A Personalized ECG Signal Monitoring Method Based on Federated Learning

Abstract

A personalized ECG signal monitoring method based on federated learning. The initialization model trained by the cloud is split into two parts: the basic feature extraction network and the personalized classification network. Only the federated strategy is used to train the basic feature extraction network uploaded by multiple terminals. . Since each terminal only uploads part of the model parameters, malicious personnel can be prevented from restoring the original data through the entire model parameters. Model training through federated policies has higher communication efficiency than passive sequential parameter updates. In order to better learn the personalized features of individuals and improve the accuracy of the classification model, multiple terminals learn the basic features through collaborative training of the basic feature extraction network, and the personalized classification network is only reserved in the terminal for the extraction of personalized features. When a new individual joins, the basic feature extraction network can be directly used as the basic feature extraction part of the new individual, and only a small amount of individual data is needed to obtain a personalized ECG signal monitoring model.

Description

technical field [0001] The invention relates to the field of electrocardiographic signal monitoring, in particular to a personalized electrocardiographic signal monitoring method based on federated learning. Background technique [0002] Electrocardiogram signals represent the electrical activity of the heart. Artificial intelligence technology combined with intelligent ECG signal monitoring terminals and cloud big data platforms deployed in the Internet of Things, as well as various mobile wearable devices, has been widely used in the monitoring and monitoring of individual ECG signals. classification, such as figure 1 . By deploying a lightweight network model on the terminal, the burden of data processing on the cloud can be reduced, and this ECG signal processing solution can reduce time and place constraints. There are still some challenges in solving personalized ECG signal monitoring. [0003] The traditional ECG signal classification method uses relatively small s...

AI Technical Summary

Problems solved by technology

In real scenarios, the acquisition of ECG signals by wearable sensor nodes usually requires experts to mark the ECG signals of each individual, which is a heavy workload and prone to misjudgment

From the above analysis, it can be concluded that it is very difficult to obtain large-scale and high-quality labeled data (individual and shared). If the model of an individual is collaboratively trained through data transfer between multiple individuals, there will still be problems in the process of data transfer. privacy leak

Images