Time Series Similarity Measurement And Its Application In Physiological Information Mining

In recent years, a mass of physiological information data are accessed because of the advance of computer processing and storage ability and the quick development of medical technology. It is realizable that we can analyze pathogenic factors, predict the trends disease changes, optimize the allocation of hospital resources, and effectively plan the medicine purchase, if these data are mined base on time series data mining technology and related medical knowledge, so it has great significance for developing medical research and new treatment technology, promoting hospital management level, and improving medical quality. For instance, in intensive care unit (ICU), a variety of vital sign parameters of critical patients are monitored continuously, such as ECG, blood pressure, blood oxygen, breathing, temperature and so on, so large amount of physiological data are produced. Moreover, more than one critical patient are cared by only one clinical doctor, so physiological state abrupt change of patients may not be observed instantly, then that lead to delay rescue time. So we can analysis patients'vital sign parameters and predict the trends of physiological state through data mining technology, then intelligent monitoring in ICU patients is realizable, and it will has important meaning for the development and perfection of clinical decision support system. Therefore, the following aspects of research are carried out in this paper:Firstly, the main present methods of time series similarity measurement are analyzed, including the characteristics of various methods and their application fields.Secondly, dynamic time warping (DTW) similarity measurement base on the extreme values and the vector space of statistical features similarity measurement base on symbolization are focused on and realized. At the same time, space vector model base on information retrieval model similarity measurement is putted forward and realized. And on account of the complex characteristics of trends presented by physiological time series, SYNDATA (The Control Chart Synthetic Dataset) is adopted to check and compare the performance of three algorithms. Then it is known that space vector model base on information retrieval model similarity measurement has higher recognition rate.Lastly, a small physical state sample database is set up, where the physiological information in ICU patients was accessed by using MIMIC-Ⅱ database. The standards of patients'physiological state are established, according to the basic pathophysiology knowledge and the availability of clinical information, and we try to predict physiological stable states in ICU patients, drawing lessons from internet search principle. The method is that the test physiological time series as the keywords are used to search for the related similar time series in sample database, then the search results are used to judge whether the test time series are stable or not. Thus the prediction of physiological stable state is realized. Space vector model base on information retrieval model similarity measurement put across in the above parts is adopted. The prediction results are evaluated according to the patients'actual physiological state and medical records in MIMIC-II database.