Electrocardiogram vector reconstruction method based on unsupervised learning

Abstract

The invention discloses an electrocardio vector reconstruction method based on unsupervised learning. The method carries out the reconstruction of an electrocardio vector of an inputted standard 12-lead electrocardiogram through employing a neural network. In the training process, a method of firstly mapping the standard 12 lead to the electrocardiogram vector and then restoring the 12 lead electrocardiogram by using a projection method is used, so that the problem that the traditional method depends on the corresponding data of the 12 lead and the electrocardiogram vector is solved, the utilization efficiency of the data is obviously improved, and the data cost is reduced. During reconstruction, a neural network is used for recalculating a projection vector to perform reconstruction froman electrocardiogram vector to a 12-lead electrocardiogram, and a regularization term is used for constraining the projection vector in a final loss calculation module, so that the interpretability and the accuracy of a reconstruction process are ensured while the individual difference of the electrocardiogram is solved. The multi-order differential loss is used in a final loss calculation module,so that the problems of low frequency, such as baseline interference and the like, are avoided on the basis of ensuring morphological characteristics.

Description

Technical field[0001]The invention relates to the technical field of electrocardiogram signal processing, in particular to an electrocardiogram vector reconstruction method based on unsupervised learning.Background technique[0002]The electrocardiogram has been invented in 1885 and has a history of more than one hundred years. Among the electrocardiograms composed of various leads, the conventional 12-lead electrocardiogram is the most widely used today.[0003]The vector electrocardiogram is a spatial electrocardiogram loop formed by the depolarization of the heart. The three-dimensional electrocardiogram vector loop is called the three-dimensional electrocardiogram vector. The graph generated by the projection of the three-dimensional electrocardiogram vector to the frontal, transverse and lateral planes is called a flat electrocardiogram. At present, the medical community has reached a consensus that vector ECG is better than ECG in the diagnosis of ventricular hypertrophy, bundle b...

AI Technical Summary

Problems solved by technology

[0006] Design the reconstruction method of the projection matrix through the projection relationship between the ECG vector and the leads: In actual situations, due to the difference in body posture of different people, when the ECG lead line is placed on the body surface of the subject, compared with the standard placement method There is a positional deviation. Although the collected 12-lead ECG does not affect the final diagnosis, it will cause relatively large errors in the reconstruction of the vector cardiogram, which will affect the final diagnosis.

Moreover, since the collection of ECG will have various interference problems such as baseline drift, these interferences are easily overlooked by physicians in the interpretation of 12-lead ECG, but the ECG reconstructed according to the projection matrix often has obvious electrical axis offset, etc. and other issues have a non-negligible impact on the actual interpretation

[0007] Using the hybrid ECG collected synchronously from the 12-lead ECG and the vector ECG as the supervised training data to train the neural network to reconstruct the ECG vector: this kind of data has high requirements on the acquisition equipment, and the data cost in the actual clinical data It is also relatively high, and since the collected ECG and 12-lead ECG have various interferences such as baseline drift, in order to enable the neural network to have the ability to filter interference, a large-scale data set is required

And because the method of reconstructing the ECG vector based on the neural network does not consider modeling the interference, the fitting effect of this method is not ideal even in the mixed ECG

Due to the lack of fitting ability, data interference and high data cost, the current supervised learning method is used to train the neural network for ECG vector reconstruction, and the effect is not ideal.

[0008] The above two methods are sensitive to the noise of ECG acquisition. Although signal processing methods such as filtering can be used to preprocess the ECG to reduce the influence of interference, the ECG filtering algorithm will be more or less Affect the morphological characteristics of the ECG. For example, the EMG filtering algorithm will have a greater impact on the shape of the small q wave and the notch of the QRS complex. Combining the various shortcomings of the above two methods, how can we only use the unlabeled It is particularly important to design a method for finely reconstructing the 12-lead ECG for low-cost data such as the 12-lead ECG