Shorten G P, Burke M J
Department of Electronic and Electrical Engineering, Trinity College Dublin, Dublin 2, Ireland.
Annu Int Conf IEEE Eng Med Biol Soc. 2011;2011:4980-3. doi: 10.1109/IEMBS.2011.6091234.
Pattern recognition, and in particular dynamic time warping has been applied to the ECG for many different purposes over the last decade. Significant research on creating adaptive, feature based, and more complex forms of the algorithm in order to increase its ability to classify an ECG signal accurately has been performed. Despite this increase in complexity and in the number of variations of the dynamic time warping algorithm there has been less focus on actually using the results of dynamic time warping to relate the reference and test signals to each other as accurately as possible. The majority of dynamic time warping algorithms published in the literature, even the most complex of them, classify the most accurate match to a reference signal based only on resulting Euclidean distance or slope difference between samples of the known reference and unknown query signal. This article demonstrates how a combination of measurements including heart-rate, amplitude and required warping time alignment can be used to reduce the resulting error in the classification of a query signal after the query and reference signals have been warped together. Its benefits are verified with significant testing.
在过去十年中，模式识别，尤其是动态时间规整，已被应用于心电图领域以实现多种不同目的。为了提高其准确分类心电图信号的能力，人们对创建自适应、基于特征且更复杂形式的算法进行了大量研究。尽管动态时间规整算法的复杂度有所增加且变体数量增多，但较少关注如何实际利用动态时间规整的结果，尽可能准确地将参考信号和测试信号相互关联起来。文献中发表的大多数动态时间规整算法，即使是最复杂的算法，也只是基于已知参考信号和未知查询信号样本之间的欧几里得距离或斜率差异，来对与参考信号的最准确匹配进行分类。本文展示了在查询信号和参考信号进行规整之后，如何结合心率、幅度和所需的规整时间对齐等测量方法，来减少查询信号分类中的误差。通过大量测试验证了其优势。
