基于BiLSTM-Attention的房颤的检测模型

Bidirectional LSTM with Attention Mechanism for Detection of Atrial Fibrillation

心房颤动(Atrial Fibrillation,AF)是一种常见的心律失常疾病,会严重影响患者的日常生活,甚至引发包括中风、血栓堵塞等的并发症。因此,对房颤早期准确的诊疗非常重要。但是,算法对大规模的心率数据运行效率较低,因此房颤的诊断仍面临一定的挑战。针对上述挑战,文章提出了一种新型的基于深度学习的房颤检测架构BiLSTM-Attention。该架构包含双向长短期记忆网络(Bi-directional Long Short-Term Memory,BiLSTM)和Attention机制。在BiLSTM-Attention中,BiLSTM用于访问心率序列中的前项和后项数据,Attention机制用于给数据特征分配不同的权重,最终通过顶部全连接层分类房颤。在MIT-BIHAF数据库上对该架构进行了交叉验证,取得了98.54%的准确率。实验结果表明,BiLSTM-Attention架构在测试数据集上表现良好,为进一步探索智慧医疗迈出了坚实的一步。

Atrial fibrillation(AF) is a common disease of cardiac arrhythmia, which seriously affects daily life of patients.and may leads to other complications, such as stroke and thrombosis. Therefore, accurate and early diagnosis and treatment of AF is significant. However, due to the low efficiency of the algorithm for large-scale heart rate data, AF diagnosis faces difficult challenges. Therefore, this paper proposes a new architecture for AF detection based on deep learning, BiLSTM-Attention including Bi-bidirectional Long and Short Term Memory(BiLSTM) network and Attention mechanism. In BiLSTM-Attention,BiLSTM is used to access both the preceding and succeeding data in the heart rate sequence. The Attention mechanism is used to assign different weights to the data features. Finally, AF classification is carried out through the top full connection layer.Experiment was conducted on MIT-BIH AF database, and the architecture was cross-verified, and the accuracy was 98.54%,respectively. The experimental results show that the BiLSTM-Attention architecture performs well in the test data set, which makes a solid step for the further exploration of intelligent medical treatment.