Method for Visual Detection of Similarities in Medical Streaming Data

摘要

The analysis of medical streaming data is quite difficult when the problem is to estimate health-state situations in real time streaming data in accordance with the previously detected and estimated streaming data of various patients. This paper deals with the multivariate time series analysis seeking to compare the current situation (sample) with that in chronologically collected historical data and to find the subsequences of the multivariate time series most similar to the sample. A visual method for finding the best subsequences matching to the sample is proposed. Using this method, an investigator can consider the results of comparison of the sample and some subsequence of the series from the standpoint of several measures that may be supplementary to one another or may be contradictory among themselves. The advantage of the visual analysis of the data, presented on the plane, is that we can see not only the subsequence best matching to the sample (such a subsequence can be found in an automatic way), but also we can see the distribution of subsequences that are similar to the sample in accordance with different similarity measures. It allows us to evaluate differences among the subsequences and among the measures.