The analysis of intracardiac signals, or electrograms (EGM), is one of the most promising tools to guide catheter ablation of atrial fibrillation and improve the success of this procedure. Given the nonlinear nature of EGM signals, several studies have conducted fractal and multifractal analyses to extract nonlinear features related with critical activity. However, the fractal exponent or the multifractal spectrum fail to provide information about the temporal behavior of these signals. To overcome this limitation, the Time-Singularity Multifractal Spectrum Distribution (TS-MFSD) was recently introduced. In this paper, a feature extraction scheme is proposed to compute descriptors from the TS-MFSD that could be used in a classification scheme. The results show that features extracted from the TS-MFSD would serve to classify EGM signals into four classes depending on their level of fragmentation. In addition, the $k$-Nearest Neighbors and Support Vector Machines classifiers employed in this study, along with the optimal feature subset, achieved an accuracy of $81.37\pm 0.95$% and $83.35\pm 1.04$%, respectively. This finding is comparable with those of other works that have used features based on the morphology of local activation waves and amplitude thresholds.

心内信号或电描记图（EGM）的分析是指导心房纤颤导管消融并提高该过程成功率的最有前途的工具之一。考虑到EGM信号的非线性特性，一些研究已经进行了分形和多重分形分析，以提取与临界活动有关的非线性特征。但是，分形指数或多重分形频谱无法提供有关这些信号的时间行为的信息。为了克服此限制，最近引入了时间奇异多重分形频谱分布（TS-MFSD）。在本文中，提出了一种特征提取方案来从TS-MFSD计算描述符，该描述符可用于分类方案。结果表明，从TS-MFSD中提取的特征将根据其碎片级别将EGM信号分为四类。除此之外$ķ$-本研究中使用的最近邻和支持向量机分类器以及最佳特征子集实现了 $81.37\pm 0.95$％和 $83.35\pm 1.04$％， 分别。这一发现与其他使用基于局部激活波的形态和幅度阈值的特征的作品的发现是可比的。