Method and device for assisting in predicting recessive atrial fibrillation and electronic equipment

Abstract

The invention provides a method and device for assisting in predicting recessive atrial fibrillation and electronic equipment, and is applied to the technical field of electromagnetic induction. The method comprises the following steps: acquiring an initial electromagnetic echo signal; performing signal separation processing on the initial electromagnetic echo signal to obtain a target electromagnetic echo signal corresponding to the motion state of the heart area of the measured object in a predetermined time period; performing first feature extraction processing on the target electromagnetic echo signal by using a space-time encoder of the information detection model to generate a plurality of mechanical motion space-time features related to normal heartbeat and abnormal heartbeat of the heart area; performing second feature extraction processing on the plurality of mechanical motion time-space features by using a feature extraction encoder of the information detection model to generate target mechanical motion time features related to abnormal heartbeat of the heart area; and performing feature conversion mapping processing on the target mechanical motion time features by using a detection evaluation sub-model of the information detection model to obtain target prediction heartbeat information of the detected object.

Description

Technical Field

[0001] The present invention relates to the field of electromagnetic induction technology, and more specifically to a method, a device and an electronic device for assisting in predicting latent atrial fibrillation. Background Art

[0002] Atrial fibrillation is a common type of arrhythmia and also a common type of cardiovascular disease. It increases the risk of stroke or other serious problems. Electrocardiogram, as a common auxiliary detection image, is used to help make a targeted diagnosis of whether the target has atrial fibrillation. However, due to the paroxysmal and hidden nature of latent atrial fibrillation, even if the target is subjected to multiple ECG examinations, there is still a high probability that the target's latent atrial fibrillation cannot be detected, resulting in an increased risk of serious problems.

[0003] In the process of realizing the above-mentioned inventive concept, the inventors found that: in the related technology, due to the paroxysmal and concealed characteristics of latent atrial fibrillation, it is impossible to intuitively obtain accurate auxiliary images containing latent atrial fibrillation information, thereby failing to timely assist relevant personnel in making predictions and judgments. In addition, while obtaining latent atrial fibrillation information, it is necessary to use contact equipment to perform multiple detections on the target, resulting in technical problems of low processing efficiency and poor flexibility. Summary of the invention

[0004] In view of the above problems, the present invention provides a method, a device and an electronic device for assisting in predicting latent atrial fibrillation.

[0005] According to a first aspect of the present invention, there is provided a method for assisting the prediction of latent atrial fibrillation, comprising: acquiring an initial electromagnetic echo signal, wherein the initial electromagnetic echo signal represents a plurality of signals that are spatially overlapped with each other and reflected by a plurality of body regions of a subject under test within a predetermined time period; performing signal separation processing on the initial electromagnetic echo signal to obtain a target electromagnetic echo signal corresponding to the motion state of a heart region of the subject under test within the predetermined time period; performing a first feature extraction processing on the target electromagnetic echo signal using a spatiotemporal encoder of an information detection model to generate a plurality of mechanical motion spatiotemporal features related to normal heartbeats and abnormal heartbeats of the heart region; performing a second feature extraction processing on the plurality of mechanical motion spatiotemporal features using a feature extraction encoder of the information detection model to generate a target mechanical motion time feature related to an abnormal heartbeat of the heart region; performing feature conversion mapping processing on the target mechanical motion time feature using a detection evaluation submodel of the information detection model to obtain target auxiliary prediction heartbeat information of the subject under test.

[0006] According to an embodiment of the present invention, a signal separation process is performed on an initial electromagnetic echo signal to obtain a target electromagnetic echo signal corresponding to the motion state of a heart region of a measured object within a predetermined time period, including: performing beamforming processing on the initial electromagnetic echo signal to obtain a plurality of intermediate electromagnetic echo signals, wherein the intermediate electromagnetic echo signal represents a plurality of four-dimensional signals spatially separated from each other and reflected by a plurality of regions within a predetermined time period; extracting a three-dimensional electromagnetic echo signal corresponding to the heart region of the measured object from the plurality of intermediate electromagnetic echo signals; and filtering the three-dimensional electromagnetic echo signal using a second-order differential time-domain filtering algorithm to obtain a target electromagnetic echo signal.

[0007] According to an embodiment of the present invention, a three-dimensional electromagnetic echo signal corresponding to the heart region of the object under test is extracted from a plurality of intermediate electromagnetic echo signals, including: superimposing the plurality of intermediate electromagnetic echo signals to obtain an intermediate electromagnetic echo superposition signal; processing the intermediate electromagnetic echo superposition signal using a constant false alarm rate algorithm to obtain a distance electromagnetic echo signal corresponding to the chest region of the object under test, wherein the distance electromagnetic echo signal is a three-dimensional signal; and clustering the distance electromagnetic echo signal using a density clustering algorithm to obtain a three-dimensional electromagnetic echo signal corresponding to the heart region of the object under test.

[0008] According to an embodiment of the present invention, a first feature extraction process is performed on a target electromagnetic echo signal using a space-time encoder of an information detection model to generate a plurality of mechanical motion space-time features related to normal heartbeat and abnormal heartbeat in the heart region, including: based on a predetermined coding period, splitting the target electromagnetic echo signal into target electromagnetic echo sub-signals of a plurality of time periods; respectively inputting the target electromagnetic echo sub-signals of the plurality of time periods into the space-time encoder of the information detection model for a first feature extraction process to generate a plurality of mechanical motion space-time features related to normal heartbeat and abnormal heartbeat in the heart region, wherein the plurality of mechanical motion space-time features correspond to the target electromagnetic echo sub-signals of the plurality of time periods.

[0009] According to an embodiment of the present invention, the detection and evaluation submodel includes a first double-layer fully connected network and a second double-layer fully connected network, and the first double-layer fully connected network and the second double-layer fully connected network both include two linear layers; the detection and evaluation submodel of the information detection model is used to perform feature conversion mapping processing on the target mechanical motion time characteristics to obtain the target auxiliary predicted heartbeat information of the object under test, including: inputting the target mechanical motion time characteristics into the first linear layer of the first double-layer fully connected network and the first linear layer of the second double-layer fully connected network respectively for dimensionality reduction processing to obtain the first reduced dimensionality mechanical motion state characteristics and the second reduced dimensionality mechanical motion time characteristics; inputting the first reduced dimensionality mechanical motion state characteristics into the second linear layer of the first double-layer fully connected network for feature conversion mapping processing to obtain the auxiliary atrial beating information of the object under test; inputting the second reduced dimensionality mechanical motion time characteristics into the second linear layer of the second double-layer fully connected network for feature conversion mapping processing to obtain the auxiliary atrial fibrillation information of the object under test.

[0010] According to an embodiment of the present invention, the information detection model is trained in the following manner, including: obtaining an initial model to be trained and a sample training data set, wherein the initial model to be trained includes an initial spatiotemporal encoder, a feature extraction encoder to be trained and a detection evaluation sub-model to be trained, and the sample training data set includes electromagnetic echo signal samples, latent atrial fibrillation disease period information samples, atrial beat information samples and atrial fibrillation information samples; based on predetermined sample encoding period information, splitting the electromagnetic echo signal samples into a plurality of electromagnetic echo signal sub-samples; inputting the plurality of electromagnetic echo signal sub-samples into the initial spatiotemporal encoder of the initial model to be trained for first feature extraction processing, and generating a positive correlation with the heart area; A plurality of training mechanical motion spatiotemporal features related to normal heartbeat and abnormal heartbeat; inputting the plurality of mechanical motion spatiotemporal training features into the feature extraction encoder to be trained for a second feature extraction process, generating training mechanical motion time features related to abnormal heartbeat in the heart area; inputting the training mechanical motion time features into the detection and evaluation sub-model to be trained for feature conversion and mapping process, and obtaining training detection heartbeat information; based on the detection and evaluation loss function, obtaining the detection and evaluation loss value according to the training detection heartbeat information, latent atrial fibrillation disease period information samples, atrial beat information samples and atrial fibrillation information samples; according to the detection and evaluation loss value, adjusting the model parameters of the initial model, and obtaining a trained information detection model.

[0011] According to an embodiment of the present invention, the initial space-time encoder is pre-trained, and the pre-training method includes the following operations: obtaining a basic space-time encoder to be trained; inputting multiple electromagnetic echo signal subsamples into the basic space-time encoder to be trained for first feature extraction processing to generate multiple basic training mechanical motion space-time features related to normal heartbeat and abnormal heartbeat in the heart area; based on a preset latent atrial fibrillation period comparison weight function, according to the latent atrial fibrillation disease period information sample, obtaining the latent atrial fibrillation period comparison weight; based on a preset space-time loss function, performing regression comparison learning processing on the latent atrial fibrillation period comparison weight and multiple basic training mechanical motion space-time features to obtain a space-time loss value; according to the space-time loss value, adjusting the latent atrial fibrillation period comparison weight of the basic space-time encoder to be trained to obtain a trained initial space-time encoder.

[0012] According to an embodiment of the present invention, training detection heartbeat information includes training atrial beating information and training atrial fibrillation information; based on the detection evaluation loss function, according to the training detection heartbeat information, the latent atrial fibrillation disease period information sample, the atrial beating information sample and the atrial fibrillation information sample, obtaining the detection evaluation loss value includes multiple iteration rounds, wherein the operation of any t-th iteration round includes: in the t-th iteration round, according to the training atrial beating information and the atrial beating information sample of the t-th iteration round, obtaining the cross entropy loss value corresponding to the t-th iteration round, wherein t≥1, t is a positive integer; according to the multiple training atrial fibrillation information and atrial fibrillation information samples of the t-th iteration round, obtaining the mean square error loss value corresponding to the t-th iteration round; according to the average cross entropy loss corresponding to the t-th iteration round and the average cross entropy loss corresponding to the t-1 iteration round, determining the atrial beating loss change rate corresponding to the t-th iteration round; according to the mean square error loss corresponding to the t-th iteration round The method comprises the following steps: determining the atrial fibrillation loss change rate corresponding to the t-th iteration round by adding the average value and the mean square error loss average value corresponding to the t-1 iteration rounds; obtaining the training loss change rate corresponding to the t-th iteration round according to the atrial beat loss change rate corresponding to the t-th iteration round and the atrial fibrillation loss change rate corresponding to the t-th iteration round; obtaining the atrial beat loss weighting factor and the atrial fibrillation loss weighting factor of the t-th iteration round according to the training loss change rate corresponding to the t-th iteration round, the atrial beat loss change rate corresponding to the t-th iteration round and the atrial fibrillation loss change rate corresponding to the t-th iteration round based on a preset beat loss change rate function and a preset fibrillation loss change rate function; obtaining the detection evaluation loss value of the t-th iteration round according to the cross entropy loss value corresponding to the t-th iteration round, the mean square error loss value corresponding to the t-th iteration round, the atrial beat loss weighting factor of the t-th iteration round and the atrial fibrillation loss weighting factor of the t-th iteration round based on the detection evaluation function.

[0013] The second aspect of the present invention provides a device for assisting the prediction of latent atrial fibrillation, characterized in that it includes: an acquisition module for acquiring an initial electromagnetic echo signal, wherein the initial electromagnetic echo signal represents a plurality of signals that are spatially overlapped and reflected by a plurality of body regions of a subject within a predetermined time period; a separation module for performing signal separation processing on the initial electromagnetic echo signal to obtain a target electromagnetic echo signal corresponding to the motion state of the heart region of the subject within the predetermined time period; a first extraction module for performing a first feature extraction processing on the target electromagnetic echo signal using a spatiotemporal encoder of an information detection model to generate a plurality of mechanical motion spatiotemporal features related to normal heartbeats and abnormal heartbeats of the heart region; a second extraction module for performing a second feature extraction processing on the plurality of mechanical motion spatiotemporal features using a feature extraction encoder of the information detection model to generate a target mechanical motion time feature related to an abnormal heartbeat of the heart region; and an acquisition module for performing feature conversion mapping processing on the target mechanical motion time feature using a detection evaluation submodel of the information detection model to obtain target auxiliary prediction heartbeat information of the subject.

[0014] A third aspect of the present invention provides an electronic device, comprising: one or more processors; a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the above method.

[0015] The fourth aspect of the present invention also provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, causes the processor to execute the above method.

[0016] The fifth aspect of the present invention also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0017] According to the method, device and electronic device for assisting the prediction of latent atrial fibrillation of the present invention, an initial electromagnetic echo signal is acquired, and signal separation processing is performed on the initial electromagnetic echo signal to obtain a target electromagnetic echo signal corresponding to the motion state of the heart region of the object under test within a predetermined time period, and a first feature extraction processing is performed on the target electromagnetic echo signal using a spatiotemporal encoder of an information detection model to generate a plurality of mechanical motion spatiotemporal features related to normal heartbeat and abnormal heartbeat of the heart region, and a second feature extraction processing is performed on the plurality of mechanical motion spatiotemporal features using a feature extraction encoder of the information detection model to generate a target mechanical motion time feature related to the abnormal heartbeat of the heart region, and a detection evaluation submodel of the information detection model is used to perform feature conversion mapping processing on the target mechanical motion time feature to obtain the target auxiliary prediction heartbeat information of the object under test, thereby realizing the characteristics that latent atrial fibrillation can cause changes in the heart structure and that mechanical activity abnormalities still exist even after sinus rhythm is restored, and using pre-trained The spatiotemporal encoder of the trained information detection model (latent atrial fibrillation detection model) extracts all mechanical activities (abnormal activities and normal activities) related to the heart area from the two dimensions of space and time, and then uses the feature extraction encoder to further separate and extract all the features containing the two dimensions of space and time, so as to extract the time characteristics of the abnormal heartbeat, that is, the heartbeat related to latent atrial fibrillation, and predict the object under test based on the time characteristics of the heartbeat related to latent atrial fibrillation, so as to obtain target-assisted detection heartbeat information with high accuracy and reliability. At the same time, when the feature extraction encoder is used to separate and extract the time characteristics of the heartbeat related to latent atrial fibrillation, the characteristics of the mechanical activities based on persistent atrial fibrillation and the time characteristics of the heartbeat related to persistent atrial fibrillation can also be separated and extracted, so as to assist relevant professionals in subsequent processing, without the need for repeated detection of the object under test, thereby improving processing efficiency and saving costs.

[0018] According to the embodiments of the present invention, further, since the information detection model is constructed based on the spatiotemporal encoder, the feature extraction encoder and the detection evaluation sub-model, a large amount of model learning training is performed on the constructed information detection model in advance so that the information detection model can accurately obtain the target-assisted predicted heartbeat information of the object under test, thereby obtaining an information detection model with high robustness, high reliability, high efficiency and high output accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The above contents and other objects, features and advantages of the present invention will become more apparent through the following description of the embodiments of the present invention with reference to the accompanying drawings, in which:

[0020] Figure 1 An application scenario diagram of a method for assisting in predicting latent atrial fibrillation according to an embodiment of the present invention is shown;

[0021] Figure 2 A flow chart of a method for assisting in predicting latent atrial fibrillation according to an embodiment of the present invention is shown;

[0022] Figure 3a A schematic diagram showing a comparison between an electrocardiogram of latent atrial fibrillation and an electrocardiogram of normal sinus rhythm according to an embodiment of the present invention;

[0023] Figure 3b A schematic diagram showing a comparison between a target electromagnetic echo signal of latent atrial fibrillation and a target electromagnetic echo signal of normal sinus rhythm according to an embodiment of the present invention;

[0024] Figure 4 A schematic diagram showing feature visualization of a feature extraction encoder of the present invention during a training process according to an embodiment of the present invention is shown;

[0025] Figure 5a A schematic diagram showing the output result of an untrained spatiotemporal encoder according to an embodiment of the present invention;

[0026] Figure 5b A schematic diagram showing the output result of a spatiotemporal encoder trained by the prior art according to an embodiment of the present invention;

[0027] Figure 5c A schematic diagram showing the output result of a spatiotemporal encoder trained by the method of the present invention according to an embodiment of the present invention;

[0028] Figure 6a A schematic diagram showing the output results of an untrained detection evaluation sub-model according to an embodiment of the present invention;

[0029] Figure 6b A schematic diagram showing the output of the detection evaluation sub-model after training with the prior art according to an embodiment of the present invention;

[0030] Figure 6c A schematic diagram showing the output of the detection evaluation sub-model after training by the method of the present invention according to an embodiment of the present invention;

[0031] Figure 7a A schematic diagram showing the excellence of various prediction indicators obtained after auxiliary detection using the method of the present invention according to an embodiment of the present invention;

[0032] Figure 7b A schematic diagram showing target auxiliary predicted heartbeat information obtained after auxiliary detection using the method of the present invention with increased monitoring time according to an embodiment of the present invention;

[0033] Figure 8A schematic diagram showing a comparison between various indicators of target auxiliary predicted heartbeat information obtained by auxiliary detection using an electrocardiogram of the prior art and a method of the present invention according to an embodiment of the present invention;

[0034] Fig. 9 A schematic diagram showing a process for assisting in predicting latent atrial fibrillation according to an embodiment of the present invention;

[0035] Fig.10 A structural block diagram of a device for assisting in predicting latent atrial fibrillation according to an embodiment of the present invention is shown;

[0036] Fig.11 A block diagram of an electronic device for assisting in predicting latent atrial fibrillation according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0037] Below, embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present invention. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of embodiments of the present invention. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of concepts of the present invention.

[0038] The terms used herein are only for describing specific embodiments and are not intended to limit the present invention. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0039] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0040] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0041] In the technical solution of the present invention, the user information (including but not limited to user personal information, user image information, user device information, such as location information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0042] Atrial fibrillation is a common type of arrhythmia and also a common type of cardiovascular disease. It increases the risk of stroke or other serious problems. Electrocardiograms can sense the electrical activity of the heart through electrodes attached to the body, thereby detecting rhythm abnormalities caused by atrial fibrillation attacks. However, due to the paroxysmal and hidden nature of atrial fibrillation, the electrocardiogram detection effect of latent atrial fibrillation in atrial fibrillation is poor, resulting in the failure of relevant personnel to accurately diagnose the target object based on the obtained electrocardiogram, delaying subsequent treatment. Accurate electrocardiogram monitoring model results can be obtained by increasing the frequency and time of electrocardiogram examinations to help relevant personnel make judgments, but multiple-frequency detection leads to poor efficiency and cannot be applied on a large scale.

[0043] On this basis, according to the characteristics of latent atrial fibrillation, the existing technology captures the tiny signal changes that appear in the electrocardiogram signal, so as to assist relevant personnel in diagnosing the abnormal heartbeat information of the captured target object. However, the above-mentioned existing technology still cannot accurately obtain relevant auxiliary information, resulting in the inability to judge the heartbeat condition of the target object, delaying subsequent processing, and cannot be widely used in actual operations. In the process of research and development, the researchers found that in the relevant technology, due to the paroxysmal and hidden characteristics of latent atrial fibrillation, it is impossible to intuitively obtain accurate auxiliary images containing latent atrial fibrillation information, thereby failing to timely assist relevant personnel in predicting and judging, and, while obtaining the latent atrial fibrillation information, it is necessary to use contact equipment to perform multiple detections on the target, resulting in technical problems of low processing efficiency and poor flexibility.

[0044] In view of this, an embodiment of the present invention provides a method for assisting the prediction of latent atrial fibrillation, comprising: acquiring an initial electromagnetic echo signal, wherein the initial electromagnetic echo signal represents a plurality of signals that are spatially overlapped with each other and reflected by a plurality of body regions of a subject under test within a predetermined time period; performing signal separation processing on the initial electromagnetic echo signal to obtain a target electromagnetic echo signal corresponding to the motion state of a heart region of the subject under test within a predetermined time period; performing a first feature extraction processing on the target electromagnetic echo signal using a spatiotemporal encoder of an information detection model to generate a plurality of mechanical motion spatiotemporal features related to a normal heartbeat and an abnormal heartbeat of the heart region; performing a second feature extraction processing on the plurality of mechanical motion spatiotemporal features using a feature extraction encoder of the information detection model to generate a target mechanical motion time feature related to an abnormal heartbeat of the heart region; performing feature conversion mapping processing on the target mechanical motion time feature using a detection evaluation submodel of the information detection model to obtain target auxiliary prediction heartbeat information of the subject under test.

[0045] Figure 1 A diagram showing an application scenario of a method for assisting in predicting latent atrial fibrillation according to an embodiment of the present invention is shown.

[0046] like Figure 1 As shown, the application scenario according to this embodiment may include a first radar device 101, a target object 102, and a receiver 103. The first radar device 101 is used to send an electromagnetic wave signal to the target object 102.

[0047] The user may use the first radar device 101 to interact with the target object 102 and the receiver 103 to receive or send signals, etc.

[0048] The receiver 103 may be a receiver for receiving various electromagnetic echo signals, for example, receiving and processing the electromagnetic echo signal sent by the first radar device 101 (for example only). The receiver 103 may analyze and process the received electromagnetic echo signal and other data, and feed back the processing result (for example, a web page, information, or data obtained or generated according to a user request) to the terminal device.

[0049] It should be noted that the method for assisting the prediction of latent atrial fibrillation provided in the embodiment of the present invention can generally be performed by the receiver 103. Accordingly, the device for assisting the prediction of latent atrial fibrillation provided in the embodiment of the present invention can generally be arranged in the receiver 103. The method for assisting the prediction of latent atrial fibrillation provided in the embodiment of the present invention can also be performed by a receiver or a receiver cluster that is different from the receiver 103 and can communicate with the first radar device 101 and / or the receiver 103. Accordingly, the device for assisting the prediction of latent atrial fibrillation provided in the embodiment of the present invention can also be arranged in a receiver or a receiver cluster that is different from the receiver 103 and can communicate with the first radar device 101 and / or the receiver 103.

[0050] It should be understood that Figure 1 The number of the first radar device, the target object and the receiver in the embodiment is only for illustration. According to the implementation requirements, there may be any number of radar devices, observation objects and receivers.

[0051] The following will be based on Figure 1 The scene described by Figure 2~Figure 9 The method for assisting in predicting latent atrial fibrillation according to an embodiment of the present invention is described in detail.

[0052] Figure 2 A flow chart of a method for assisting in predicting latent atrial fibrillation according to an embodiment of the present invention is shown.

[0053] like Figure 2 As shown, the method for assisting prediction of latent atrial fibrillation in this embodiment includes operations S210 to S250.

[0054] In operation S210, an initial electromagnetic echo signal is acquired.

[0055] According to an embodiment of the present invention, the initial electromagnetic echo signal represents a plurality of signals spatially overlapped with each other reflected by a plurality of body regions of the measured object within a predetermined period of time.

[0056] According to an embodiment of the present invention, a millimeter-wave radar can be used to transmit electromagnetic signals to the object under test. Since the millimeter-wave radar can be located directly above the object under test, multiple body regions of the object under test can receive the electromagnetic signal and generate electromagnetic echo signals. When the multiple electromagnetic echo signals are transmitted in the channel, they are spatially overlapped with each other to obtain an initial electromagnetic echo signal. The millimeter-wave radar can have multiple transmitting antennas and multiple receiving antennas.

[0057] In operation S220, the initial electromagnetic echo signal is subjected to signal separation processing to obtain a target electromagnetic echo signal corresponding to the motion state of the heart region of the measured object within a predetermined period of time.

[0058] According to an embodiment of the present invention, since the initial electromagnetic echo signal includes electromagnetic echo signals of non-cardiac areas, it is necessary to separate multiple spatially overlapped electromagnetic echo signals in order to extract the initial electromagnetic echo signal corresponding only to the heart area, and then process the initial electromagnetic echo signal corresponding to the heart area to obtain a target electromagnetic echo signal corresponding to the motion state of the heart area.

[0059] In operation S230, the target electromagnetic echo signal is subjected to a first feature extraction process using the spatiotemporal encoder of the information detection model to generate a plurality of mechanical motion spatiotemporal features related to normal heartbeats and abnormal heartbeats in the heart region. According to an embodiment of the present invention, the information detection model can be characterized as a latent atrial fibrillation detection model, and the spatiotemporal encoder of the information detection model can be an encoder built on the R(2+1)D architecture, wherein the R(2+1)D architecture includes a 2D spatial convolutional network and a 1D temporal convolutional network. By using the spatiotemporal encoder built on the R(2+1)D architecture to perform feature extraction processing on the target electromagnetic echo signal, it is possible to obtain a plurality of mechanical motion spatiotemporal features while reducing the computational complexity and increasing the number of nonlinear layers, so as to facilitate the model to learn a more complex spatiotemporal feature representation, wherein the generated mechanical motion spatiotemporal features can be 128-dimensional feature vectors.

[0060] According to an embodiment of the present invention, the abnormal heartbeat may include the beat information of latent atrial fibrillation and the beat information of persistent atrial fibrillation.

[0061] In operation S240, a feature extraction encoder of the information detection model is used to perform a second feature extraction process on the plurality of mechanical motion spatiotemporal features to generate a target mechanical motion time feature related to an abnormal heartbeat in the heart region.

[0062] According to an embodiment of the present invention, the feature extraction encoder of the information detection model can be constructed by an encoder of a neural network (Transformer) architecture based on a self-attention mechanism, and multiple mechanical motion spatiotemporal features are subjected to feature extraction processing of time changes related to heartbeat information by the feature extraction encoder of the information detection model to generate target mechanical motion time features related to abnormal heartbeats in the heart area, wherein the target mechanical motion time features can be a feature vector of the number of mechanical motion spatiotemporal features × the dimension of the mechanical motion spatiotemporal features. For example, there are a total of 5 mechanical motion spatiotemporal features, each of which is a 128-dimensional feature vector, and the feature extraction encoder of the information detection model outputs a 5×128 feature vector matrix of the target mechanical motion time features.

[0063] According to an embodiment of the present invention, the relevant characteristics of the temporal changes in the beating information of latent atrial fibrillation can be characterized as the weak peristalsis of the mechanical movement of the heart in a short period of time, that is, the signal change amplitude is very small in a short period of time, and the relevant characteristics of the temporal changes in the beating information of persistent atrial fibrillation can be characterized as the weak peristalsis of the mechanical movement of the heart in a longer period of time, that is, the signal change amplitude is very small over a long period of time.

[0064] In operation S250, the detection evaluation sub-model of the information detection model is used to perform feature conversion mapping processing on the target mechanical motion time characteristics to obtain the target auxiliary predicted heartbeat information of the object under test.

[0065] According to an embodiment of the present invention, the target-assisted prediction information may include whether the heartbeat of the measured object belongs to the first predetermined category. For example, in the case of belonging to the first predetermined category, the obtained target-assisted prediction heartbeat information may include a prediction that the measured object may carry latent atrial fibrillation, and in the case of not belonging to the first predetermined category, the obtained target-assisted prediction heartbeat information may include a prediction that the measured object may carry persistent atrial fibrillation or that the measured object does not carry atrial fibrillation. Further, in the case where the target-assisted prediction heartbeat information is a prediction that the measured object may carry latent atrial fibrillation, the target-assisted prediction heartbeat information may also include a prediction of whether the latent atrial fibrillation carried by the measured object belongs to any of a plurality of degree levels. For example, in the case of the first degree level, the target-assisted predicted heartbeat information obtained may include predicting that the degree of latent atrial fibrillation that the measured object may carry is relatively mild, in the case of the second degree level, the target-assisted predicted heartbeat information obtained may include predicting that the degree of latent atrial fibrillation that the measured object may carry is average, and in the case of the third degree level, the target-assisted predicted heartbeat information obtained may include predicting that the degree of latent atrial fibrillation that the measured object may carry is relatively serious. Here, it should be noted that the target-assisted predicted heartbeat information obtained here is only used as reference information for medical staff to make a diagnosis, and is not used as a direct diagnosis result.

[0066] For example, an electromagnetic signal is transmitted to the body area of ​​the object to be measured, and the receiving antenna receives multiple electromagnetic echo signals that are spatially overlapped with each other and transmitted in the channel, that is, the initial electromagnetic echo signal. The initial electromagnetic echo signal is separated and processed to extract the target electromagnetic echo signal that corresponds only to the heartbeat movement state of the heart area of ​​the object to be measured. Then, the target electromagnetic echo signal is processed by using the spatiotemporal encoder, feature extraction encoder and detection evaluation sub-model in the information detection model to perform feature extraction, feature conversion mapping and other processing on the target electromagnetic echo signal to obtain whether the heartbeat of the object to be measured belongs to the first predetermined category and predict the degree of latent atrial fibrillation carried by the object to be measured, that is, the target assisted prediction of the heartbeat information, so that scientists or professionals in related fields can identify the heartbeat of the object to be measured based on the target assisted prediction of the heartbeat information, and make a judgment suitable for the object to be measured.

[0067] According to an embodiment of the present invention, an initial electromagnetic echo signal is acquired, and signal separation processing is performed on the initial electromagnetic echo signal to obtain a target electromagnetic echo signal corresponding to the motion state of the heart region of the object under test within a predetermined time period, and a first feature extraction processing is performed on the target electromagnetic echo signal using a spatiotemporal encoder of an information detection model to generate a plurality of mechanical motion spatiotemporal features related to normal heartbeats and abnormal heartbeats in the heart region, and a second feature extraction processing is performed on the plurality of mechanical motion spatiotemporal features using a feature extraction encoder of the information detection model to generate a target mechanical motion time feature related to an abnormal heartbeat in the heart region, and a detection evaluation submodel of the information detection model is used to perform feature conversion mapping processing on the target mechanical motion time feature to obtain the target auxiliary predicted heartbeat information of the object under test, thereby realizing the characteristics that latent atrial fibrillation can cause changes in the heart structure and that mechanical activity abnormalities still exist even after sinus rhythm is restored, and using a pre-trained information detection model ( The spatiotemporal encoder of the latent atrial fibrillation detection model extracts all mechanical activities (abnormal activities and normal activities) related to the heart area from the two dimensions of space and time, and then uses the feature extraction encoder to further separate and extract all the features containing the two dimensions of space and time, so as to extract the time characteristics of the abnormal heartbeat, that is, the heartbeat related to latent atrial fibrillation, and predicts the object under test based on the time characteristics of the heartbeat related to latent atrial fibrillation, so as to obtain the target auxiliary detection heartbeat information with high accuracy and high reliability. At the same time, when the time characteristics of the heartbeat related to latent atrial fibrillation are separated and extracted by the feature extraction encoder, the characteristics of the mechanical activities based on persistent atrial fibrillation and the time characteristics of the heartbeat related to persistent atrial fibrillation can also be separated and extracted, so as to assist relevant professionals in subsequent processing, without the need for repeated detection of the object under test, thereby improving processing efficiency and saving costs.

[0068] According to the embodiments of the present invention, further, since the information detection model is constructed based on the spatiotemporal encoder, the feature extraction encoder and the detection evaluation sub-model, a large amount of model learning and training is performed on the constructed information detection model in advance so that the information detection model can accurately obtain the target-assisted detection heartbeat information of the object under test, thereby obtaining an information detection model with high robustness, high reliability, high efficiency and high output accuracy.

[0069] According to the embodiments of the present invention, further, the present invention utilizes contactless radar equipment to perform auxiliary detection on the object to be detected, which can improve the detection experience of the object to be detected and the application scenarios and scope of auxiliary detection of latent atrial fibrillation of the object to be detected.

[0070] It should be noted that the target-assisted detection heartbeat information obtained by the present invention is only an intermediate result, and the diagnosis result or health status cannot be directly derived from the target-assisted detection heartbeat information obtained by the method of the present invention.

[0071] According to an embodiment of the present invention, a method for performing signal separation processing on an initial electromagnetic echo signal to obtain a target electromagnetic echo signal corresponding to a motion state of a heart region of a measured object within a predetermined period of time includes the following operations.

[0072] According to an embodiment of the present invention, beamforming processing is performed on the initial electromagnetic echo signal to obtain a plurality of intermediate electromagnetic echo signals.

[0073] According to an embodiment of the present invention, the intermediate electromagnetic echo signal represents a plurality of four-dimensional signals spatially separated from each other reflected by a plurality of regions within a predetermined period of time.

[0074] According to an embodiment of the present invention, the initial electromagnetic echo signal may be a four-dimensional signal. The initial electromagnetic echo signal may be expressed according to formula (1). Formula (1) is as follows.

[0075] (1);

[0076] Among them, b(t) can be represented as the initial electromagnetic echo signal, h can be represented as the energy attenuation of the electromagnetic echo signal in the propagation path, j can be represented as an imaginary unit, and d tx It can be characterized as the propagation distance of the electromagnetic signal from the transmitting antenna to the object under test, d rx It can be characterized as the distance from the electromagnetic echo signal reflected from the object to the receiving antenna, d e (t) can be represented as the motion change information of objects in the environment, which includes heartbeat, breathing and other motion information. Can be characterized as the wavelength of an electromagnetic signal.

[0077] According to an embodiment of the present invention, based on the spatial resolution provided by the antenna array of the radar device and the large bandwidth of the radar, the initial electromagnetic echo signal is subjected to beamforming processing, so that the electromagnetic echo signals reflected from different body regions to the receiving antenna, that is, the initial electromagnetic echo signals, can be separated to obtain multiple intermediate electromagnetic echo signals, so as to facilitate the extraction of the intermediate electromagnetic echo signals corresponding to the heart region.

[0078] According to an embodiment of the present invention, a three-dimensional electromagnetic echo signal corresponding to the heart region of the measured object is extracted from a plurality of intermediate electromagnetic echo signals.

[0079] According to an embodiment of the present invention, the distance information of each intermediate electromagnetic echo signal can be determined in combination with the distance information between the object under test and the radar receiving antenna array to distinguish and determine the three-dimensional electromagnetic echo signal corresponding to the heart area.

[0080] According to an embodiment of the present invention, the intermediate electromagnetic echo signal after beamforming processing includes information in four dimensions, namely, distance dimension information, azimuth dimension information, elevation dimension information and Doppler frequency shift dimension information. Through the distance information between the object under test and the radar receiving antenna array, the three-dimensional electromagnetic echo signal corresponding to the heart area can be determined, and the three-dimensional electromagnetic echo signal can be characterized as a signal including information in three dimensions, namely, azimuth dimension information, elevation dimension information and Doppler frequency shift dimension information.

[0081] According to an embodiment of the present invention, a second-order differential time-domain filtering algorithm is used to filter the three-dimensional electromagnetic echo signal to obtain a target electromagnetic echo signal.

[0082] According to an embodiment of the present invention, the three-dimensional electromagnetic echo signal corresponding to the heart region may include heartbeat motion features, respiratory motion features and other motion features. Based on the different accelerations between different motion features, the motion features other than the heartbeat motion features with larger accelerations may be eliminated using a second-order differential time domain filtering algorithm, thereby obtaining a target electromagnetic echo signal that only includes the heartbeat motion state corresponding to the heart region of the measured object.

[0083] According to an embodiment of the present invention, the second-order differential signal used in the second-order differential time-domain filtering algorithm can be expressed according to formula (2), and formula (2) is as follows.

[0084] (2);

[0085] Among them, s ’’s(l,t) can be characterized as the second-order difference signal of the tth frame at position l, △t can be characterized as the time interval between two adjacent frames of three-dimensional electromagnetic echo signals, s(l,t) can be characterized as the tth frame three-dimensional electromagnetic echo signal after beamforming processing at position l, s(l,t+1) can be characterized as the t+1th frame three-dimensional electromagnetic echo signal after beamforming processing at position l, and s(l,t-1) can be characterized as the t-1th frame three-dimensional electromagnetic echo signal after beamforming processing at position l. wave signal, s(l,t+2) can be characterized as the t+2th frame three-dimensional electromagnetic echo signal after beamforming processing at position l, s(l,t-2) can be characterized as the t-2th frame three-dimensional electromagnetic echo signal after beamforming processing at position l, s(l,t+3) can be characterized as the t+3th frame three-dimensional electromagnetic echo signal after beamforming processing at position l, s(l,t-3) can be characterized as the t-3th frame three-dimensional electromagnetic echo signal after beamforming processing at position l,

[0086] According to an embodiment of the present invention, beamforming is performed on an initial electromagnetic echo signal to obtain a plurality of intermediate electromagnetic echo signals, and a three-dimensional electromagnetic echo signal corresponding to the heart region of the object under test is extracted from the plurality of intermediate electromagnetic echo signals. The three-dimensional electromagnetic echo signal is filtered using a second-order differential time-domain filtering algorithm to obtain a target electromagnetic echo signal, thereby achieving accurate separation and extraction of the three-dimensional electromagnetic echo signal corresponding only to the heart region from the spatially aliased initial electromagnetic echo signal, and eliminating features other than heartbeat features from the three-dimensional electromagnetic echo signal to obtain a heartbeat signal corresponding only to the heart region, i.e., the target electromagnetic echo signal, so as to facilitate analysis and processing of the target electromagnetic echo signal to obtain target-assisted predicted heartbeat information of the object under test.

[0087] According to an embodiment of the present invention, a method for extracting a three-dimensional electromagnetic echo signal corresponding to a heart region of a measured object from a plurality of intermediate electromagnetic echo signals includes the following operations.

[0088] According to an embodiment of the present invention, a plurality of intermediate electromagnetic echo signals are superimposed to obtain an intermediate electromagnetic echo superimposed signal.

[0089] According to an embodiment of the present invention, a constant false alarm rate algorithm is used to process the intermediate electromagnetic echo superposition signal to obtain a distance electromagnetic echo signal corresponding to the chest area of ​​the measured object.

[0090] According to an embodiment of the present invention, based on the fact that the chest region has a large receiving area and good reflection characteristics, the intensity of the reflected electromagnetic echo signal is relatively large. According to the distance information between the object under test and the radar receiving antenna array, the CFAR algorithm (Constant False Alarm Rate) is used to determine the distance unit information of each intermediate electromagnetic echo signal in the intermediate electromagnetic echo superposition signal, thereby determining multiple distance electromagnetic echo signals corresponding to the chest region of the object under test, wherein the distance unit information of the distance electromagnetic echo signal may be consistent with the distance information between the object under test and the radar receiving antenna array.

[0091] According to an embodiment of the present invention, the range electromagnetic echo signal is a three-dimensional signal, that is, a signal including information in three dimensions: azimuth dimension information, elevation dimension information, and Doppler frequency shift dimension information.

[0092] According to an embodiment of the present invention, a density clustering algorithm is used to perform clustering processing on the range electromagnetic echo signal to obtain a three-dimensional electromagnetic echo signal corresponding to the heart area of ​​the measured object.

[0093] According to an embodiment of the present invention, since there may be multiple distance unit information consistent with the distance information between the object to be measured and the radar receiving antenna array, it is possible to cluster the multiple distance electromagnetic echo signals using the DBSCAN algorithm (Density-Based Spatial Clustering of Applications with Noise) based on the angle information between the object to be measured and the radar receiving antenna array and based on the consistency of the amplitude of the chest movement, to obtain a three-dimensional electromagnetic echo signal corresponding to the heart area of ​​the object to be measured.

[0094] According to an embodiment of the present invention, a plurality of intermediate electromagnetic echo signals are superimposed to obtain an intermediate electromagnetic echo superposition signal, and the intermediate electromagnetic echo superposition signal is processed using a constant false alarm rate algorithm to obtain a distance electromagnetic echo signal corresponding to the chest region of the object under test, wherein the distance electromagnetic echo signal is a three-dimensional signal; the distance electromagnetic echo signal is clustered using a density clustering algorithm to obtain a three-dimensional electromagnetic echo signal corresponding to the heart region of the object under test, thereby achieving accurate determination of a three-dimensional electromagnetic echo signal corresponding to the heart region based on dimensional feature information such as distance and angle, combined with the chest region and reflection characteristics and motion characteristics.

[0095] According to an embodiment of the present invention, a method for performing a first feature extraction process on a target electromagnetic echo signal using a spatiotemporal encoder of an information detection model to generate multiple spatiotemporal features of mechanical motion related to normal heartbeats and abnormal heartbeats in a heart area includes the following operations.

[0096] According to an embodiment of the present invention, based on a predetermined coding period, the target electromagnetic echo signal is split into target electromagnetic echo sub-signals of a plurality of time periods.

[0097] For example, the initial electromagnetic echo signal initially acquired is 10s, then the target electromagnetic echo signal after processing is also 10s, and the predetermined coding period is 2s. Based on the predetermined coding period, the 10s target electromagnetic echo signal is split into target electromagnetic echo sub-signals of 0-2s, target electromagnetic echo sub-signals of 2-4s, target electromagnetic echo sub-signals of 4-6s, target electromagnetic echo sub-signals of 6-8s, and target electromagnetic echo sub-signals of 8-10s, so that the space-time encoder of the information detection model can perform feature extraction processing on the target electromagnetic echo sub-signals.

[0098] According to an embodiment of the present invention, target electromagnetic echo sub-signals of multiple time periods are respectively input into the spatiotemporal encoder of the information detection model for first feature extraction processing to generate multiple mechanical motion spatiotemporal features related to normal heartbeat and abnormal heartbeat in the heart area.

[0099] According to an embodiment of the present invention, a plurality of mechanical motion spatiotemporal features correspond to target electromagnetic echo sub-signals in a plurality of time periods.

[0100] According to an embodiment of the present invention, the spatiotemporal characteristics of mechanical motion may be characterized as accurate motion characteristics of the heart of the measured object in the time dimension and the space dimension, wherein the spatiotemporal characteristics of mechanical motion may be a 128-dimensional feature vector.

[0101] According to an embodiment of the present invention, the target electromagnetic echo signal is split into target electromagnetic echo sub-signals of multiple time periods based on a predetermined coding time period, so that the space-time encoder of the information detection model can perform feature extraction processing on the target electromagnetic echo sub-signals, thereby improving the speed and efficiency of feature extraction of the space-time encoder, and then the target electromagnetic echo sub-signals of the multiple time periods are respectively input into the space-time encoder of the information detection model for a first feature extraction process, thereby generating multiple mechanical motion space-time features related to normal heartbeats and abnormal heartbeats in the heart area, thereby realizing the conversion of electromagnetic echo signals into accurate motion features of the heart in the time and space dimensions through the information detection model, so as to facilitate analysis and processing based on real and accurate motion features.

[0102] According to an embodiment of the present invention, the detection evaluation sub-model includes a first double-layer fully connected network and a second double-layer fully connected network, and the first double-layer fully connected network and the second double-layer fully connected network both include two linear layers.

[0103] According to an embodiment of the present invention, a method for obtaining target-assisted predicted heartbeat information of a detected object by performing feature conversion mapping processing on target mechanical motion time features using a detection evaluation sub-model of an information detection model includes the following operations.

[0104] According to an embodiment of the present invention, the target mechanical motion time characteristics are respectively input into the first linear layer of the first double-layer fully connected network and the first linear layer of the second double-layer fully connected network for dimensionality reduction processing to obtain the first reduced dimensionality mechanical motion state characteristics and the second reduced dimensionality mechanical motion time characteristics.

[0105] According to an embodiment of the present invention, the first reduced dimensionality mechanical motion state feature and the second reduced dimensionality mechanical motion time feature after dimensionality reduction processing are both 10-dimensional feature vectors, wherein the first reduced dimensionality mechanical motion state feature may be a feature related to the mechanical motion of the heart when an abnormal heartbeat occurs, and the second reduced dimensionality mechanical motion time feature may be a feature related to the mechanical motion change trend of the heart during the period when an abnormal heartbeat occurs.

[0106] According to an embodiment of the present invention, the first dimension-reduced mechanical motion state feature is input into the second linear layer of the first two-layer fully connected network for feature conversion mapping processing to obtain the auxiliary atrial beating information of the measured object.

[0107] According to an embodiment of the present invention, the second linear layer of the first two-layer fully connected network performs feature conversion mapping processing on the input first dimension reduction mechanical motion state feature to generate the target heart beating state feature of the object under test, wherein the target heart beating state feature is a 2-dimensional feature vector, and the target heart beating state feature can characterize whether the target electromagnetic echo signal of the object under test contains the target mechanical motion time feature related to the abnormal heartbeat, thereby obtaining the auxiliary atrial beating information of the object under test based on the target heart beating state feature. For example, in the case where the target electromagnetic echo signal of the object under test contains the target mechanical motion time feature related to the abnormal heartbeat, the auxiliary atrial beating information of the object under test may belong to the first predetermined category, that is, the object under test may carry auxiliary information of latent atrial fibrillation. On this basis, it should be noted that the auxiliary atrial beating information is only used as an intermediate result, and the diagnosis result or health status cannot be directly obtained from the target auxiliary detection heartbeat information obtained according to the method of the present invention.

[0108] According to an embodiment of the present invention, the second dimension-reduced mechanical motion time feature is input into the second linear layer of the second two-layer fully connected network for feature conversion mapping processing to obtain auxiliary atrial fibrillation information of the measured object.

[0109] According to an embodiment of the present invention, the second linear layer of the second double-layer fully connected network performs feature conversion mapping processing on the input second dimension reduction mechanical motion time feature to generate a target heart beat change trend feature of the measured object, wherein the target heart beat change trend feature is a 1-dimensional feature vector, and the target heart beat change trend feature can characterize the change trend of the signal related to the abnormal heartbeat in the target electromagnetic echo signal of the measured object, thereby obtaining the auxiliary atrial fibrillation information of the measured object based on the target heart beat change trend feature. For example, in the case where the target heart beat change trend feature of the measured object is that the mechanical movement of the heart in a short period of time is weak peristalsis, the auxiliary atrial fibrillation information of the measured object can belong to the first degree level, that is, the degree of latent atrial fibrillation that the measured object may carry is relatively light; in the case where the target heart beat change trend feature of the measured object is that the mechanical movement of the heart in a long period of time is weak peristalsis, the auxiliary atrial fibrillation information of the measured object can belong to the third degree level, that is, the degree of latent atrial fibrillation that the measured object may carry is relatively serious. On this basis, it should be noted that the auxiliary atrial fibrillation information is only an intermediate result, and the diagnosis result or health status cannot be directly derived from the target auxiliary predicted heartbeat information obtained according to the method of the present invention.

[0110] According to an embodiment of the present invention, the target mechanical motion time feature is respectively input into the first linear layer of the first double-layer fully connected network and the first linear layer of the second double-layer fully connected network for dimensionality reduction processing to obtain the first reduced dimensionality mechanical motion state feature and the second reduced dimensionality mechanical motion time feature, the first reduced dimensionality mechanical motion state feature is input into the second linear layer of the first double-layer fully connected network for feature conversion mapping processing to obtain the auxiliary atrial beating information of the object under test, and the second reduced dimensionality mechanical motion time feature is input into the second linear layer of the second double-layer fully connected network for feature conversion mapping processing to obtain the auxiliary atrial fibrillation information of the object under test, thereby achieving the dimensionality reduction effect of the target mechanical motion time feature and extracting different features to output different auxiliary information through the first linear layer of the two double-layer fully connected networks, and then performing secondary dimensionality reduction through the second linear layer, and finally outputting multiple target auxiliary information with high accuracy related to the heart in parallel, thereby improving the model processing efficiency.

[0111] According to an embodiment of the present invention, the information detection model is trained in the following manner, and the training method includes the following operations.

[0112] According to an embodiment of the present invention, an initial model to be trained and a sample training data set are obtained.

[0113] According to an embodiment of the present invention, the initial model to be trained includes an initial spatiotemporal encoder, a feature extraction encoder to be trained and a detection evaluation sub-model to be trained, and the sample training data set includes electromagnetic echo signal samples, latent atrial fibrillation disease period information samples, atrial beat information samples and atrial fibrillation information samples.

[0114] According to an embodiment of the present invention, the latent atrial fibrillation period information sample can be characterized as a period sample of the time when the sample subject suffers from latent atrial fibrillation, the atrial beat information sample can be characterized as an information sample of whether the sample subject suffers from latent atrial fibrillation, and the atrial fibrillation information sample can be characterized as an information sample of the degree of latent atrial fibrillation suffered by the sample subject.

[0115] According to an embodiment of the present invention, the electromagnetic echo signal sample is split into a plurality of electromagnetic echo signal sub-samples based on predetermined sample encoding period information.

[0116] According to an embodiment of the present invention, multiple electromagnetic echo signal subsamples are input into the initial spatiotemporal encoder of the initial model to be trained for first feature extraction processing to generate multiple training mechanical motion spatiotemporal features related to normal heartbeats and abnormal heartbeats in the heart area.

[0117] According to an embodiment of the present invention, the spatiotemporal features of the trained mechanical motion may be 128-dimensional feature vectors.

[0118] According to an embodiment of the present invention, a plurality of mechanical motion spatiotemporal training features are input into a feature extraction encoder to be trained for second feature extraction processing to generate training mechanical motion time features related to abnormal heartbeats in the cardiac region.

[0119] According to an embodiment of the present invention, the training mechanical motion temporal feature may be a feature vector of the number of training mechanical motion spatiotemporal features×the dimension of the training mechanical motion spatiotemporal features.

[0120] According to an embodiment of the present invention, the training mechanical motion time feature is input into the detection evaluation sub-model to be trained for feature conversion mapping processing to obtain the training detection heartbeat information.

[0121] According to an embodiment of the present invention, the training detected heartbeat information may include training atrial beat information and training atrial fibrillation information.

[0122] According to an embodiment of the present invention, based on the detection evaluation loss function, a detection evaluation loss value is obtained according to training detection heartbeat information, latent atrial fibrillation disease period information samples, atrial beat information samples and atrial fibrillation information samples.

[0123] According to an embodiment of the present invention, the model parameters of the initial model are adjusted according to the detection evaluation loss value to obtain a trained information detection model.

[0124] According to an embodiment of the present invention, based on the detection evaluation loss value, the weight parameters of the initial spatiotemporal encoder, the feature extraction encoder to be trained, and the detection evaluation sub-model to be trained can be adjusted respectively, so as to obtain an information detection model whose detection evaluation loss value is less than a predetermined evaluation loss threshold, that is, a trained information detection model.

[0125] According to an embodiment of the present invention, in addition to setting a predetermined assessment loss threshold to help determine whether the information detection model reaches the target standard, the training can also be completed by setting the number of training times to help determine whether the information detection model reaches the target standard.

[0126] According to an embodiment of the present invention, by utilizing samples in a sample training data set to train an initial spatiotemporal encoder in an initial model to be trained, a feature extraction encoder to be trained, and a detection evaluation sub-model to be trained, and then combining with information samples in the sample training data set, based on the calculated detection evaluation loss value, it is possible to obtain accurate features and auxiliary result information that can be output based on electromagnetic echo signals, thereby improving the robustness and maturity of the information detection model in the process of learning and training, so as to help relevant professionals improve processing efficiency and the efficiency of auxiliary detection of the objects under test.

[0127] According to an embodiment of the present invention, the initial spatiotemporal encoder is pre-trained, and the pre-training method includes the following operations.

[0128] According to an embodiment of the present invention, a basic spatiotemporal encoder to be trained is obtained.

[0129] According to an embodiment of the present invention, the basic spatiotemporal encoder to be trained may be a basic encoder built on the R(2+1)D architecture. By utilizing the output features of the basic spatiotemporal encoder, features including both time and space dimensions may be output.

[0130] According to an embodiment of the present invention, multiple electromagnetic echo signal subsamples are input into a basic spatiotemporal encoder to be trained for first feature extraction processing to generate multiple basic training mechanical motion spatiotemporal features related to normal heartbeats and abnormal heartbeats in the heart area.

[0131] According to an embodiment of the present invention, the spatiotemporal features of basic training mechanical motion may be a 128-dimensional feature vector.

[0132] According to an embodiment of the present invention, based on a preset latent atrial fibrillation period comparison weight function, a latent atrial fibrillation period comparison weight is obtained according to a latent atrial fibrillation disease period information sample.

[0133] According to an embodiment of the present invention, the contrast weight of the latent atrial fibrillation period can be calculated according to formula (3), which is shown below.

[0134] (3);

[0135] Among them, w i，a It can be represented as the contrast weight of latent atrial fibrillation period, y a It can be represented as the information sample of the ath hidden atrial fibrillation period, y i It can be represented as the information sample of the i-th latent atrial fibrillation period, and eps can be represented as a parameter that is not 0.

[0136] According to an embodiment of the present invention, based on a preset spatiotemporal loss function, regression contrast learning processing is performed on the contrast weights of the latent atrial fibrillation period and the spatiotemporal features of multiple basic training mechanical movements to obtain a spatiotemporal loss value.

[0137] According to an embodiment of the present invention, the spatiotemporal loss value may be calculated according to formula (4), which is shown below.

[0138] (4);

[0139] Among them, L regcon can be represented as a spatiotemporal loss value, I can be represented as a total of I sample indexes, that is, there are a total of I samples of latent atrial fibrillation disease period information, A(i) can be represented as a set of latent atrial fibrillation disease period information samples except the i-th latent atrial fibrillation disease period information sample, z i It can be represented as the basic training mechanical motion spatiotemporal feature after the basic spatiotemporal encoder feature extraction corresponding to the information sample of the i-th hidden atrial fibrillation period, z a It can be represented as the basic training mechanical motion spatiotemporal features corresponding to the information sample of the ath hidden atrial fibrillation period after the basic spatiotemporal encoder feature extraction, z b It can be represented as the basic training mechanical motion spatiotemporal features corresponding to the bth hidden atrial fibrillation period information sample after the basic spatiotemporal encoder feature extraction, It can be represented as the temperature coefficient used in contrast learning, and can be set to 0.07 in specific implementation.

[0140] According to an embodiment of the present invention, based on regression contrast learning technology, under the guidance of multiple latent atrial fibrillation disease period information samples, the basic spatiotemporal encoder to be trained is adjusted according to the spatiotemporal loss value, so that the feature differences between the multiple basic training mechanical motion spatiotemporal features output by the basic spatiotemporal encoder and the multiple latent atrial fibrillation disease period information samples are shortened or driven away.

[0141] According to an embodiment of the present invention, the latent atrial fibrillation period comparison weight of the basic spatiotemporal encoder to be trained is adjusted according to the spatiotemporal loss value to obtain a trained initial spatiotemporal encoder.

[0142] According to an embodiment of the present invention, in the process of training a basic space-time encoder to be trained, the basic space-time encoder can be adjusted once to obtain an initial space-time encoder, and then in the process of learning and training the information detection model to be trained that includes the initial space-time encoder, the initial space-time encoder and other structural models are adjusted once again to obtain a trained information detection model.

[0143] According to an embodiment of the present invention, a basic space-time encoder to be trained is obtained, and electromagnetic echo signal subsamples and latent atrial fibrillation disease period information samples are used to perform encoder learning training on the basic space-time encoder to be trained, so that the basic space-time encoder can output basic training mechanical motion space-time features, and then regression contrast learning is used to perform loss calculation on the basic space-time encoder that can output basic training mechanical motion space-time features, so as to adjust the weight parameters of the basic space-time encoder, thereby enhancing the contrast loss between the features of the latent atrial fibrillation disease period information samples and the basic training mechanical motion space-time features, weakening the contrast loss between the features of the latent atrial fibrillation disease period information samples with a large difference and the basic training mechanical motion space-time features, and improving the feature extraction accuracy of the initial space-time encoder and the robustness of the initial space-time encoder.

[0144] According to an embodiment of the present invention, the training detection heartbeat information includes training atrial beat information and training atrial fibrillation information.

[0145] According to an embodiment of the present invention, based on the detection evaluation loss function, according to the training detection heartbeat information, latent atrial fibrillation disease period information samples, atrial beat information samples and atrial fibrillation information samples, the detection evaluation loss value is obtained, including multiple iteration rounds, wherein the tth iteration round includes the following operations, wherein t≥1, t is a positive integer.

[0146] According to an embodiment of the present invention, in the tth iteration round, a cross entropy loss value corresponding to the tth iteration round is obtained based on the training atrial beat information and atrial beat information samples of the tth iteration round.

[0147] According to an embodiment of the present invention, a mean square error loss value corresponding to the t-th iteration round is obtained based on a plurality of training atrial fibrillation information and atrial fibrillation information samples of the t-th iteration round.

[0148] According to an embodiment of the present invention, the training atrial fibrillation information may be the time of carrying latent atrial fibrillation obtained by training the first double-layer fully connected network, and the atrial fibrillation information sample may be the time when the sample object actually carries latent atrial fibrillation.

[0149] According to an embodiment of the present invention, the atrial beat loss change rate corresponding to the tth iteration round is determined based on the average cross entropy loss corresponding to t iteration rounds and the average cross entropy loss corresponding to t-1 iteration rounds.

[0150] According to an embodiment of the present invention, the atrial beat loss change rate corresponding to the t-th iteration round can be calculated according to formula (5), which is shown as follows.

[0151] (5);

[0152] in, It can be represented as the change rate of atrial beating loss corresponding to the tth iteration round, L 1 (t) can be represented as the average cross entropy loss corresponding to t iterations, L 1 (t-1) can be characterized as the average cross entropy loss corresponding to t-1 iterations.

[0153] According to an embodiment of the present invention, the atrial fibrillation loss change rate corresponding to the tth iteration round is determined based on the mean square error loss average value corresponding to t iteration rounds and the mean square error loss average value corresponding to t-1 iteration rounds.

[0154] According to an embodiment of the present invention, the atrial fibrillation loss change rate corresponding to the t-th iteration round can be calculated according to formula (6), which is shown below.

[0155] (6);

[0156] in, It can be represented as the change rate of atrial fibrillation loss corresponding to the tth iteration round, L 2 (t) can be characterized as the average mean square error loss corresponding to t iterations, L 2 (t-1) can be characterized as the average value of the mean square error loss corresponding to t-1 iterations.

[0157] According to an embodiment of the present invention, the training loss change rate corresponding to the t-th iteration round is obtained based on the atrial beat loss change rate corresponding to the t-th iteration round and the atrial fibrillation loss change rate corresponding to the t-th iteration round.

[0158] According to an embodiment of the present invention, the training loss change rate corresponding to the t-th iteration round is obtained by adding the atrial beat loss change rate corresponding to the t-th iteration round and the atrial fibrillation loss change rate corresponding to the t-th iteration round.

[0159] According to an embodiment of the present invention, based on a preset beating loss change rate function and a preset fibrillation loss change rate function, according to the training loss change rate corresponding to the t-th iteration round, the atrial beating loss change rate corresponding to the t-th iteration round, and the atrial fibrillation loss change rate corresponding to the t-th iteration round, the atrial beating loss weighting factor and the atrial fibrillation loss weighting factor of the t-th iteration round are obtained.

[0160] According to an embodiment of the present invention, the atrial beat loss weighting factor of the t-th iteration round can be calculated according to formula (7), which is shown as follows.

[0161] (7);

[0162] Among them, w 1 (t) can be represented as the atrial beat loss weighting factor of the t-th iteration round, N can be represented as a total of N loss change rates, that is, the t-th iteration round contains an atrial fibrillation loss change rate and an atrial beat loss change rate, and n can be represented as the n-th loss change rate, that is, the atrial fibrillation loss change rate or the atrial beat loss change rate of the t-th iteration round.

[0163] According to an embodiment of the present invention, the atrial fibrillation loss weighting factor of the t-th iteration round can be calculated according to formula (8), which is shown as follows.

[0164] (8);

[0165] Among them, w 2 (t) can be characterized as the atrial fibrillation loss weighting factor of the t-th iteration round.

[0166] According to an embodiment of the present invention, based on the detection evaluation loss function, the detection evaluation loss value of the tth iteration round is obtained according to the cross entropy loss value corresponding to the tth iteration round, the mean square error loss value corresponding to the tth iteration round, the atrial beat loss weighting factor of the tth iteration round, and the atrial fibrillation loss weighting factor of the tth iteration round.

[0167] According to an embodiment of the present invention, the detection evaluation loss value of the t-th iteration round can be calculated according to formula (9), which is as follows.

[0168] (9);

[0169] Among them, L total (t) can be represented as the detection evaluation loss value of the t-th iteration round, w n(t) can be characterized as the loss weighting factor of the t-th iteration round, including the atrial fibrillation loss weighting factor of the t-th iteration round and the atrial beat loss weighting factor of the t-th iteration round, L n (t) can be characterized as the loss value corresponding to the t-th iteration round, including the cross entropy loss value corresponding to the t-th iteration round and the mean square error loss value corresponding to the t-th iteration round.

[0170] According to an embodiment of the present invention, when the detection evaluation loss value of the t-th iteration round is greater than a predetermined evaluation loss threshold, the weights of the atrial beat loss weighting factor and the atrial fibrillation loss weighting factor are adjusted to obtain a trained information detection model. For example, when the loss output by any two-layer fully connected network changes rapidly, it means that the two-layer fully connected network has a larger optimization space, and thus a larger weight is given; conversely, if the loss no longer changes significantly, it means that the two-layer fully connected network has tended to be optimal. At this time, the model optimization of the information detection model will focus more on the parameter adjustment of other two-layer fully connected networks or the initial spatiotemporal encoder, so as to obtain an information detection model that can output high accuracy.

[0171] According to an embodiment of the present invention, by calculating the respective loss values ​​and loss weighting factors between the first double-layer fully connected network and the second double-layer fully connected network, the total loss value of the detection evaluation sub-model to be trained is obtained according to the respective loss values ​​and loss weighting factors, thereby adjusting the weight parameters in the detection evaluation sub-model to be trained and the weight parameters of the initial spatiotemporal encoder, so that the information detection model can accurately obtain the target-assisted predicted heartbeat information of the object under test, thereby obtaining an information detection model with high robustness, high reliability, high efficiency and high output accuracy.

[0172] Figure 3a A schematic diagram showing a comparison between an electrocardiogram of latent atrial fibrillation and an electrocardiogram of normal sinus rhythm according to an embodiment of the present invention is shown. Figure 3b A schematic diagram showing a comparison between a target electromagnetic echo signal of latent atrial fibrillation and a target electromagnetic echo signal of normal sinus rhythm according to an embodiment of the present invention is shown.

[0173] like Figure 3a~3b As shown, Figure 3a~3b The figure shows the comparison of the electrocardiogram of latent atrial fibrillation and the electrocardiogram of normal sinus rhythm, and the comparison of the target electromagnetic echo signal of latent atrial fibrillation and the target electromagnetic echo signal of normal sinus rhythm. Figure 3a It can be seen that there is almost no difference between the ECG of the subject with normal sinus rhythm and the ECG of the subject with latent atrial fibrillation. The ECG rhythm of both subjects is normal, and the electrical signals corresponding to the atrial beating exist. Figure 3bIt can be seen that in the signal diagram of the electromagnetic echo signal, there is an obvious difference between the normal electromagnetic echo signal of a subject with normal sinus rhythm and the target electromagnetic echo signal of a subject with latent atrial fibrillation. The target electromagnetic echo signal of the subject with latent atrial fibrillation has a very small signal change amplitude in a short period of time, and the atrium is in a weak peristaltic state in a short period of time, while the normal electromagnetic echo signal has obvious signal fluctuations, and the atrial contraction function is normal.

[0174] Figure 4 A schematic diagram of feature visualization of a feature extraction encoder of the present invention during a training process according to an embodiment of the present invention is shown.

[0175] like Figure 4 As shown, Figure 4 The figure shows the feature visualization of the feature extraction encoder during the training process. The horizontal axis can be represented as the distribution rate of the model visualization clustering area after training, and the vertical axis can be represented as the distribution rate of the model visualization clustering area after training. Blue represents normal heart rate, black represents latent atrial fibrillation, and red represents persistent atrial fibrillation. After one round of training of the feature extraction encoder, the feature extraction encoder cannot clearly distinguish and output normal heart rate, latent atrial fibrillation and persistent atrial fibrillation. After multiple rounds of training of the feature extraction encoder, the feature extraction encoder can clearly distinguish and output normal heart rate, latent atrial fibrillation and persistent atrial fibrillation.

[0176] Figure 5a A schematic diagram showing the output result of an untrained spatiotemporal encoder according to an embodiment of the present invention. Figure 5b A schematic diagram showing the output result of a spatiotemporal encoder trained with the prior art according to an embodiment of the present invention. Figure 5c A schematic diagram showing the output result of a spatiotemporal encoder trained by the method of the present invention according to an embodiment of the present invention is shown.

[0177] like Figure 5a~5c As shown, Figure 5a~5cThe prediction results of three spatiotemporal encoder outputs are shown, the horizontal axis can be represented as real information, the vertical axis can be represented as target-assisted detection heartbeat information output by the information detection model, NAF can be represented as non-latent atrial fibrillation, SAF can be represented as latent atrial fibrillation, the upper left square in each 2×2 figure can be represented as the number of objects whose real information and target-assisted detection heartbeat information are both non-latent atrial fibrillation, the upper right square can be represented as the number of objects whose real information is latent atrial fibrillation and whose target-assisted detection heartbeat information is non-latent atrial fibrillation, the lower left square can be represented as the number of objects whose real information is non-latent atrial fibrillation and whose target-assisted detection heartbeat information is latent atrial fibrillation, the lower right square can be represented as the number of objects whose real information and target-assisted detection heartbeat information are both latent atrial fibrillation. It can be seen from the figure that the result output by the spatiotemporal encoder trained by the method of the present invention has the highest accuracy when used to obtain target-assisted detection heartbeat information, and can distinguish between normal heartbeats and heartbeats that may carry latent atrial fibrillation. The indicator information of the prediction results output by the spatiotemporal encoder under the three training conditions is shown in Table 1.

[0178] Table 1

[0179]

[0180] It can also be seen from the table that various indicators of the results output by the spatiotemporal encoder trained by the method of the present invention are good.

[0181] Figure 6a A schematic diagram showing the results output by an untrained detection evaluation sub-model according to an embodiment of the present invention. Figure 6b A schematic diagram showing the output results of the detection evaluation sub-model after training with the prior art according to an embodiment of the present invention. Figure 6c A schematic diagram showing the results output by the detection evaluation sub-model after being trained by the method of the present invention according to an embodiment of the present invention.

[0182] like Figure 6a~6c As shown, Figure 6a~6cThe prediction results output by three detection evaluation sub-models are shown. The horizontal axis can be represented as real information, and the vertical axis can be represented as target-assisted detection heartbeat information output by the information detection model. NAF can be represented as non-latent atrial fibrillation, and SAF can be represented as latent atrial fibrillation. The square in the upper left corner of each 2×2 figure can be represented as the number of objects whose real information and target-assisted detection heartbeat information are both non-latent atrial fibrillation, the square in the upper right corner can be represented as the number of objects whose real information is latent atrial fibrillation and whose target-assisted detection heartbeat information is non-latent atrial fibrillation, the square in the lower left corner can be represented as the number of objects whose real information is non-latent atrial fibrillation and whose target-assisted detection heartbeat information is latent atrial fibrillation, and the square in the lower right corner can be represented as the number of objects whose real information and target-assisted detection heartbeat information are both latent atrial fibrillation. It can be seen from the figure that the accuracy of the results output by the detection evaluation sub-model after training by the method of the present invention is the highest, and it can distinguish between normal heartbeats and heartbeats that may carry latent atrial fibrillation. The indicator information of the prediction results output by the detection evaluation sub-models under the three training conditions is shown in Table 2.

[0183] Table 2

[0184]

[0185] It can also be seen from the table that the various indicators of the results output by the detection and evaluation sub-model after training with the method of the present invention are good.

[0186] Figure 7a A schematic diagram showing the excellence of various prediction indicators obtained after auxiliary detection using the method of the present invention according to an embodiment of the present invention. Figure 7b A schematic diagram showing target auxiliary predicted heartbeat information obtained after auxiliary detection using the method of the present invention with increased monitoring time according to an embodiment of the present invention is shown.

[0187] like Figure 7a~7b As shown, Figure 7a~7b shows a schematic diagram after auxiliary detection of multiple objects to be detected, Figure 7b The horizontal axis can represent the time, and the vertical axis can represent the result output by the method of the present invention. It can be seen from the figure that the performance of the information detection model used for auxiliary detection in the six tested objects is stable and excellent. Even if the output target auxiliary predicted heartbeat information is not accurate enough, the auxiliary prediction effect can be improved by a longer time. Specifically, except for the tested object 5, the performance of the information detection model is stable and excellent, with an average accuracy of 99% and an average F1 score of 0.76. When the auxiliary detection of the tested object 5 is carried out for a longer time, the success rate of the target auxiliary predicted heartbeat information output by the information detection model of the tested object 5 is also improved.

[0188] Figure 8A schematic diagram showing comparison between various indicators of target auxiliary detection heartbeat information obtained after auxiliary prediction using the electrocardiogram of the prior art and the method of the present invention according to an embodiment of the present invention is shown.

[0189] like Figure 8 As shown, Figure 8 The various indicators of the target assisted detection heartbeat information obtained after auxiliary detection using the electrocardiogram of the prior art and the method of the present invention are shown. It can be seen from the figure that the target assisted prediction heartbeat information output by the method of the present invention is superior to the auxiliary information based on the electrocardiogram output in terms of the four evaluation indicators of accuracy, sensitivity, specificity and F1 score. Among them, the method of the present invention has the greatest improvement in F1 score, which is 108% higher than that based on the electrocardiogram method.

[0190] Fig. 9 A schematic diagram of a process for assisting in predicting latent atrial fibrillation according to an embodiment of the present invention is shown.

[0191] like Fig. 9 As shown, Fig. 9 The present invention shows a process for assisting in predicting latent atrial fibrillation, S901 uses a millimeter wave radar to transmit an electromagnetic signal to a measured object, and after receiving the electromagnetic signal, the measured object generates an initial electromagnetic echo signal, and after performing signal separation processing on the initial electromagnetic echo signal, a target electromagnetic echo signal corresponding to the motion state of the heart area is obtained, S902 the target electromagnetic echo signal is input into a spatiotemporal encoder of an information detection model that has been trained by regression contrast learning, and a plurality of mechanical motion spatiotemporal features related to normal heartbeats and abnormal heartbeats in the heart area are output, and then the plurality of mechanical motion spatiotemporal features are input into a feature extraction encoder for feature extraction processing to obtain target mechanical motion time features related to abnormal heartbeats, S903 the target mechanical motion time features related to abnormal heartbeats are input into a first double-layer fully connected network to obtain auxiliary atrial beating information of the measured object, and S904 the target mechanical motion time features related to abnormal heartbeats are input into a second double-layer fully connected network to obtain auxiliary atrial fibrillation information of the measured object.

[0192] Fig.10 A structural block diagram of a device for assisting in predicting latent atrial fibrillation according to an embodiment of the present invention is shown.

[0193] like Fig.10 As shown, the device for evaluating the ejection function value based on the semi-supervised deep segmentation model of this embodiment includes: an acquisition module 1010 , a separation module 1020 , a first extraction module 1030 , a second extraction module 1040 and a obtaining module 1050 .

[0194] The acquisition module 1010 is used to acquire an initial electromagnetic echo signal, wherein the initial electromagnetic echo signal represents a plurality of spatially overlapped signals reflected by a plurality of body regions of the subject within a predetermined period of time. The acquisition module 1010 can be used to perform the operation S210 described above, which will not be described in detail here.

[0195] The separation module 1020 is used to perform signal separation processing on the initial electromagnetic echo signal to obtain a target electromagnetic echo signal corresponding to the motion state of the heart region of the measured object within a predetermined period of time. The separation module 1020 can be used to perform the operation S220 described above, which will not be described in detail here.

[0196] The first extraction module 1030 is used to perform a first feature extraction process on the target electromagnetic echo signal using the spatiotemporal encoder of the information detection model to generate a plurality of mechanical motion spatiotemporal features related to normal heartbeats and abnormal heartbeats in the heart region. The first extraction module 1030 can be used to perform the operation S230 described above, which will not be described in detail here.

[0197] The second extraction module 1040 is used to perform a second feature extraction process on the multiple mechanical motion spatiotemporal features using the feature extraction encoder of the information detection model to generate a target mechanical motion time feature related to the abnormal heartbeat in the heart area. The second extraction module 1040 can be used to perform the operation S240 described above, which will not be described in detail here.

[0198] The obtaining module 1050 is used to perform feature conversion mapping processing on the target mechanical motion time feature using the detection evaluation submodel of the information detection model to obtain the target auxiliary predicted heartbeat information of the object under test. The obtaining module 1050 can be used to perform the operation S250 described above, which will not be repeated here.

[0199] According to an embodiment of the present invention, the separation module 1020 includes: a first beamforming submodule, a first extraction submodule and a first filtering submodule.

[0200] The first beamforming submodule is used to perform beamforming processing on the initial electromagnetic echo signal to obtain multiple intermediate electromagnetic echo signals, wherein the intermediate electromagnetic echo signal represents multiple four-dimensional signals separated from each other and reflected by multiple regions within a predetermined time period.

[0201] The first extraction submodule is used to extract a three-dimensional electromagnetic echo signal corresponding to the heart area of ​​the object under test from the multiple intermediate electromagnetic echo signals.

[0202] The first filtering submodule is used to filter the three-dimensional electromagnetic echo signal using a second-order differential time domain filtering algorithm to obtain a target electromagnetic echo signal.

[0203] According to an embodiment of the present invention, the first extraction submodule includes: a first superposition unit, a first processing unit and a second processing unit.

[0204] The first superposition unit is used to perform superposition processing on multiple intermediate electromagnetic echo signals to obtain an intermediate electromagnetic echo superposition signal.

[0205] The first processing unit is used to process the intermediate electromagnetic echo superposition signal using a constant false alarm rate algorithm to obtain a distance electromagnetic echo signal corresponding to the chest area of ​​the object being measured, wherein the distance electromagnetic echo signal is a three-dimensional signal.

[0206] The second processing unit is used to perform clustering processing on the range electromagnetic echo signal by using a density clustering algorithm to obtain a three-dimensional electromagnetic echo signal corresponding to the heart area of ​​the object under test.

[0207] According to an embodiment of the present invention, the first extraction module 1030 includes: a first splitting submodule and a second extraction submodule.

[0208] The first splitting submodule is used to split the target electromagnetic echo signal into target electromagnetic echo sub-signals of multiple time periods based on a predetermined coding time period.

[0209] The second extraction submodule is used to input the target electromagnetic echo sub-signals of multiple time periods into the space-time encoder of the information detection model for first feature extraction processing, so as to generate multiple mechanical motion space-time features related to normal heartbeat and abnormal heartbeat in the heart area, wherein the multiple mechanical motion space-time features correspond to the target electromagnetic echo sub-signals of multiple time periods.

[0210] According to an embodiment of the present invention, the obtaining module 1050 includes: a first dimensionality reduction submodule, a first conversion submodule and a second conversion submodule.

[0211] The first dimensionality reduction submodule is used to input the target mechanical motion time characteristics into the first linear layer of the first double-layer fully connected network and the first linear layer of the second double-layer fully connected network for dimensionality reduction processing, so as to obtain the first reduced dimensionality mechanical motion state characteristics and the second reduced dimensionality mechanical motion time characteristics.

[0212] The first conversion submodule is used to input the first dimension-reduced mechanical motion state feature into the second linear layer of the first two-layer fully connected network for feature conversion mapping processing to obtain the auxiliary atrial beating information of the object under test.

[0213] The second conversion submodule is used to input the second dimension-reduced mechanical motion time feature into the second linear layer of the second double-layer fully connected network for feature conversion mapping processing to obtain auxiliary atrial fibrillation information of the object under test.

[0214] According to an embodiment of the present invention, the apparatus for assisting in predicting latent atrial fibrillation further includes: a training module.

[0215] Training module, used to train information detection models.

[0216] According to an embodiment of the present invention, the training module includes: a first acquisition submodule, a second splitting submodule, a third extraction submodule, a fourth extraction submodule, a first mapping submodule, a first obtaining submodule and a first adjusting submodule.

[0217] The first acquisition submodule is used to acquire an initial model to be trained and a sample training data set, wherein the initial model to be trained includes an initial spatiotemporal encoder, a feature extraction encoder to be trained and a detection and evaluation submodel to be trained, and the sample training data set includes electromagnetic echo signal samples, latent atrial fibrillation disease period information samples, atrial beat information samples and atrial fibrillation information samples.

[0218] The second splitting submodule is used to split the electromagnetic echo signal sample into a plurality of electromagnetic echo signal sub-samples based on predetermined sample encoding period information.

[0219] The third extraction submodule is used to input multiple electromagnetic echo signal subsamples into the initial spatiotemporal encoder of the initial model to be trained for first feature extraction processing, so as to generate multiple training mechanical motion spatiotemporal features related to normal heartbeat and abnormal heartbeat in the heart area.

[0220] The fourth extraction submodule is used to input the multiple mechanical motion spatiotemporal training features into the feature extraction encoder to be trained for second feature extraction processing to generate training mechanical motion time features related to abnormal heartbeats in the heart area.

[0221] The first mapping submodule is used to input the training mechanical motion time feature into the detection evaluation submodel to be trained for feature conversion mapping processing to obtain the training detection heartbeat information.

[0222] The first submodule is used to obtain a detection evaluation loss value based on a detection evaluation loss function and according to training detection heartbeat information, latent atrial fibrillation disease period information samples, atrial beat information samples and atrial fibrillation information samples.

[0223] The first adjustment submodule is used to adjust the model parameters of the initial model according to the detection evaluation loss value to obtain a trained information detection model.

[0224] According to an embodiment of the present invention, the training module further includes: a second acquisition submodule, a fifth extraction submodule, a second obtaining submodule, a first regression comparison submodule and a second adjustment submodule.

[0225] The second acquisition submodule is used to acquire the basic spatiotemporal encoder to be trained.

[0226] The fifth extraction submodule is used to input multiple electromagnetic echo signal subsamples into the basic spatiotemporal encoder to be trained for first feature extraction processing to generate multiple basic training mechanical motion spatiotemporal features related to normal heartbeats and abnormal heartbeats in the heart area.

[0227] The second obtaining submodule is used to obtain the latent atrial fibrillation period comparison weight based on the preset latent atrial fibrillation period comparison weight function and the latent atrial fibrillation disease period information sample.

[0228] The first regression comparison submodule is used to perform regression comparison learning processing on the latent atrial fibrillation period comparison weights and multiple basic training mechanical motion spatiotemporal features based on a preset spatiotemporal loss function to obtain a spatiotemporal loss value.

[0229] The second adjustment submodule is used to adjust the latent atrial fibrillation period comparison weight of the basic spatiotemporal encoder to be trained according to the spatiotemporal loss value to obtain a trained initial spatiotemporal encoder.

[0230] According to an embodiment of the present invention, the first obtaining submodule includes: a first obtaining unit, a second obtaining unit, a first determining unit, a second determining unit, a third obtaining unit, a fourth obtaining unit and a fifth obtaining unit.

[0231] The first obtaining unit is used to obtain a cross entropy loss value corresponding to the t-th iteration round according to the training atrial beating information and atrial beating information samples of the t-th iteration round in the t-th iteration round, wherein t≥1 and t is a positive integer.

[0232] The second obtaining unit is used to obtain a mean square error loss value corresponding to the t-th iteration round according to multiple training atrial fibrillation information and atrial fibrillation information samples of the t-th iteration round.

[0233] The first determination unit is used to determine the atrial beat loss change rate corresponding to the t-th iteration round according to the average value of the cross entropy loss corresponding to the t-th iteration round and the average value of the cross entropy loss corresponding to the t-1-th iteration round.

[0234] The second determining unit is used to determine the atrial fibrillation loss change rate corresponding to the t-th iteration round according to the mean square error loss average value corresponding to the t iteration rounds and the mean square error loss average value corresponding to the t-1 iteration rounds.

[0235] The third obtaining unit is used to obtain the training loss change rate corresponding to the t-th iteration round according to the atrial beat loss change rate corresponding to the t-th iteration round and the atrial fibrillation loss change rate corresponding to the t-th iteration round.

[0236] The fourth obtaining unit is used to obtain the atrial beat loss weighting factor and the atrial fibrillation loss weighting factor of the t-th iteration round based on the preset beat loss change rate function and the preset fibrillation loss change rate function, according to the training loss change rate corresponding to the t-th iteration round, the atrial beat loss change rate corresponding to the t-th iteration round, and the atrial fibrillation loss change rate corresponding to the t-th iteration round.

[0237] The fifth obtaining unit is used to obtain the detection evaluation loss value of the tth iteration round based on the detection evaluation function, according to the cross entropy loss value corresponding to the tth iteration round, the mean square error loss value corresponding to the tth iteration round, the atrial beat loss weighting factor of the tth iteration round, and the atrial fibrillation loss weighting factor.

[0238] According to an embodiment of the present invention, any multiple modules among the acquisition module 1010, the separation module 1020, the first extraction module 1030, the second extraction module 1040 and the acquisition module 1050 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present invention, at least one of the acquisition module 1010, the separation module 1020, the first extraction module 1030, the second extraction module 1040 and the acquisition module 1050 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation modes of software, hardware and firmware or in a proper combination of any of them. Alternatively, at least one of the acquisition module 1010, the separation module 1020, the first extraction module 1030, the second extraction module 1040 and the obtaining module 1050 may be at least partially implemented as a computer program module, and when the computer program module is executed, a corresponding function may be performed.

[0239] Fig.11 A block diagram of an electronic device for assisting in predicting latent atrial fibrillation according to an embodiment of the present invention is shown.

[0240] like Fig.11As shown, the electronic device according to an embodiment of the present invention includes a processor 1101, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1102 or a program loaded from a storage part 1108 to a random access memory (RAM) 1103. The processor 1101 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a dedicated microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1101 may also include an onboard memory for caching purposes. The processor 1101 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present invention.

[0241] In RAM 1103, various programs and data required for the operation of the electronic device are stored. Processor 1101, ROM 1102 and RAM 1103 are connected to each other via bus 1104. Processor 1101 performs various operations of the method flow according to the embodiment of the present invention by executing the program in ROM 1102 and / or RAM 1103. It should be noted that the program can also be stored in one or more memories other than ROM 1102 and RAM 1103. Processor 1101 can also perform various operations of the method flow according to the embodiment of the present invention by executing the program stored in the one or more memories.

[0242] According to an embodiment of the present invention, the electronic device may further include an input / output (I / O) interface 1105, which is also connected to the bus 1104. The electronic device may further include one or more of the following components connected to the I / O interface 1105: an input portion 1106 including a keyboard, a mouse, etc.; an output portion 1107 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 1108 including a hard disk, etc.; and a communication portion 1109 including a network interface card such as a LAN card, a modem, etc. The communication portion 1109 performs communication processing via a network such as the Internet. A drive 1110 is also connected to the I / O interface 1105 as needed. A removable medium 1111, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1110 as needed, so that a computer program read therefrom is installed into the storage portion 1108 as needed.

[0243] The present invention also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiment; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present invention is implemented.

[0244] According to an embodiment of the present invention, the computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus or a device. For example, according to an embodiment of the present invention, the computer-readable storage medium may include the ROM 1102 and / or RAM 1103 described above and / or one or more memories other than ROM 1102 and RAM 1103.

[0245] The embodiment of the present invention also includes a computer program product, which includes a computer program, and the computer program contains program code for executing the method shown in the flowchart. When the computer program product is run in a computer system, the program code is used to enable the computer system to implement the method for assisting in predicting latent atrial fibrillation provided by the embodiment of the present invention.

[0246] The computer program executes the above functions defined in the system / device of the embodiment of the present invention when it is executed by the processor 1101. According to the embodiment of the present invention, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0247] In one embodiment, the computer program may rely on tangible storage media such as optical storage devices, magnetic storage devices, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of signals on a network medium, and downloaded and installed through the communication part 1109, and / or installed from a removable medium 1111. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0248] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 1109, and / or installed from the removable medium 1111. When the computer program is executed by the processor 1101, the above functions defined in the system of the embodiment of the present invention are performed. According to the embodiment of the present invention, the system, device, means, module, unit, etc. described above can be implemented by a computer program module.

[0249] It will be appreciated by those skilled in the art that the features described in the various embodiments of the present invention may be combined and / or combined in various ways, even if such combinations and / or combinations are not explicitly described in the present invention. In particular, without departing from the spirit and teachings of the present invention, the features described in the various embodiments of the present invention may be combined and / or combined in various ways. All of these combinations and / or combinations fall within the scope of the present invention.

Claims

1. A method for assisting in the prediction of latent atrial fibrillation, characterized in that: include: Acquiring an initial electromagnetic echo signal, wherein the initial electromagnetic echo signal represents a plurality of signals spatially overlapped with each other reflected by a plurality of body regions of the measured object within a predetermined time period; Performing signal separation processing on the initial electromagnetic echo signal to obtain a target electromagnetic echo signal corresponding to the motion state of the heart region of the measured object within the predetermined time period; Performing a first feature extraction process on the target electromagnetic echo signal using a spatiotemporal encoder of the information detection model to generate a plurality of mechanical motion spatiotemporal features related to normal heartbeats and abnormal heartbeats in the heart region; Performing a second feature extraction process on the plurality of mechanical motion spatiotemporal features using a feature extraction encoder of the information detection model to generate a target mechanical motion time feature related to an abnormal heartbeat in the heart region; The detection evaluation sub-model of the information detection model is used to perform feature conversion mapping processing on the target mechanical motion time characteristics to obtain the target auxiliary predicted heartbeat information of the detected object.

2. The method according to claim 1, characterized in that: The performing signal separation processing on the initial electromagnetic echo signal to obtain a target electromagnetic echo signal corresponding to the motion state of the heart region of the measured object within the predetermined time period includes: Performing beamforming processing on the initial electromagnetic echo signal to obtain a plurality of intermediate electromagnetic echo signals, wherein the intermediate electromagnetic echo signals represent a plurality of four-dimensional signals that are spatially separated from each other and reflected by a plurality of regions within a predetermined time period; extracting a three-dimensional electromagnetic echo signal corresponding to the heart region of the measured object from the plurality of intermediate electromagnetic echo signals; The three-dimensional electromagnetic echo signal is filtered using a second-order differential time-domain filtering algorithm to obtain the target electromagnetic echo signal.

3. The method according to claim 2, characterized in that The step of extracting a three-dimensional electromagnetic echo signal corresponding to the heart region of the measured object from the plurality of intermediate electromagnetic echo signals comprises: Performing superposition processing on the multiple intermediate electromagnetic echo signals to obtain an intermediate electromagnetic echo superposition signal; The intermediate electromagnetic echo superposition signal is processed by using a constant false alarm rate algorithm to obtain a distance electromagnetic echo signal corresponding to the chest area of ​​the measured object, wherein the distance electromagnetic echo signal is a three-dimensional signal; The distance electromagnetic echo signal is clustered using a density clustering algorithm to obtain a three-dimensional electromagnetic echo signal corresponding to the heart area of ​​the measured object.

4. The method according to claim 1, characterized in that: The spatiotemporal encoder using the information detection model performs a first feature extraction process on the target electromagnetic echo signal to generate a plurality of mechanical motion spatiotemporal features related to the normal heartbeat and the abnormal heartbeat of the heart area, including: Based on a predetermined coding period, splitting the target electromagnetic echo signal into target electromagnetic echo sub-signals of a plurality of time periods; The target electromagnetic echo sub-signals of the multiple time periods are respectively input into the space-time encoder of the information detection model for a first feature extraction process to generate a plurality of mechanical motion space-time features related to the normal heartbeat and abnormal heartbeat of the heart area, wherein the plurality of mechanical motion space-time features correspond to the target electromagnetic echo sub-signals of the multiple time periods.

5. The method according to claim 1, characterized in that The detection evaluation sub-model includes a first double-layer fully connected network and a second double-layer fully connected network, wherein the first double-layer fully connected network and the second double-layer fully connected network each include two linear layers; The method of using the detection evaluation sub-model of the information detection model to perform feature conversion mapping processing on the target mechanical motion time feature to obtain the target auxiliary predicted heartbeat information of the detected object includes: Inputting the target mechanical motion time feature into the first linear layer of the first double-layer fully connected network and the first linear layer of the second double-layer fully connected network for dimensionality reduction processing respectively, to obtain a first reduced-dimensionality mechanical motion state feature and a second reduced-dimensionality mechanical motion time feature; Inputting the first dimension-reduced mechanical motion state feature into the second linear layer of the first double-layer fully connected network for feature conversion mapping processing to obtain auxiliary atrial beating information of the measured object; The second dimension-reduced mechanical motion time feature is input into the second linear layer of the second two-layer fully connected network for feature conversion mapping processing to obtain auxiliary atrial fibrillation information of the measured object.

6. The method according to claim 1, characterized in that The information detection model is trained using the following methods, including: Acquire an initial model to be trained and a sample training data set, wherein the initial model to be trained includes an initial spatiotemporal encoder, a feature extraction encoder to be trained, and a detection evaluation sub-model to be trained, and the sample training data set includes electromagnetic echo signal samples, latent atrial fibrillation disease period information samples, atrial beat information samples, and atrial fibrillation information samples; Based on predetermined sample encoding period information, splitting the electromagnetic echo signal sample into a plurality of electromagnetic echo signal sub-samples; Inputting the plurality of electromagnetic echo signal subsamples into the initial spatiotemporal encoder of the initial model to be trained for first feature extraction processing to generate a plurality of training mechanical motion spatiotemporal features related to normal heartbeats and abnormal heartbeats in the heart region; Inputting the plurality of mechanical motion spatiotemporal training features into a feature extraction encoder to be trained for second feature extraction processing to generate training mechanical motion time features related to abnormal heartbeats in the heart region; Inputting the training mechanical motion time feature into the detection evaluation sub-model to be trained for feature conversion mapping processing to obtain training detection heartbeat information; Based on the detection evaluation loss function, a detection evaluation loss value is obtained according to the training detection heartbeat information, the latent atrial fibrillation disease period information sample, the atrial beat information sample and the atrial fibrillation information sample; According to the detection evaluation loss value, the model parameters of the initial model are adjusted to obtain a trained information detection model.

7. The method according to claim 6, characterized in that The initial spatiotemporal encoder is pre-trained, and the pre-training method includes the following operations: Obtain the basic spatiotemporal encoder to be trained; Inputting the plurality of electromagnetic echo signal subsamples into the basic spatiotemporal encoder to be trained for first feature extraction processing to generate a plurality of basic training mechanical motion spatiotemporal features related to normal heartbeats and abnormal heartbeats in the heart region; Based on a preset latent atrial fibrillation period comparison weight function, and according to the latent atrial fibrillation disease period information sample, a latent atrial fibrillation period comparison weight is obtained; Based on a preset spatiotemporal loss function, performing regression contrast learning processing on the latent atrial fibrillation period contrast weight and the plurality of basic training mechanical motion spatiotemporal features to obtain a spatiotemporal loss value; According to the spatiotemporal loss value, the latent atrial fibrillation period comparison weight of the basic spatiotemporal encoder to be trained is adjusted to obtain the trained initial spatiotemporal encoder.

8. The method according to claim 6, characterized in that The training and detecting heartbeat information includes training atrial beating information and training atrial fibrillation information; The step of obtaining the detection evaluation loss value based on the detection evaluation loss function according to the training detection heartbeat information, the latent atrial fibrillation disease period information sample, the atrial beat information sample and the atrial fibrillation information sample comprises multiple iteration rounds, wherein the operation of the t-th iteration round comprises: In the t-th iteration round, according to the training atrial beat information and the atrial beat information sample of the t-th iteration round, a cross entropy loss value corresponding to the t-th iteration round is obtained, wherein t≥1 and t is a positive integer; Obtaining a mean square error loss value corresponding to the t-th iteration round according to the plurality of training atrial fibrillation information and the atrial fibrillation information samples of the t-th iteration round; Determine the atrial beating loss change rate corresponding to the t-th iteration round according to the average value of the cross entropy loss corresponding to the t-th iteration round and the average value of the cross entropy loss corresponding to the t-1-th iteration round; Determine the atrial fibrillation loss change rate corresponding to the t-th iteration round according to the mean square error loss average value corresponding to the t iteration rounds and the mean square error loss average value corresponding to the t-1 iteration rounds; Obtaining a training loss change rate corresponding to the t-th iteration round according to the atrial beat loss change rate corresponding to the t-th iteration round and the atrial fibrillation loss change rate corresponding to the t-th iteration round; Based on a preset beating loss change rate function and a preset fibrillation loss change rate function, according to the training loss change rate corresponding to the t-th iteration round, the atrial beating loss change rate corresponding to the t-th iteration round, and the atrial fibrillation loss change rate corresponding to the t-th iteration round, obtaining an atrial beating loss weighting factor and an atrial fibrillation loss weighting factor of the t-th iteration round; Based on the detection evaluation loss function, the detection evaluation loss value of the tth iteration round is obtained according to the cross entropy loss value corresponding to the tth iteration round, the mean square error loss value corresponding to the tth iteration round, the atrial beat loss weighting factor of the tth iteration round, and the atrial fibrillation loss weighting factor of the tth iteration round.

9. A device for assisting in predicting latent atrial fibrillation, characterized in that: include: An acquisition module, configured to acquire an initial electromagnetic echo signal, wherein the initial electromagnetic echo signal represents a plurality of signals spatially overlapped with each other and reflected by a plurality of body regions of the subject within a predetermined period of time; A separation module, configured to perform signal separation processing on the initial electromagnetic echo signal to obtain a target electromagnetic echo signal corresponding to the motion state of the heart region of the measured object within the predetermined time period; a first extraction module, configured to perform a first feature extraction process on the target electromagnetic echo signal using a spatiotemporal encoder of the information detection model, and generate a plurality of mechanical motion spatiotemporal features related to a normal heartbeat and an abnormal heartbeat in the heart region; A second extraction module, configured to perform a second feature extraction process on the plurality of mechanical motion spatiotemporal features using a feature extraction encoder of the information detection model, to generate a target mechanical motion time feature related to an abnormal heartbeat in the heart region; The obtaining module is used to perform feature conversion mapping processing on the target mechanical motion time characteristics by using the detection evaluation sub-model of the information detection model to obtain the target auxiliary predicted heartbeat information of the object under test.

10. An electronic device, characterized in that: include: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors execute the method according to any one of claims 1 to 8.

Images