Abstract

Info

Author(s):: ZHANG Ruifang1; LIANG Yongbo1; 2; 3; CUI Mou1; CHEN Zhencheng1; 2; 3; 1. School of Life and Environmental Sciences, GuiLin University of Electronic Technology, GuiLin 541004, China 2. Guangxi Colleges and Universities Key Laboratory of Biomedical Sensors and Intelligent Instruments, Guilin 541004, China 3. Guangxi Engineering Technology Research Center of Human Physiological Information Noninvasive Detection, Guilin 541004, China

Keywords:: atrial fibrillation deep learning pulse wave Resnet-CBAM Gramian angular field

Abstract:: To address the problems in diagnosis and detection of atrial fibrillation (AF) and invasive pathological examination, a model for AF classification based on pulse waves and deep learning is constructed to realize the accurate prediction of AF. The data collected from the photoplethysmography (PPG) acquisition device and the MIMIC-III database data are used to establish PPG-AF dataset, and a ResNet-CBAM-1DCNN dual-channel convolutional neural network for AF classification is constructed based on the Pytorch deep learning framework. The established dataset is divided into a training set, a validation set and a test set in a ratio of 8:1:1. The PPG and its corresponding Gramian angular field map are taken as input. After the optimization of network structure and hyperparameters, the proposed model obtains a F1 score of 97.30% in the test set, and has an accuracy of 98.12% for AF classification. The multimodal fusion approach based on PPG and dual-channel convolutional neural network can achieve the accurate diagnosis of AF, which is expected to provide an important basis for decision-making in clinic.