Electrocardiogram detection method and device based on magnetoresistive sensor

Abstract

The present disclosure provides a magnetoresistive sensor-based electrocardiogram detection method and device. The magnetoresistive sensor-based electrocardiogram detection method comprises: setting a magnetic marker on a predetermined region of the precordial region of the measured object; setting a magnetoresistive sensor array at a predetermined distance of the precordial region of the measured object, and collecting magnetoresistive sensor signals containing ballistocardiogram signals and noise signals in real time through the magnetic field coupling between the magnetoresistive sensor array and the magnetic marker; preprocessing the magnetoresistive sensor signals, and extracting the ballistocardiogram signals from the preprocessed magnetoresistive sensor signals; and inputting the ballistocardiogram signals into a reconstruction model to output electrocardiogram waveforms. The method of collecting ballistocardiogram signals in real time and converting the ballistocardiogram signals into electrocardiogram signals through a model algorithm improves the accuracy and stability of the measurement, and is suitable for early prevention and long-term monitoring of cardiovascular diseases.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the field of medicine. In particular, it relates to a magnetoresistive sensor based electrocardiogram detection method and device. BACKGROUND

[0002] In the diagnosis of cardiovascular diseases (CVD), electrocardiogram (ECG) is an important biomedical signal, which provides important information for the diagnosis of cardiovascular diseases. Experiments show that by continuously monitoring and analyzing electrocardiogram, cardiovascular diseases can be better diagnosed, controlled and prevented. In each cardiac cycle, the result of myocardial depolarization and repolarization can be used to detect small electrical signal changes on the body electrode in the method of electrocardiogram monitoring.

[0003] At present, with the advancement of sensor and communication technology, real-time monitoring of electrocardiogram has used various low-power biomedical sensors and energy-saving wireless transmission, allowing lightweight systems to integrate with miniaturized components.

[0004] However, these schemes based on adhesive electrocardiogram electrodes are limited by the nature of skin attachment in continuous monitoring, and there are problems of monitoring reliability, adaptability and continuity. First, adhesive electrocardiogram electrodes and invasive monitoring systems can make patients feel uncomfortable, reducing the subjective willingness to wear such devices. Second, long-term continuous monitoring of electrocardiogram is required to capture occasional cardiac dysfunction events, but the limitations of the battery and the falling of adhesive electrodes will degrade continuity and reliability. In addition, in some cases such as burn patients, highly infected patients and premature infants, it is not feasible to attach adhesive electrocardiogram electrodes. Therefore, there is a need for an improved method and device for measuring electrocardiogram. SUMMARY

[0005] The present disclosure proposes a wireless magnetoresistive sensor based electrocardiogram detection method and device to overcome the limitations of traditional electrocardiogram measurement requiring contact electrodes. By real-time acquisition of ballistocardiogram signals and conversion of the ballistocardiogram signals into electrocardiogram waveforms through a model algorithm, the accuracy and stability of the measurement are improved, which is suitable for early prevention and long-term monitoring of cardiovascular diseases.

[0006] In one general aspect, there is provided a magnetoresistive sensor based electrocardiogram detection method, the electrocardiogram detection method comprising: disposing a magnetic marker on a predetermined region of the precordial region of a subject; disposing a magnetoresistive sensor array at a predetermined distance from the precordial region of the subject, real-time acquisition of a magnetoresistive sensor signal containing a ballistocardiogram signal and a noise signal through magnetic field coupling between the magnetoresistive sensor array and the magnetic marker; preprocessing the magnetoresistive sensor signal, and extracting the ballistocardiogram signal from the preprocessed magnetoresistive sensor signal; inputting the ballistocardiogram signal into a reconstruction model to output an electrocardiogram waveform.

[0007] Optionally, the pre-processing of the magnetoresistive sensor signal comprises: removing low-frequency baseline drift caused by body motion of the measured object from the magnetoresistive sensor signal by using a filter; separating time-frequency features from the filtered magnetoresistive sensor signal to extract a ballistocardiogram signal; identifying noise signals from the filtered magnetoresistive sensor signal; and performing motion artifact suppression on the ballistocardiogram signal based on the identified noise signals.

[0008] Optionally, the reconstruction model comprises a hybrid model fusing a convolutional neural network and a Transformer model, the hybrid model comprising a one-dimensional convolutional layer, a multi-head self-attention mechanism, and a fully connected layer, and the inputting of the ballistocardiogram signal into the reconstruction model to output an electrocardiogram waveform comprises: extracting spatial local features of the ballistocardiogram signal by the one-dimensional convolutional layer; modeling global temporal dependencies of the ballistocardiogram signal by the multi-head self-attention mechanism; and concatenating the spatial local features and the global temporal dependencies by the fully connected layer to output a voltage amplitude sequence of the electrocardiogram as the electrocardiogram waveform.

[0009] Optionally, the reconstruction model is trained by: obtaining sample data and label data, the label data comprising a contact ECG signal synchronized with the data of the magnetoresistive sensor signal; inputting the sample data into the reconstruction model to obtain an electrocardiogram waveform output by the reconstruction model through a spatio-temporal attention module jump connection between an encoder-decoder in the one-dimensional convolutional layer; determining a loss value by a joint loss function based on the output electrocardiogram waveform and the label data; and adjusting waveform reconstruction error and detection accuracy of key feature points of the hybrid model based on the loss value.

[0010] Optionally, the joint loss function comprises: a time-domain loss function for determining a first loss based on L1 loss and L2 loss of the electrocardiogram waveform; a frequency-domain loss function for determining a second loss based on spectral energy distribution consistency constraint of short-time Fourier transform; and a feature adversarial loss function for determining a third loss by learning a mapping from the ballistocardiogram signal to the electrocardiogram waveform through a discriminator, wherein the step of determining the loss value by the joint loss function comprises: obtaining the loss value according to a weighted sum of the first loss, the second loss, and the third loss.

[0011] Optionally, the magnetoresistive sensor array comprises at least three magnetoresistive sensors pointing in different directions from each other.

[0012] In one general aspect, there is provided a magnetoresistive sensor based electrocardiogram detection device, comprising: a magnetic marker arranged at a predetermined region of a precordial body surface of a subject; a magnetoresistive sensor array arranged at a predetermined distance from the precordial region of the subject, for real-time acquisition of magnetoresistive sensor signals containing ballistocardiogram signals and noise signals via magnetic field coupling between the magnetoresistive sensor array and the magnetic marker; an analog signal processing unit comprising multiple stages of amplifiers and filter circuits, and for extracting ballistocardiogram signals from the pre-processed magnetoresistive sensor signals; and a computing unit comprising a reconstruction model, for inputting the ballistocardiogram signals into the reconstruction model to output electrocardiogram waveforms.

[0013] Optionally, the analog signal processing unit comprises, in sequence: a first stage of high-pass filtering and low-noise amplification with a cutoff frequency of 0.1 Hz and a gain ≥ 60 dB; a second stage of band-pass filtering and amplification with a gain range of 20-40 dB; and a third stage of notch filtering for suppressing power frequency interference and environmental magnetic field noise.

[0014] Optionally, the magnetoresistive sensor array is arranged to cover the precordial region projection area in orthogonal three-axes at a predetermined distance from the precordial region of the subject, and the distance between adjacent magnetoresistive sensors is less than 40 mm.

[0015] In one general aspect, there is provided a computer readable storage medium storing a computer program which, when executed by a processor, implements the magnetoresistive sensor based electrocardiogram detection method as described above.

[0016] The magnetoresistive sensor based electrocardiogram detection method and device according to the present disclosure improve the convenience, accuracy and stability of measurement.

[0017] Effects of the present disclosure are not limited to what has been described herein above and other benefits and effects will also be apparent from those skilled in the art upon consideration of the specification. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings, which are included to provide a further understanding of the disclosure and are incorporated in and constitute a part of this specification, illustrate embodiments of the disclosure and together with the description serve to explain the principles of the disclosure. In the drawings:

[0019] Figure 1 is a flowchart illustrating a magnetoresistive sensor based electrocardiogram detection method according to an embodiment of the present disclosure.

[0020] Figure 2 is a flowchart illustrating pre-processing of magnetoresistive sensor signals according to an embodiment of the present disclosure.

[0021] Figure 3 is a schematic diagram of magnetoresistive sensor arrangement according to an embodiment of the present disclosure.

[0022] Figure 4 is a model block diagram of generating an electrocardiogram, illustrating embodiments of the present disclosure.

[0023] Figure 5 is a flowchart of pre-processing a BCG signal, illustrating embodiments of the present disclosure. DETAILED DESCRIPTION

[0024] The following detailed description is provided to aid in understanding the method, apparatus, and / or system described herein. However, various changes, modifications, and equivalents of the method, apparatus, and / or system described herein can be apparent to those skilled in the art after understanding the disclosure provided herein. For example, the order of the operations described herein is merely an example and is not limited to those set forth herein, but can be changed as would be apparent to one of ordinary skill in the art after understanding the disclosure provided herein. Moreover, descriptions of features known to those of ordinary skill in the art can be omitted for the sake of clarity and conciseness.

[0025] The features described herein can be implemented in different forms and should not be construed as limited to the examples described herein. Rather, these examples are provided as illustrative of only some of the many possible implementations of the method, apparatus, and / or system described herein.

[0026] As used herein, the term "and / or" includes any one of the associated listed items, as well as any combination of any two or more of the associated listed items.

[0027] Although terms such as "first", "second", and "third" can be used herein to describe various components, assemblies, regions, layers, or sections, these components, assemblies, regions, layers, or sections should not be limited by these terms. Rather, these terms are used only to distinguish one component, assembly, region, layer, or section from another component, assembly, region, layer, or section. Thus, a component, assembly, region, layer, or section referred to as a first component, a first assembly, a first region, a first layer, or a first section in the examples described herein can also be referred to as a second component, a second assembly, a second region, a second layer, or a second section without departing from the teachings of the examples.

[0028] In the specification, when an element (such as a layer, a region, or a substrate) is referred to as "on" another element, "connected to" or "coupled to" another element, it can be directly on, directly connected to, or directly coupled to the other element, or one or more other elements can be interposed therebetween. In contrast, when an element is referred to as being "directly on," "directly connected to," or "directly coupled to" another element, no other elements are interposed therebetween.

[0029] The terminology used herein is for the purpose of describing various examples only and is not intended to be limiting of the disclosure. Singular forms are intended to include the plural forms as well, unless the context clearly indicates otherwise. The terms "comprises," "comprising," "includes," and "including" specify the presence of stated features, numbers, operations, components, elements, and / or a combination thereof, but do not preclude the presence or addition of one or more other features, numbers, operations, components, elements, and / or a combination thereof.

[0030] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs when read in light of the present disclosure. Unless explicitly defined otherwise, terms such as, for example, "a", "an", and "the" are not intended to be limiting, but rather are to be read in the context of the relevant specification and claims to reflect the intended meaning. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and the present disclosure, and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0031] Further, in the description of the examples, detailed descriptions of well-known related structures or functions are omitted when it is deemed that such detailed description will cause ambiguous interpretation of the present disclosure.

[0032] In wireless sensing technology, a magnetoresistive sensor (TMR, GMR, or AMR) is a sensor based on the principle of magnetoresistance change, with characteristics of high sensitivity, low power consumption, fast response, etc., including magnetic memory, magnetic sensor, magnetic head, magnetic disk drive, magnetic resonance imaging, etc. The core principle of the magnetoresistive sensor is the tunnel magnetoresistance effect (i.e., the resistance of the material changes under the action of an applied magnetic field), which makes it possible to indirectly measure the magnetic field by measuring the resistance change of the material. The high sensitivity of the magnetoresistive sensor is suitable for measuring weak magnetic field changes, the low power consumption is suitable for battery-powered application scenarios, the fast response is suitable for real-time measurement of magnetic field changes, and the wide working temperature range is suitable for working in various environments.

[0033] The present disclosure proposes to apply wireless sensing technology to electrocardiogram (ECG) waveform measurement. Real-time ballistocardiogram (BCG) signal waveforms are obtained by a magnetoresistive sensor, and after performing average processing on the real-time BCG signal waveforms, key waveforms (e.g., P wave, Q wave, R wave, S wave, T wave, and U wave, etc.) in the electrocardiogram waveform can be distinguished for monitoring and diagnosis of heart activity. The present disclosure proposes to measure a ballistocardiogram signal by a magnetoresistive sensor, and to convert the ballistocardiogram signal into an electrocardiogram waveform through model training, breaking through the limitation of traditional electrocardiogram monitoring requiring a contact sensor, and providing an innovative solution for early prevention and long-term monitoring of cardiovascular diseases.

[0034] The present disclosure proposes a magnetoresistive sensor-based electrocardiogram detection method, Figure 1 is a flowchart of a magnetoresistive sensor-based electrocardiogram detection method illustrating an embodiment of the present disclosure. Referring toFigure 1 The magnetoresistive sensor-based electrocardiogram detection method comprises the following steps:

[0035] In step 101, a magnetic marker is arranged on a predetermined region of the precordial surface of the subject.

[0036] In step 102, a magnetoresistive sensor array is arranged at a predetermined distance from the precordium of the subject, and a magnetoresistive sensor signal containing a ballistocardiogram signal and a noise signal is collected in real time through magnetic field coupling between the magnetoresistive sensor array and the magnetic marker. The noise signal in the magnetoresistive sensor signal can include a respiratory signal and a motion artifact, etc.

[0037] In step 103, the magnetoresistive sensor signal is preprocessed, and the ballistocardiogram signal is extracted from the preprocessed magnetoresistive sensor signal.

[0038] The preprocessing of the magnetoresistive sensor signal can include dynamic baseline correction, and the ballistocardiogram signal is extracted from the magnetoresistive sensor signal by wavelet packet decomposition, but is not limited thereto.

[0039] In step 104, the ballistocardiogram signal is input into a reconstruction model to output an electrocardiogram waveform.

[0040] The reconstruction model can be a pre-trained deep neural network model, and the training can be performed by synchronously calibrating the BCG signal and the ECG signal.

[0041] Specifically, Figure 2 is a flowchart showing the preprocessing of the magnetoresistive sensor signal, as Figure 2 shown, according to an embodiment, the preprocessing of the magnetoresistive sensor signal can include:

[0042] In step 201, a filter is used to eliminate low-frequency baseline drift in the magnetoresistive sensor signal caused by body movement of the subject. In step 202, the time-frequency features are separated from the filtered magnetoresistive sensor signal to extract the ballistocardiogram signal. For example, the time-frequency features of the ballistocardiogram signal in the magnetoresistive sensor signal can be separated by empirical mode decomposition (EMD). In step 203, the ballistocardiogram signal is subjected to motion artifact suppression based on the identified noise signal. For example, the periodic template matching algorithm can be used to suppress the motion artifact of the ballistocardiogram.

[0043] According to one embodiment of the present disclosure, a magnetoresistive sensor-based electrocardiogram detection device is provided, which comprises a magnetic marker, a magnetoresistive sensor, an analog signal processing unit, and a computing unit.

[0044] The magnetic marker can be arranged on a predetermined region of the precordial surface of the subject.

[0045] As an example, the magnetic marker can be a permanent magnet material (e.g., NdFeB permanent magnet or SmCo permanent magnet), which can be attached to the heart or abdominal region of the measured object, and the signal-to-noise ratio can be improved by enhancing the local magnetic field strength.

[0046] The magnetoresistance sensor array can be arranged at a predetermined distance from the precordial region of the measured object, and the magnetoresistance sensor signals containing the ballistocardiogram signals and noise signals can be collected in real time through the magnetic field coupling between the magnetoresistance sensor array and the magnetic marker.

[0047] As an example, the magnetoresistance sensor can be a high-sensitivity sensor arranged on a fixed support, kept at a distance of less than 0.5 meters from the precordial region of the measured object, and covering the front, left, and right three-dimensional directions of the heart region of the measured object in space, for detecting the magnetic field signals generated by the magnetic marker.

[0048] The analog signal processing unit can include a multi-stage amplifier and filter circuit, and extract the ballistocardiogram signals from the preprocessed magnetoresistance sensor signals.

[0049] As an example, the analog signal processing unit can include a preamplifier, a filter, and an analog-to-digital conversion unit. The preamplifier can adopt a low-noise design with a gain ≥ 60 dB, capable of lifting the nV-level sensor output to the collectable range. The filter can integrate high-pass filtering (cutoff frequency 0.1 Hz) and band-pass filtering, capable of suppressing baseline drift and high-frequency noise.

[0050] According to an embodiment, the analog signal processing unit includes, in sequence: a first-stage high-pass filter and low-noise amplifier with a cutoff frequency of 0.1 Hz and a gain ≥ 60 dB; a second-stage band-pass filter and amplifier with a gain range of 20-40 dB; and a third-stage notch filter for suppressing power frequency interference and environmental magnetic field noise.

[0051] As an example, the specific topology of the amplifier and filter in the analog signal processing unit can include six elements of a first-stage high-pass filter, a first-stage amplifier, a first-stage low-pass filter, a second-stage high-pass filter, a second-stage amplifier, and a second-stage low-pass filter. The first-stage amplifier can be selected as an AD8429 chip, for example, which is capable of measuring tiny signals with extremely low noise, only 1 nV / Hz input noise performance, while having a very high common-mode rejection ratio (CMRR) to prevent unnecessary signal distortion. The high-pass filter and the low-pass filter in the circuit can be combined to form a band-pass filter, because the BCG signal is mainly concentrated in the frequency range of 0.1-10 Hz, therefore the passband of the circuit is set to the range of 0.05-30 Hz, which can eliminate the influence of 50 Hz power frequency noise and completely cover the frequency range of the BCG signal.

[0052] The computing unit can comprise a reconstruction model, and the ballistocardiogram signal is input into the reconstruction model to output an electrocardiogram waveform.

[0053] Formula 1 below is a calculation formula of the sensitivity S of the magnetoresistive sensor:

[0054] (1)

[0055] wherein, is a voltage change value output by the magnetoresistive sensor, is a change value of the magnetic field intensity, is an operating voltage applied to a power supply end of the magnetoresistive sensor. In the measurement process, the magnetic field is generated by a Helmholtz coil with known calibration parameters, and the magnetic field intensity B generated by the coil is related to the current input into the Helmholtz coil, as shown in formula 2 below:

[0056] (2)

[0057] wherein, is the current intensity of the input coil, and is related to the voltage input into the coil and the resistance of the coil itself, is a coil coefficient, representing a conversion coefficient of the current and the magnetic field, and is related to parameters such as the number of turns, thickness, and spacing of the coil. The sensitivity measurement of the magnetoresistive sensor can be performed as follows: within a magnetic field range of ±20 uT, the input direct current signal is sequentially increased, and then the input direct current signal is continuously decreased, the coil voltage and the voltage parameter output by the sensor are recorded, and the hysteresis curve of the sensor is obtained. It can be known from the sensitivity obtained by measurement and calculation that the circuit noise is mainly derived from the sensor, and therefore the magnetic noise data of the sensor can be converted from the circuit noise.

[0058] The analog signal processing unit can further comprise an analog-to-digital converter, and the analog-to-digital converter has a 24-bit resolution and a sampling rate ≥1 kHz. The analog signal processing unit can perform amplification and filtering processing on the sensing signal of the magnetoresistive sensor, and the digital signal can be output to the digital signal processing unit through the analog-to-digital converter.

[0059] Optionally, the electrocardiogram detection device based on the magnetoresistive sensor can comprise a digital signal processing unit, the digital signal processing unit can adopt a high-precision acquisition device (for example, a PowerLab system), record multiple-channel signals in real time, and realize signal visualization, heart rate calculation, and preliminary analysis through a computer. The digital signal processing unit can output a ballistocardiogram signal.

[0060] According to an embodiment, the magnetoresistive sensor array is arranged to cover the precordial projection area of the precordial region of the measured object at a predetermined distance in the orthogonal three-axial manner, and the distance between adjacent magnetoresistive sensors is less than 40 mm.

[0061] Figure 3is a schematic diagram of a magnetoresistive sensor arrangement illustrating embodiments of the present disclosure.

[0062] Referring to Figure 3 , the magnetic marker 100 can be attached to the heart projection area magnetic marker for enhancing the target magnetic signal. The magnetoresistive sensor array 200 can be arranged to fix the multi-channel magnetoresistive sensor array to a special bracket, adjusted to a predetermined distance (for example, 1-5 cm apart) from the precordial region of the measured object, and ensure that the magnetic marker is covered in three-dimensional space. The magnetoresistive sensor array can include at least three magnetoresistive sensors that are different from each other. As shown in Figure 3 , the magnetoresistive sensor array 200 can include 9 sensors in the form of a 3X3 array, and their detection directions are different from each other in three-dimensional space, for example, it can include at least three sensors respectively facing the three coordinate axis directions of an orthogonal coordinate system.

[0063] According to an embodiment, the reconstruction model includes a hybrid model that fuses a convolutional neural network and a Transformer model, the hybrid model including a one-dimensional (1D) convolutional layer, a multi-head self-attention mechanism, and a fully connected layer, and inputting the ballistocardiogram signal into the reconstruction model to output an electrocardiogram waveform includes: extracting spatial local features of the ballistocardiogram signal through the one-dimensional convolutional layer; modeling global temporal dependencies of the ballistocardiogram signal through the multi-head self-attention mechanism; and concatenating the spatial local features and the global temporal dependencies through the fully connected layer to output a voltage amplitude sequence of the electrocardiogram as the electrocardiogram waveform.

[0064] Figure 4 is a model block diagram for generating an electrocardiogram, illustrating embodiments of the present disclosure.

[0065] As an example, referring to Figure 4 , the reconstruction model can be designed based on a deep learning cross-modal generation hybrid architecture.

[0066] The reconstruction model as a whole can adopt a 1D U-Net framework: a 1D U-Net structure with a contraction-expansion path, which fuses shallow details (such as waveform steep changes) and deep semantic features through a skip connection, and is suitable for pixel-level temporal mapping from BCG to ECG. At the same time, local features and global dependencies are jointly modeled by combining convolutional neural networks (CNN) and Transformers.

[0067] In addition, a PatchGAN discriminator (not shown in Figure 4 ) can be used to distinguish between cross-modal generated ECG signals and real ECG signals to achieve high-fidelity conversion. By learning the mapping from BCG to ECG signals through the discriminator, the complex nonlinear relationship between the original ballistocardiogram signal and the electrocardiogram signal is solved, and a more realistic and physiologically consistent electrocardiogram signal can be generated.

[0068] The hybrid encoder (BCG signal coding) is composed of an input processing layer and a plurality of down-sampling layers, each of which includes a global feature extraction module and a down-sampling module. The input processing layer uses a 1D convolution layer and a max-pooling layer to compress and encode the input BCG signal and extract local features. The global feature extraction module is stacked with a Transformer encoder, including a multi-head attention mechanism and position encoding, and then connected with a residual connection, layer normalization and a feedforward network to obtain global correlation features. The down-sampling module uses a 1D convolution layer with a step size of 2 to halve the sequence length.

[0069] The multi-scale decoder (ECG generation) is composed of a plurality of up-sampling layers, each of which includes a feature fusion module and a waveform generation module. The feature fusion module is stacked with a Transformer decoder, and the cross-modal attention mechanism has Query from the decoder and Key and Value from the encoder. The waveform generation module uses a 1D deconvolution layer to gradually restore the time sequence dimension of the signal features to the original length.

[0070] In the training process of the precordial electrocardiogram data measurement, the measured object is kept in a prone position or a supine position, a seat, a yoga mat or a mattress is used to stabilize the body position, a multi-channel signal is recorded synchronously by starting the acquisition system, and the multi-channel signal at least includes a magnetoresistance sensor signal and a contact ECG signal synchronized with the data of the magnetoresistance sensor signal.

[0071] According to the embodiment, the reconstruction model is trained by the following steps: obtaining sample data and label data, the label data including a contact ECG signal synchronized with the data of the magnetoresistance sensor signal; inputting the sample data into the reconstruction model to obtain an electrocardiogram waveform output by the reconstruction model through a spatio-temporal attention module jump connection between the encoder-decoder in the one-dimensional convolution layer; determining a loss value based on the output electrocardiogram waveform and the label data through a joint loss function; and adjusting the waveform reconstruction error of the hybrid model and the detection accuracy of the key feature points based on the loss value. Through the jump connection, high-frequency details can be preserved, and multi-task optimization is performed on multiple key feature points such as QRS complexes and T waves.

[0072] According to an embodiment, the reconstruction model is trained by using a joint loss function, wherein the joint loss function comprises: a time domain loss function, a first loss is determined based on L1 loss and L2 loss of electrocardiogram waveform; a frequency domain loss function, a second loss is determined based on consistency constraint of short-time Fourier transform spectrum energy distribution; and a feature adversarial loss function, a third loss is determined by learning the mapping from ballistocardiogram signal to electrocardiogram waveform through a discriminator, wherein the step of determining the loss value by using the joint loss function comprises: obtaining the loss value according to the weighted sum of the first loss, the second loss and the third loss. The joint loss function can solve the complex nonlinear relationship between the original ballistocardiogram signal and the electrocardiogram waveform, and can generate more real and physiological electrocardiogram waveform.

[0073] As an example, the multi-objective loss function can include a time domain loss, a frequency domain loss and a feature adversarial loss.

[0074] The time domain loss, i.e., waveform reconstruction loss, uses weighted L1 loss (also known as absolute value loss function) and L2 loss (also known as least square error) to constrain signal amplitude matching, and the time domain loss L rec As shown in the following formula 3:

[0075] (3)

[0076] Wherein, λ1 and λ2 represent weights.

[0077] The frequency domain loss calculates the spectrum difference by short-time Fourier transform (STFT), and ensures the consistency of energy distribution in key frequency bands (such as 0.5-40Hz), and the frequency domain loss L spec As shown in the following formula 4:

[0078] (4)

[0079] Wherein, x represents the sample input of the model, and y represents the output generated by the model.

[0080] The feature adversarial loss can automatically capture complex features in the ballistocardiogram signal through the adversarial process between the generator (U-Ne) and the discriminator, improve the accuracy and stability of the conversion process, and thus more effectively reconstruct the real electrocardiogram waveform, and the feature adversarial loss L adv As shown in the following formula 5:

[0081] (5)

[0082] Wherein, x represents the sample input of the model, and y represents the output generated by the model. G ( x ) is the sample generated by the generator; D ( y) is the judgment result of the discriminator on the sample (usually 1, indicating true); D ( G ( x )) is the judgment result of the discriminator on the generated sample (usually 0, indicating fake).

[0083] In summary, the total loss is as follows formula 6:

[0084] (6)

[0085] The training steps can be divided into stages, that is, first pre-training the reconstruction loss, and then introducing the discriminator. First, the time domain loss and the frequency domain loss are used to pre-train the encoder-decoder (U-Ne). Then, the discriminator is trained using a fixed encoder, and end-to-end joint fine-tuning is performed. The training steps can also include an early stopping strategy (patience = 20 epochs), which triggers early stopping if the loss function value does not significantly decrease.

[0086] Figure 5 is a flowchart showing the preprocessing of the BCG signal according to an embodiment of the present disclosure, as shown in Figure 5 According to the embodiment, preprocessing the BCG signal can include:

[0087] According to an embodiment of the present disclosure, in the training of the reconstruction model, before training the reconstruction model to convert the ballistocardiogram signal into an electrocardiogram waveform, the BCG signal needs to be preprocessed. The steps of preprocessing can include: in step 501, performing signal alignment, which can use dynamic time warping (DTW) to align the BCG and ECG signals; in step 502, using band-pass filtering to remove noise; in step 503, unifying the sampling rate, which can perform resampling on the collected BCG signal with respect to the reference sample frequency; in step 504, performing data augmentation, which performs random time warping, adds Gaussian noise, and performs channel random offset processing on the data; in step 505, standardization processing, which performs z-score normalization for individuals.

[0088] The present application balances the local feature capture and global temporal modeling capability through a hybrid architecture, combines the discriminator to learn the complex nonlinear relationship between the original ballistocardiogram signal and the electrocardiogram waveform, and ensures that the generated electrocardiogram waveform is highly similar to the real signal in structure and morphology. The electrocardiogram detection method and device based on magnetic resistance sensors according to the present disclosure improve the convenience, accuracy and stability of measurement.

[0089] An exemplary embodiment according to the present application also provides a computer-readable storage medium storing a computer program. The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to execute the magnetoresistive sensor-based electrocardiogram detection method according to the present application. The computer-readable recording medium is any data storage device that can store data readable by a computer system. Examples of the computer-readable recording medium include a read-only memory, a random access memory, a read-only optical disk, a magnetic tape, a floppy disk, an optical data storage device, and a carrier wave such as data transmission through an Internet via a wired or wireless transmission path.

[0090] An exemplary embodiment according to the present application also provides a computer device. The computer device includes a processor and a memory. The memory stores a computer program. The computer program is executed by the processor to cause the processor to execute the magnetoresistive sensor-based electrocardiogram detection method according to the present application.

[0091] Accordingly, example embodiments of the present disclosure can be implemented as methods implemented in a computer or a non-transitory computer readable medium storing computer executable instructions. In example embodiments, the computer readable instructions, when executed by a processor, can perform a method according to at least one aspect of the present disclosure.

[0092] Further, the method according to example embodiments of the present disclosure can be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer readable medium.

[0093] The computer readable medium can include program instructions, data files, data structures, etc., alone or in combination. The program instructions recorded on the computer readable medium can be specifically designed and configured for example embodiments of the present disclosure, or can be known and available to those skilled in the computer software field. The computer readable recording medium can include a hardware device configured to store and execute program instructions. For example, the computer readable recording medium can be or include a magnetic medium such as a hard disk, a floppy disk, and a magnetic tape, an optical medium such as a CD-ROM and a DVD, a magneto-optical medium such as a magneto-optical floppy disk, a ROM, a RAM, a flash memory, etc. The program instructions can include not only machine language codes generated by a compiler, but also high-level language codes executable by a computer through an interpreter, etc.

[0094] While the application includes specific examples, it will be apparent to one of ordinary skill in the art that various changes in form and detail can be made without departing from the spirit and scope of the claims and their equivalents. The examples described herein are to be considered in a descriptive sense only and not for purposes of limitation. Descriptions of features or aspects in each example are to be considered as being applicable to similar features or aspects in other examples. Proper results can be obtained if the described techniques are performed in a different order, and / or if the components of the described systems, architectures, devices or circuits are combined in a different manner, and / or if the described systems, architectures, devices or circuits are implemented using different components or their equivalents. Therefore, the scope of the application is defined not by the specific embodiments, but by the appended claims and their equivalents, with all variations alterations and modifications to the examples described herein being considered to be within the scope of the application.

Claims

1. A magnetoresistive sensor based electrocardiogram detection method, characterized by, The electrocardiogram detection method comprises: setting a magnetic marker on a predetermined region of the precordial surface of the measured object; setting a magnetoresistive sensor array at a predetermined distance from the precordial region of the measured object, and acquiring magnetoresistive sensor signals containing ballistocardiogram signals and noise signals through magnetic field coupling between the magnetoresistive sensor array and the magnetic marker; preprocessing the magnetoresistive sensor signals, and extracting the ballistocardiogram signals from the preprocessed magnetoresistive sensor signals; inputting the ballistocardiogram signals into a reconstruction model to output electrocardiogram waveforms.

2. The electrocardiogram detection method according to claim 1, characterized by, The preprocessing of the magnetoresistive sensor signals comprises: eliminating low-frequency baseline drift caused by body movement of the measured object from the magnetoresistive sensor signals by using a filter; separating time-frequency features from the filtered magnetoresistive sensor signals to extract the ballistocardiogram signals; identifying noise signals from the filtered magnetoresistive sensor signals; and performing motion artifact suppression on the ballistocardiogram signals based on the identified noise signals.

3. The electrocardiogram detection method according to claim 1, characterized by, The reconstruction model comprises a hybrid model combining a convolutional neural network and a Transformer model, and the hybrid model comprises a one-dimensional convolutional layer, a multi-head self-attention mechanism, and a fully connected layer. The inputting of the ballistocardiogram signals into the reconstruction model to output electrocardiogram waveforms comprises: extracting spatial local features of the ballistocardiogram signals through the one-dimensional convolutional layer; modeling global time sequence dependencies of the ballistocardiogram signals through the multi-head self-attention mechanism; and concatenating the spatial local features and the global time sequence dependencies through the fully connected layer to output a voltage amplitude sequence of the electrocardiogram as the electrocardiogram waveforms.

4. The electrocardiogram detection method according to claim 3, characterized by, The reconstruction model is trained by the following steps: obtaining sample data and label data, the label data comprising a contact ECG signal synchronized with the data of the magnetoresistive sensor signals; inputting the sample data into the reconstruction model to obtain electrocardiogram waveforms output by the reconstruction model through a spatio-temporal attention module jump connection between an encoder-decoder in the one-dimensional convolutional layer; determining a loss value based on the output electrocardiogram waveforms and the label data through a joint loss function; and adjusting waveform reconstruction errors and detection accuracy of key feature points of the hybrid model based on the loss value.

5. The electrocardiogram detection method according to claim 4, characterized by, The joint loss function comprises: a time-domain loss function for determining a first loss through point-by-point matching based on L1 loss and L2 loss of the electrocardiogram waveforms; a frequency-domain loss function for determining a second loss based on spectral energy distribution consistency constraints of a short-time Fourier transform; and a feature adversarial loss function for determining a third loss through a discriminator learning mapping from the ballistocardiogram signals to the electrocardiogram waveforms, wherein the step of determining the loss value based on the joint loss function comprises obtaining the loss value according to a weighted sum of the first loss, the second loss, and the third loss.

6. The electrocardiogram detection method according to claim 1, characterized by, The magnetoresistive sensor array comprises at least three magnetoresistive sensors pointing in different directions.

7. A magnetoresistive sensor based electrocardiogram detection device, characterized by The electrocardiogram detection device comprises: a magnetic marker arranged on a predetermined region of the precordial surface of the measured object; a magnetoresistive sensor array arranged at a predetermined distance from the precordial region of the measured object, and acquiring magnetoresistive sensor signals containing ballistocardiogram signals and noise signals through magnetic field coupling between the magnetoresistive sensor array and the magnetic marker; an analog signal processing unit comprising multiple stages of amplifiers and filter circuits and extracting ballistocardiogram signals from the pre-processed magnetoresistive sensor signals; a computing unit comprising a reconstruction model into which the ballistocardiogram signals are inputted to output electrocardiogram waveforms.

8. The electrocardiogram detection device according to claim 7, characterized in that The analog signal processing unit comprises sequentially arranged: a first stage of high-pass filtering and low-noise amplification with a cut-off frequency of 0.1 Hz and a gain ≥ 60 dB; a second stage of band-pass filtering and amplification with a gain range of 20-40 dB; and a third stage of notch filtering for suppressing power-line interference and ambient magnetic field noise. The magnetoresistive sensor array is arranged to cover the precordial projection area in a predetermined distance from the precordium of the subject in an orthogonal three-axis configuration with a sensor spacing of less than 40 mm.

9. The electrocardiogram detection apparatus according to claim 7, characterized by The computer program, when executed by a processor, implements the magnetoresistive sensor based electrocardiogram detection method as claimed in any one of claims 1 to 6.

10. A computer readable storage medium storing a computer program, characterized in that, ​

Images