A Clustering Analysis Strategy For On-line Dynamic Electrocardiogram Waveform

Through developed in the last 40 years, Ambulatory Electrocardiogram has become one of most common detection tools in the field of electrocardiogram non-invasive detection. The most valuable application of Ambulatory Electrocardiogram is to detect the abnormal waveforms. These abnormal waveforms are hardly to be detected in clinical diagnosis because they occur with low probability. However, Ambulatory Electrocardiogram records as long as 24 hours heart data about 100,000 waveforms. Just because of the massive data it sampled, the real-time analysis and detection is a so hard work. How to find out the typical waveforms from massive data rapidly, accurately and reliably has become a research hotspot.Through analysing the waveform characteristics of Ambulatory Electrocardiogram, This paper makes a thorough study of Feature Extraction and clustering and then proposes a new on-line real-time clustering strategy.After achieving the Ambulatory Electrocardiogram, the traditional analysis methods directly process the high-dimensional vectors. However, in the Ambulatory Electrocardiogram Analysis, these traditional methods are not suitable because of the huge data and high-dimensional vectors. Through contrasting the several algorithms of reduction of dimension, this paper chooses Sanger algorithm to extract the low-dimensional vectors from complete and high-dimensional data. In the experiment, it has been proved that the clustering results used the low-dimensional vectors are close to the results used the original high-dimensional vectors and the accuracy loss can be accepted in the medical specialty.How to improve the accuracy of the Electrocardiogram clustering is the spot in this paper. The improved clustering algorithm based on simulating annealing is used in Ambulatory Electrocardiogram analysis. In the experiment, it has been proved that the new algorithm is more accurate than traditional algrithms, so this algorithm can be used appropriately in new analyisis strategy.Based on the Sanger algorithm and improved clustering algorithm, a new on-line in-time Clustering Analysis Strategy for On-line Ambulatory Electrocardiogram is proposed in this paper. This strategy achieves the low-dimensional vectors of Ambulatory Electrocardiogram by neural network with litttle loss of accuracy at first. low-dimensional vectors can reduce the time-comsuming. And then the clustering algorithm, improved by simulate annealing algorithm, has been used to analyse in low-dimensional vectors. The experiment results has prove that the new strategy has more accurate. This algorithm can be embedded in the Electrocardiogram sample devices and become a part of the pre-processing, and makes the real-time analysis and detection possible.