HADLN: Hybrid Attention-Based Deep Learning Network for Automated Arrhythmia Classification

Abstract

In recent years, with the development of artificial intelligence, deep learning model has achieved initial success in ECG data analysis, especially the detection of atrial fibrillation. In order to solve the problems of ignoring the correlation between contexts and gradient dispersion in traditional deep convolution neural network model, the hybrid attention-based deep learning network (HADLN) method is proposed to implement arrhythmia classification. The HADLN can make full use of the advantages of residual network (ResNet) and bidirectional long–short-term memory (Bi-LSTM) architecture to obtain fusion features containing local and global information and improve the interpretability of the model through the attention mechanism. The method is trained and verified by using the PhysioNet 2017 challenge dataset. Without loss of generality, the ECG signal is classified into four categories, including atrial fibrillation, noise, other, and normal signals. By combining the fusion features and the attention mechanism, the learned model has a great improvement in classification performance and certain interpretability. The experimental results show that the proposed HADLN method can achieve precision of 0.866, recall of 0.859, accuracy of 0.867, and F1-score of 0.880 on 10-fold cross-validation.

Artificial intelligence for atrial fibrillation detection, prediction, and treatment: A systematic review of the last decade (2013–2023)

“…Some studies detect AF from a single‐lead ECG (M. Jiang et al, 2021; Z. Li, Feng, et al, 2019; Yu et al, 2022; O. Zhang, Ding, et al, 2021). Fang et al (2023) developed a dual‐channel neural network for AF detection.…”

“…Their analysis highlighted that the simple model remained the gold standard for risk prediction of AF. M. Jiang et al (2021) created a convolutional neural network predicting AF recurrence risk post‐catheter ablation using 12‐lead ECG data. Using a multi‐modal ML approach incorporating ECG and cardiac imaging data (radiomics), Pujadas et al (2022) reported better AF prediction than models using ECG features alone.…”

“…Some studies detect AF from a single-lead ECG (M. Jiang et al, 2021;Z. Li, Feng, et al, 2019;Yu et al, 2022;O.…”

Hemodynamic functional connectivity optimization of frequency EEG microstates enables attention LSTM framework to classify distinct temporal cortical communications of different cognitive tasks

“…Recently, the attention mechanism has been introduced [ 55 ] to improve the performance of deep learning models; it highlights the more informative feature and subsequently gives higher weights to the corresponding original feature sequence. It has been embedded with the LSTM architecture in several EEG studies [ 21 , 28 , 48 , 61 , 62 , 66 , 67 ] by effectively selecting the feature information and observed with significantly improved efficiency and performance accuracy of deep learning systems.…”

Application of Dense Neural Networks for Detection of Atrial Fibrillation and Ranking of Augmented ECG Feature Set

“…The interest to the AF detection problem is intensified during the last three years after the Challenge, showing an advanced progress in ECG signal processing by DNN. Some basic neural architectures are further disclosed as an overview of the enhanced ECG representation of local and global rhythm information in deep layers, including: hybrid attention-based DNN (HADLN), embedding residual network (ResNet) and bidirectional Bi-LSTM architectures with attention mechanism to improve the interpretability of the model [ 42 ]; beat–interval–texture convolutional neural network (BIT-CNN) for image analysis of 2D time–morphology representations of the heartbeats (electrocardiomatrix) [ 43 ]; MultiFusionNet for multiplicative fusion of two DNNs trained on different sources of knowledge, including raw ECG data and large feature set from the Poincaré plot, average beat, cross correlation, fiducial points (intervals and amplitudes), presence of P-waves and atypical ventricular morphologies [ 44 ]; and an end-to-end 1D CNN with 10 convolutional blocks, two fully-connected layers, and an output Softmax classifier that uses only raw ECG input with data length normalization [ 45 ].…”