Search

CN-122004889-A - Reconstruction method, device and system of twelve-lead signal

CN122004889ACN 122004889 ACN122004889 ACN 122004889ACN-122004889-A

Abstract

The invention discloses a method, a device and a system for reconstructing twelve-lead signals, belonging to the technical field of biomedical signal detection; the method inputs the single-lead electrocardiosignal into a trained twelve-lead signal reconstruction model, and can synthesize a clinically available twelve-lead electrocardiogram. The twelve-lead signal reconstruction model comprises a pre-coding module, a migration module and a reconstruction module which are sequentially connected, global features are extracted through collaborative coding, a domain self-adaptive module is utilized to map target lead features, and finally, the twelve-lead signal conforming to the electrophysiological rule is generated by denoising of a conditional diffusion model. Compared with the traditional method, the method has the advantages that the signal fidelity, the heart beat maintainability and the diagnosis consistency are obviously improved on the multi-data set, the generalization capability under the open scene is considered, and the application value of the wearable equipment in the remote monitoring and early screening of cardiovascular diseases is effectively enhanced.

Inventors

  • SHEN GANG
  • Jin Fangzhuo
  • CHEN SHICHENG
  • QIU JIAFENG

Assignees

  • 华中科技大学

Dates

Publication Date
20260512
Application Date
20260129

Claims (10)

  1. 1. A method for reconstructing a twelve lead signal, comprising: Single lead electrocardiosignal of object to be detected Inputting the trained twelve-lead signal reconstruction model to obtain twelve-lead electrocardio target signals ; The twelve-lead signal reconstruction model comprises a pre-coding module, a migration module and a reconstruction module which are sequentially connected, wherein the twelve-lead signal reconstruction model is used for minimizing twelve-lead electrocardiographic target signals output by training in the training process True twelve-lead electrocardiosignal input by training The error between them is the target; Wherein the pre-coding module is used for carrying out pre-coding on the single-lead electrocardiosignals Preliminary reconstruction is carried out to obtain twelve-lead electrocardiosignal The coding module is used for generating twelve-lead electrocardiosignals Space-time joint coding to obtain global potential characteristics The migration module is used for integrating the global potential features Mapping to target lead domain to obtain domain self-adaptive characteristic The reconstruction module is used for using the single-lead electrocardiosignal And the domain adaptation feature For the diffusion joint condition, obtaining eleven-lead reconstruction results from Gaussian noise by step-by-step sampling, and obtaining the single-lead electrocardiosignal And the eleven-lead reconstruction result are fused to obtain the twelve-lead electrocardio target signal 。
  2. 2. The method for reconstructing twelve-lead signal according to claim 1, wherein the precoding module comprises a main UNet network With secondary UNet network When the input is twelve-lead electrocardiographic real signals In the case of a single training sample of (a), The main UNet network The first twelve-lead electrocardiosignal with single lead electrocardiosignal in single training sample as input and output The difference between the training sample and the single training sample is minimized as a training target; The secondary UNet network Any one of the other lead electrocardiosignals except the single lead electrocardiosignal in the single training sample is taken as input, and the second twelve-lead electrocardiosignal is output The difference from the single training samples is minimized to the training objective.
  3. 3. The method for reconstructing twelve-lead signal according to claim 2, wherein the joint loss of the precoding module is during training The method comprises the following steps: ; Wherein, the The loss is re-established for the primary UNet network, , For the secondary UNet network to reestablish the loss, , The weight coefficients for the feature alignment penalty, In order to account for the loss of alignment of features, , 、 Intermediate features of the primary UNet network and the primary UNet network respectively, Representing the euclidean norm.
  4. 4. The method of claim 1, wherein the encoding module comprises an encoder and a decoder, and wherein the input is a twelve-lead electrocardiographic real signal during training The training object of the coding module is that the encoder codes the single training sample with mask to obtain global characteristics The decoder is used for the global feature The difference between the third twelve-lead electrocardiographic signal decoded and the single training sample is minimized.
  5. 5. The method for reconstructing a twelve lead signal according to claim 4, wherein the total loss of the encoding module is: ; Wherein, the Is that Is a single cardiac segment of the heart(s), For a single reconstructed segment in the third twelve-lead electrocardiosignal, n is a segment And The total number of each of the two, For the ith mask to be used, For the Hadamard element-by-element product, Representing the euclidean norm, For the loss of weights in the frequency domain, In order for the spectral alignment to be lost, , Is the amplitude spectrum obtained by short-time Fourier transform.
  6. 6. The method for reconstructing twelve-lead signals according to claim 4, wherein said migration module comprises a domain migration network Rhythm classifier After the coding module training is finished, the twelve-lead electrocardio real signals are obtained in the migration module training process Is input to the encoder in the encoding module Make it output global features , Reuse of the domain migration network Mapping it to the target lead domain to obtain And utilizes a rhythm classifier For a pair of And (5) performing category discrimination.
  7. 7. The method for reconstructing a twelve lead signal according to claim 6, wherein the total loss of the migration module is: ; Wherein, the For the domain alignment loss, , For the target lead field reference feature, , In order to classify the loss of a rhythm, The weights are lost for classification.
  8. 8. A reconstruction device of twelve-lead signals is characterized in that the reconstruction device is used for single-lead electrocardiosignals of an object to be detected Inputting the trained twelve-lead signal reconstruction model to obtain twelve-lead electrocardio target signals The twelve-lead signal reconstruction model comprises a pre-coding module, a migration module and a reconstruction module which are sequentially connected, wherein the twelve-lead signal reconstruction model is used for minimizing twelve-lead electrocardiographic target signals output by training in the training process True twelve-lead electrocardiosignal input by training The error between them is the target; Wherein the pre-coding module is used for carrying out pre-coding on the single-lead electrocardiosignals Preliminary reconstruction is carried out to obtain twelve-lead electrocardiosignal The coding module is used for generating twelve-lead electrocardiosignals Space-time joint coding to obtain global potential characteristics The migration module is used for integrating the global potential features Mapping to target lead domain to obtain domain self-adaptive characteristic The reconstruction module is used for using the single-lead electrocardiosignal And the domain adaptation feature For the diffusion joint condition, obtaining eleven-lead reconstruction results from Gaussian noise by step-by-step sampling, and obtaining the single-lead electrocardiosignal And the eleven-lead reconstruction result are fused to obtain the twelve-lead electrocardio target signal 。
  9. 9. A twelve lead signal reconstruction system comprising a memory and a processor, said memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 7 when said computer program is executed.
  10. 10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 7.

Description

Reconstruction method, device and system of twelve-lead signal Technical Field The invention belongs to the technical field of biomedical signal detection, and particularly relates to a method, a device and a system for reconstructing twelve-lead signals. Background Cardiovascular disease is one of the leading causes of global death, and is expected to result in more than two tens of millions of deaths each year in 2030, so early discovery and intervention of cardiovascular disease is of great public health importance. Twelve lead electrocardiography is a gold standard for clinical assessment of cardiac electrical activity and diagnosis of complex arrhythmias, myocardial ischemia, etc., which characterizes the three-dimensional electrical activity process of the heart from multiple angles by simultaneous measurement of electrocardiographic signals from multiple locations on the extremities and chest. However, conventional twelve-lead electrocardiographic acquisition relies on specialized equipment and operating experience, and long-term continuous monitoring is difficult to achieve. Wearable devices, while portable, typically collect only a single lead signal, limiting their diagnostic capabilities under complex cardiac pathology. To solve this contradiction, researchers have attempted to reconstruct twelve lead signals from a single lead by linear transformation or deep learning, but in practical applications, reconstructing the accuracy difference. Thus, there is a need for a system that can fully exploit the implicit correlation between twelve leads under single lead input conditions to meet the needs of remote continuous monitoring and early screening. Disclosure of Invention Aiming at the defects or improvement demands of the prior art, the invention provides a method, a device and a system for reconstructing twelve-lead signals, and aims to solve the technical problem that the conventional wearable equipment cannot reconstruct twelve-lead signals from a single lead with high precision. To achieve the above object, according to one aspect of the present invention, there is provided a method for reconstructing a twelve-lead signal, comprising: Single lead electrocardiosignal of object to be detected Inputting the trained twelve-lead signal reconstruction model to obtain twelve-lead electrocardio target signals; The twelve-lead signal reconstruction model comprises a pre-coding module, a migration module and a reconstruction module which are sequentially connected, wherein the twelve-lead signal reconstruction model is used for minimizing twelve-lead electrocardiographic target signals output by training in the training processTrue twelve-lead electrocardiosignal input by trainingThe error between them is the target; Wherein the pre-coding module is used for carrying out pre-coding on the single-lead electrocardiosignals Preliminary reconstruction is carried out to obtain twelve-lead electrocardiosignalThe coding module is used for generating twelve-lead electrocardiosignalsSpace-time joint coding to obtain global potential characteristicsThe migration module is used for integrating the global potential featuresMapping to target lead domain to obtain domain self-adaptive characteristicThe reconstruction module is used for using the single-lead electrocardiosignalAnd the domain adaptation featureFor the diffusion joint condition, obtaining eleven-lead reconstruction results from Gaussian noise by step-by-step sampling, and obtaining the single-lead electrocardiosignalAnd the eleven-lead reconstruction result are fused to obtain the twelve-lead electrocardio target signal。 Further, the precoding module comprises a main UNet networkWith secondary UNet networkWhen the input is twelve-lead electrocardiographic real signalsIn the case of a single training sample of (a), The main UNet networkThe first twelve-lead electrocardiosignal with single lead electrocardiosignal in single training sample as input and outputThe difference between the training sample and the single training sample is minimized as a training target; The secondary UNet network Any one lead electrocardiosignal except the single lead electrocardiosignal in the single training is taken as input, and the second twelve lead electrocardiosignal is outputThe difference from the single training samples is minimized to the training objective. Further, during training, joint loss of the precoding moduleThe method comprises the following steps: Wherein, the method comprises the steps of, The loss is re-established for the primary UNet network,,For the secondary UNet network to reestablish the loss,,The weight coefficients for the feature alignment penalty,In order to account for the loss of alignment of features,,、Intermediate features of the primary UNet network and the primary UNet network respectively,Representing the euclidean norm. Further, the coding module comprises an encoder and a decoder, and in the training process, when the input is twelve-