[OPTICAL REVIEW Vol. 6, No. 4 (1999) 302-307]
Hierarchical Clustering Method for Extraction of Knowledge from a Large Amount of Data
Hidekazu NISHIZAWA, Takashi OBI, Masahiro YAMAGUCHI and Nagaaki OHYAMA*
Imaging Science and Engineering Laboratory, Tokyo Institute of Technology, 4259, Nagatsuta, Midori-ku, Yokohama, 226-8503 Japan
(Received October 23, 1998; Revised March 19, 1999; Accepted March 26, 1999)
The introduction of information systems in the medical field has made it possible to accumulate a large amount of health care examination data. Analysis of such data could yield valuable new knowledge about health and disease. In this paper, we propose a method for the analysis of large amounts of medical and health care data, especially images or signals. The proposed method treats data in a multidimensional space without any pre-processing, and the data is classified into groups according to the criterion. The criterion used in this paper is to maximize likelihood calculated from the probability density, which is given by the Parzen estimation method. The result of classification is expressed by a binary tree structure as a hierarchy of clusters. We applied this method to computer-generated data and practical electrocardiogram data, and the results showed its validity.
Key words : medical, knowledge, clustering, Parzen, ECG