基于LightGBM的心电信号分类算法

ECG Signal’s Classification Algorithm Based on LightGBM

心电图(Electrocardiogram, ECG)被广泛应用于窦性心动过速、室性早搏和心房颤动等心律失常诊断中,进而在心脏疾病诊断分析方面展现巨大的临床应用价值。为提升计算机辅助诊断心电图的分类效果,本文提出一种基于LightGBM的心电信号分类算法。该算法从心电图中提取单心拍特征、心律波动性特征以及全波形特征建立混合特征集,并采用LightGBM实现正常心拍、心房颤动、其他心律不齐、噪声四个类别的分类。最终该算法的性能指标在PhysioNet/CinC Challenge数据集上达到0.824,优于CART和CatBoost算法。同时为了加快心电图特征提取的速度,本文根据特征重要性筛选关键特征来减少分类所需的特征数量,在保持分类性能的同时将特征提取时间降为原来的17.8%。

Electrocardiogram (ECG) has been widely used in the diagnosis of arrhythmia such as sinus tachycardia, ventricular premature beats and atrial fibrillation, and has shown great clinical application value in the diagnosis and analysis of heart diseases. In order to improve the classification performance of ECG computer-aided diagnosis, an ECG signal classification algorithm based on LightGBM was proposed. The algorithm extracts single heart beat features, heart rhythm volatility features and full waveform features from ECGs to establish a mixed feature set, and uses LightGBM to classify four categories: normal heart beat, atrial fibrillation, other arrhythmia, and noise. Finally, the performance parameter of the algorithm reached 0.824 on PhysioNet/CinC Challenge dataset, which was better than CART and CatBoost. At the same time, in order to accelerate the speed of ECG feature extraction, this paper screens the key features according to the importance of features to reduce the number of features required for classification, and the feature extraction time is reduced to 17.8% while maintaining the performance of classification.