direction, and 150 mm in the z-axis direction. Accordingly, the total number of voxels Q may be 3314.

Outside the conductor space, a measurement region z> 0 is arranged. A sensor space in which the plurality of core sensor 112 may be disposed in the measurement area. The core sensor 112 may be a SQUID sensor. Location information of the core sensor is already known. The SQUID sensor 112 may include a first order axial gradiometer having a baseline of 70 mm.

Since the output power of the spatial filter can estimate the magnitude of mycocardial electric activity, the spatial filter method can reconstruct the heart model. The source vector is defined as s (r, t) in discrete volume space V at position r = (x, y, z) and time t. The total number of sources is Q. The source vector may indicate the magnitude and direction of the source current.

The lead field vector l (r) calculated in one direction from one current source may be represented by Equation 2, and the total lead field vector L (r) for the entire current source Q is represented by Equation 3 It may be (S130).

[Equation 2]

[Equation 3]

If the time vector at position r = (x, y, z) is t, and you define the source vector as s (r, t), the source vectors of the atrium and ventricles are s _a (r, t) and s _v (r , t). When the lead field vector L (r) and the source vector s _a (r, t) and s _v (r, t) are linearly combined, the magnetic field of the atrium and the ventricle calculated in the sensor space is b _a (t) , b _v (t)) can be calculated as in Equation 4.

[Equation 4]

Where V represents the source space.

Spatial filters are a technology used in sensor arrays for signal transmission and reception. The spatial filter (W ^T _a, ^T W _v) is the weight vector (weight vector). Accordingly, the signal of the source current can be estimated by linearly multiplying the measured magnetic field (core signal) by the spatial filters W ^T _a and W ^T _v . That is, the estimated source current vector of the atrium and the ventricles may be represented by Equation 5.

[Equation 5]

Since the estimated source current vectors s _a (r, t) and s _v (r, t) of the atria and the ventricles are included in the total estimated source current vectors s (r, t), the power of the estimated source current vectors is increased. In general, the equations derived to obtain are described without atrial subscript (a) and ventricular subscript (v). Combining Equations 4 and 5, the estimated source current vector may be expressed as Equation 6.

[Equation 6]

Here, r is a target source position and r 'is a position other than the target source.

Is a beam response and represents the gain of misleading leakage current generated from a source other than the target source position of the spatial filter.

In order to design an ideal spatial filter, the gain of the beam response should be minimized. Therefore, a minimum-norm-based spatial filter can be derived by solving the optimization problem under the following constraint.

[Equation 7]

Represents the gain of the sensor array. In Equation 7, C (W) is a cost function and represents the sum of the total leakage currents. The cost function C (W) may be expressed as Equation 8.

[Equation 8]

tr {} is the trace of the diagonal components of the matrix, δ (r) is the delta function and H (r ') is an arbitrary weight matrix.

The gram matrix of the atrium and the ventricles (Gram matrix, G _a , G _v ) may be defined as shown in Equation 9 by combining the lead field vector with arbitrary weight matrices of the atria and ventricles.

[Equation 9]

For source currents having the same magnitude, the lead field vector has a large magnitude when the equivalent current dipole is close in the SQUID sensor. On the other hand, the lead field vector is small when it is far from the sensor. Therefore, normalizing the lead field vector for each position can relatively increase the size of the distant source. Normalized leadfield vector (

) May be represented as in Equation 10 (S140).

[Equation 10]

The normalized lead field vector (

) And the gram matrix G (r) are linearly combined to obtain a spatial filter or weight vector as follows.

[Equation 11]

Minimal variance (MV) spatial filters are the most popular adaptive spatial filters in the field of bioelectromagnetism. The covariance matrix is an ensemble average of the measured data b (t) as

Can be defined as

The minimally distributed spatial filter can be derived by solving an optimization problem under the constraint of the following equation.

[Equation 12]

The minimum distributed spatial filter is the normalized lead field vector (

) And the covariance matrix (D) can be obtained as follows.

[Equation 13]

The only differences between Equations 11 and 13 are the gram matrix and the covariance matrix. Applying Equation 4 to the covariance matrix (D) can be expressed as follows.

[Equation 14]

Comparing Equations 9 and 14, the arbitrary weight matrix H (r) is the source power matrix.

Similar to Therefore, if the weight matrix H (r) is replaced with the source power matrix, it can be expected that the performance of the minimum size spatial filter is similar to the minimum distributed spatial filter.

To get more accurate results, we need to consider the effect of noise on the gram matrix. The condition number of a function is a number that indicates how the output value changes due to a small change in the input variable. The actual number of conditions in the gram matrix is very large. Specifically, the noise component in the gram matrix is a very small value, but when converted to the gram inverse, the small noise value changes significantly. Therefore, calculating the inverse of the gram matrix results in inaccurate results. Applying a regularization method can reduce the gram inverse calculation error by removing noise components. Therefore, the modified minimum spatial gain spatial filter of the atria and the ventricles may be expressed as shown in Equation 15, respectively.

[Equation 15]

Where λ _a and λ _v are the normalization parameters of the atria and ventricles, and the normalization parameters are determined by the signal-to-noise ratio of the measured values.

Since the source current vectors s _a (r, t) and s _v (r, t) of the atria and the ventricles are unknown in Equation 14, the source currents are obtained to obtain the weighting matrix H _a (r) and H _v (r). Vector is estimated source current vector

and

It can be replaced with (S193).

The following process is required to find the optimal weight matrix H (r). First, the weight matrix is initialized to the unit matrix (S150). Second, the output value of the minimum gain spatial filter of array gain is obtained from Equation 15 (S180). Third, the source current vector estimated from Equation 5

,

To obtain (S191). Fourth, the weight matrix H (r) of Equation 9 is updated with the estimated source current vector. The weight matrix H (r) is recursively updated through the first to fourth processes until the updated weight converges to a constant value (S193). Finally, when the update is completed, an estimated source current vector may be obtained from Equation 5.

The estimated source current vector (

, Power of each source,

Wow

Can be obtained. The estimated source current power obtained may be normalized by dividing by the maximum magnitude (S194).

The first contour of the atrium and the ventricles are reconstructed by applying a first threshold set from the normalized estimated source current power from the signal-to-noise ratio. For example, when the first threshold is set to 0.5, only voxels having a value greater than or equal to the first threshold may be displayed in the source space. The outermost voxels of the voxels above the first threshold may be connected to one another to provide a first contour. The first-first threshold may be set for the atrium, and the first-second threshold may be set for the ventricles.

The first contour may be formed with respect to the atrium and the ventricles, respectively. The first threshold values of the atrium and the ventricles may be different.

[Coherence Mapping]

When calculating the power of a myocardial current source, physiological noise, such as breathing or digestion, may have a false influence on calculating the power of the source current. Therefore, coherence mapping can be further applied to create a more accurate heart model. Coherence refers to a phenomenon in which two or more waves interfere with each other in phase. In the case of the heart, the ion permeability of the cardiomyocyte membrane varies according to the area. Thus, in the proximal region, myocardial currents show similar action potentials. However, in the distant realms, they show different action potentials. Therefore, by estimating the position of a signal source that matches the maximum signal source of the atrium and the ventricles, the contour can be reconstructed into the atrium and ventricular regions, respectively.

Figure 9 illustrates coherence mapping of the ventricles in accordance with an embodiment of the present invention.

8 and 9, the coherence analysis is a representative analysis method showing the correlation between signals according to frequency. The coherence function for two signals of stationary state is a cross spectral density function and a magnetic spectral density ( It can be calculated from the auto spectral density function.

The estimated signal waveform (estimated source current vector) in the source region is reconstructed from the product of the spatial filter and the measured magnetic field from equation (5). The time series of the waveform (estimated source current vector) reconstructed at the Q position is

Appears. Waveforms (estimated source vectors) at the point where the power of the atria and ventricle source currents are maximum

Wow

It may be represented as (S210). Reconstructed Time Series Waveform (Estimate Source Vector)

,

May be transformed into S _a (f), S _v (f), and S _q (f) by fast Fourier transforms (S230).

The cross-spectrum of the atrium and the ventricles and the auto-spectrum of the Q-th source may be defined as follows (S240).

[Equation 16]

Where <> means an average and the superscript * indicates a complex conjugate.

The coherence of the atrium and the ventricles may be represented by Equation 17 as a combination of the cross-spectrum and the self-spectrum (S250).

[Equation 17]

Specifically, the coherence of the atria (Coh _aq (f)) is the cross-spectrum (Γ _aq (f)) obtained between the atria and all source spaces, respectively, for each self-spectrum (Γ _aa (f), Γ _qq (f)). Can be normalized by dividing by the square root of the product of Similarly, cochlear cohesion (Coh _vq (f)) can be standardized as above. Ventricular coherence (Coh _vq (f)) is the product of the cross-spectrum (Γ _vq (f)) obtained between the ventricle and all of the source spaces of each self-spectrum (Γ _vv (f), Γ _qq (f)). It can be normalized by dividing by the square root of.

The normalized atrial coherence and ventricular coherence may reconstruct a second contour based on each second threshold determined from a signal-to-noise ratio (S260). For example, when the second threshold is set to 0.5, only voxels having a value greater than or equal to the second threshold may be displayed in the source space. The outermost voxels of the voxels above the second threshold may be connected to one another to provide a second contour. The 2-1 threshold may be set for the atrium, and the 2-2 threshold may be set for the ventricles.

The second contour may be formed with respect to the atrium and the ventricles, respectively. The second thresholds of the atrium and the ventricles may be different.

The third contour may be reconstructed by combining the first and second contours of the atrium and the ventricles (S300). The reconstruction method may reconstruct a third contour based on each third threshold value after multiplying the normalized coherence by the power of the normalized source current of the atrium and the ventricles. The third contour may be formed with respect to the atria and the ventricles, respectively. The third thresholds of the atrium and the ventricles may be different.

Hereinafter, the results of numerical simulation and phantom experiment and numerical simulation and model experiment for demonstrating the effectiveness of the present invention will be described.

Referring to FIG. 10,

virtual hearts

204 and 206 are located in the virtual source space 202. The virtual heart includes a virtual atrium 204 and a virtual ventricle 206.

The virtual source space 202 may be a cuboid based on an xyz rectangular coordinate system. An arbitrary position of the xiphoid in the rectangular coordinate system is defined as a fiducial point. The virtual source space 202 may be composed of voxels divided by 10 mm units. The virtual source space 202 may be 150 mm in the x-axis direction, 170 mm in the y-axis direction, and 150 mm in the z-axis direction. Accordingly, the total number of voxels Q may be 3314.

The virtual atria 204 has been assigned 199 voxels, and the virtual ventricle 206 has been assigned 248 voxels. Equivalent current dipoles (ECDs) are set in the

virtual heart

204 and 206 as a source model, and the signal is a sine wave. The equivalent current dipole of the atrium is 0.25 μAm and the equivalent current dipole of the ventricles is 1 μAm. Gaussian white noise with a signal-to-noise ratio of 20 dB was added.

The center frequency of the atrial sine wave is 10 Hz, and the center frequency of the ventricular sine wave is 8 Hz. In addition, the planar body model was set as the heart model. Different phases of the sine wave were assigned to position to simulate the propagation of the heart's action potentials. In order to show the systole and relaxation of the heart, the phase was sequentially increased by 30 degrees from the cardiac base to the cardiac apex, that is, the y-axis direction. In addition, the phase was sequentially increased by 26 degrees in the transverse plane of the heart, that is, the z-axis direction.

Referring to FIG. 11, the magnetic field generated from the set source model was measured in the sensor space (S110). The power of the atrial and ventricular source currents was obtained by applying the regression update gram matrix array minimum gain spatial filter (S191). The first contour of the atrium and the ventricles was reconstructed by filtering the power of the source current by the 1-1 threshold and the 1-2 threshold set as the signal-to-noise ratio (S195). From the reconstructed signal waveforms, the coherence between the atria and the ventricles of all source regions was obtained (S250). The coherence was filtered by the 2-1 threshold and the 2-2 threshold set to the signal-to-noise ratio to reconstruct the second contour of the atrium and the ventricles (S260). Finally, the first contour and the second contour are combined to reconstruct the third contour of the heart (S300). As a result, the contour of the reconstructed heart almost coincides with the virtual heart contour.

Referring to Figure 12, the model heart (a) was made in the form of goose eggs (goose egg). Twelve current dipoles of 10 mm in length are located on 5 mm of the surface of the model heart. The model heart is located in a torso phantom made of fiberglass. The model heart is tilted 45 degrees to the coronal plane and 30 degrees to the sagittal plane. The body model is filled with 0.9% saline solution having an electrical conductivity of about 0.16 S / m.

The model heart is located in the virtual source space. The virtual source space is the same as the virtual source space that was used for the numerical simulation. Any position of the xiphoid on the surface of the trunk model b is defined as a fiducial point. In the model heart, an equivalent current dipole (ECD) was set as the source model. The size of the model cardiac equivalent current dipole is 100 μAm and the signal shape is a combination of 5 Hz and 10 Hz sine wave. In addition, the planar body model was set as the heart model.

Referring to FIG. 13, the first contour and the second contour of the heart were obtained by the method of matching with the power of the source current as in the numerical simulation. The first contour and the second contour were combined to reconstruct the third contour of the heart. The results indicate that the contour of the reconstructed heart is nearly identical to the model heart contour.

14 shows the ratio of normalization parameters to signal to noise ratios.

Referring to FIG. 14, the ratio of the normalization parameter is a ratio with respect to the maximum singular value of the gram matrix. The higher the signal-to-noise ratio, the smaller the ratio of normalization parameters. In addition, the ratio of normalization parameters is small because the signal of the atria is smaller than that of the ventricles.

Although the present invention has been described with reference to the embodiments shown in the drawings, this is merely exemplary, and it will be understood by those skilled in the art that various modifications and equivalent other embodiments are possible. Therefore, the true technical protection scope of the present invention will be defined by the technical spirit of the appended claims.

Claims

The first contour of the heart is derived from the source current power obtained using an array-gain constraint minimum-norm with recursively updated gram matrix (AGN-RUG) spatial filter. Obtaining;
Obtaining a second contour of the heart by applying a coherence mapping method; And
Combining the first contour and the second contour to form a third contour of the heart.
The method of claim 1,
Determining the first contour of the heart is:
Measuring a magnetic signal generated in the myocardium using a core diagram apparatus;
Extracting a cardiac magnetic signal of a P-wave and a T-wave interval representing atrial and ventricle from a multi-channel measured core signal;
Calculating a magnetic field (leadfield vector) using a horizontally layered volume conductor model approximating the unit source current of the heart and the position information of the core magnetic sensor; And
Normalizing the leadfield vector to compensate for the size of the signal source over distance.
The method of claim 2,
Determining the first contour of the heart is:
Initializing an arbitrary weight matrix to an identity matrix;
Obtaining a gram matrix by combining the leadfield vector with an arbitrary weight matrix;
Setting a normalization parameter in the obtained gram matrix in consideration of the signal-to-noise ratio of the measurement signal;
Obtaining an array-gain constraint minimum-norm spatial filter by combining the normalized gram matrix and the normalized leadfield vector;
Obtaining power of an estimated source current vector using the obtained array gain limit spatial filter and the measured atrial and ventricular signals;
Checking whether the power of the obtained source current vector converges to a constant value;
Updating the weighted matrix using the estimated source vector when the obtained power of the source current vector does not converge to a constant value;
Normalizing the power of the obtained estimated source current vector; And
And reconstructing the first contour of the atrium and the ventricle by applying the power of the normalized source current vector to a first threshold set to a signal-to-noise ratio.
The method of claim 1,
Applying the coherence mapping method to find the second contour of the heart:
Obtaining points where the power of the source current vector is greatest in the atria and the ventricles, respectively;
Reconstructing signal waveforms (estimated source current vectors) in all source regions by linearly combining the measured core signal and the minimum-size spatial filter of the regression update gram matrix array gain;
Fast Fourier transforming the reconstructed signal waveform (estimated source current vector);
Obtaining cross-spectrum and auto-spectrum for all source regions in the atria and ventricles using the fast Fourier transformed signal;
Combining the obtained inter-spectrum with the self-spectrum to obtain coherence of the atrium and ventricular coherence;
Normalizing the coherence of the obtained atrium and ventricles; And
And reconstructing the second contour of the atrium and the ventricles by applying a second threshold set to the normalized coherence to a signal-to-noise ratio.
Determining the heart from the source current power obtained using an array-gain constraint minimum-norm with recursively updated gram matrix (AGN-RUG) spatial filter Cardiac contour reconstruction method comprising a.
The method of claim 5,
The steps to delineating the heart are:
Extracting a cardiac magnetic signal of a P-wave and a T-wave interval representing atrial and ventricle from a multi-channel measured core signal;
Calculating a magnetic field (leadfield vector) using a horizontally layered volume conductor model approximating the unit source current of the heart and the position information of the core magnetic sensor; And
Normalizing the leadfield vector to compensate for the size of the signal source over distance.
The method of claim 6,
Initializing an arbitrary weight matrix to an identity matrix;
Obtaining a gram matrix by combining the leadfield vector with an arbitrary weight matrix;
Adding a normalization parameter to the obtained gram matrix in consideration of the signal-to-noise ratio of the measurement signal;
Obtaining an array-gain constraint minimum-norm spatial filter by combining the normalized gram matrix and the normalized leadfield vector;
Obtaining power of an estimated source current vector using the obtained array gain limit spatial filter and the measured atrial and ventricular signals;
Checking whether the power of the obtained source current vector converges to a constant value;
Updating the weighted matrix using the estimated source vector when the obtained power of the source current vector does not converge to a constant value;
Normalizing the power of the obtained estimated source current vector;
And reconstructing the contours of the atrium and ventricles by applying the threshold of the normalized source current vector to the signal-to-noise ratio.
Obtaining points where the power of the estimated source current vector is greatest in the atria and the ventricles, respectively;
Reconstructing the signal waveforms (estimated source current vectors) in all source regions by linearly combining the measured core signal and the minimum size spatial filter of the regression update gram matrix array gain;
Fast Fourier transforming the reconstructed signal waveform (estimated source current vector);
Obtaining cross-spectrum and auto-spectrum for all source regions in the atria and ventricles using the fast Fourier transformed signal;
Combining the obtained inter-spectrum with the self-spectrum to obtain coherence of the atrium and ventricular coherence;
Normalizing the coherence of the obtained atrium and ventricles; And
Reconstructing the contours of the atria and ventricles by applying a threshold set to the normalized coherence signal-to-noise ratio.
A recording medium on which a program for executing the method of any one of claims 1 to 8 is recorded on a computer.
A core measuring apparatus including a core measuring sensor and a magnetic shield room and measuring a core measuring signal; And
The core also includes a processor for processing a signal to reconstruct the outline of the heart,
The processing unit:
The first contour of the heart is derived from the source current power obtained using an array-gain constraint minimum-norm with recursively updated gram matrix (AGN-RUG) spatial filter. Finding,
Applying a coherence mapping method to find the second contour of the heart, and
And the first contour and the second contour to mutually constitute a third contour of the heart.