基于注意力机制的堆叠LSTM心电预测算法

Anomaly Prediction in ECG Signals via Stacked Long Short-term Memory with Attention Mechanism

心电信号的处理一直是一个热门的研究领域,针对日趋增长的心电数据分析需求,国内的研究大多停留在分类处理阶段,对心电异常的预测相对较少。而心电异常的提前预警对预防接下来可能出现的危险极为重要,因此,提出了一种新的心电预测算法。首先,对原始数据进行小波变换处理,经过预处理后的数据能够更好地从中提取特征进行学习。将处理后的信号输入训练模型,在训练过程中采用两个LSTM网络结构,构成一种堆叠的循环神经网络模型。输出的信号再通过注意力机制,加强重点关注区域后经由全连接层输出结果。模型采用预测准确率作为衡量模型性能的指标,并在MIT-BIH数据集上进行了测试。经过实验数据的对比,该模型在此数据集上最终预测准确率为97.7%,与传统堆叠LSTM网络相比,提升了1.3百分点;与加入注意力机制的单层LSTM网络相比,提升了0.9百分点。结果表明,该模型有效地提高了预测的准确率,充分证明了其优越性。

ECG signal processing has always been a hot research field. For the growing demand of ECG data analysis, most domestic research stays in the stage of classification, and there is relatively little prediction of ECG abnormalities. The early warning of ECG abnormality is quite essential to prevent the coming possible danger. Therefore, a new ECG prediction algorithm is proposed. Firstly, the original data is processed by wavelet transform. The preprocessed data can better extract features from them for learning. The processed signal is then input into the training model. In the training process, the model composed of two LSTM networks is used to form a stacked cyclic neural network model. The output then outputs the results through the full connection layer after the attention mechanism which can enhance the focus area. The model uses the prediction accuracy as the index to measure the performance of the model, and is tested on MIT-BIH data set. Compared with the experimental data, the final prediction accuracy of the model on this data set is 97.7%,which is 1.3% higher than that of the traditional stacked LSTM network and is 0.9% higher than that of the single-layer LSTM network with attention mechanism. The results show that such model effectively improves the accuracy of prediction and fully proves its superiority.