Abstract:Non-invasive blood pressure prediction is an important method to prevent diseases such as hypertension. This paper proposes a sub-network aggregation with large convolution kernel convolution to predict non-invasive blood pressure. First, the large convolution kernel module in the backbone network is used to extract PPG data features. Then, the multi-scale features of the backbone network are fused and then aggregated with the features extracted from the parallel subnetwork. Finally, ABP data is predicted by convolution, and then blood pressure is predicted according to the relationship between ABP data and blood pressure. In this work, the large convolution kernel is used to extract more information, and the feature extraction of subnetwork is used to help the prediction of backbone network, which further achieves improved prediction. Under the BHS standard, the prediction accuracy of blood pressure based on DBP and MAP can reach grade A. In addition, the prediction accuracy of DBP and MAP can also reach the standard in terms of AAMI standard.

A Sub-network Aggregation Neural Network for Non-invasive Blood Pressure Prediction