Hidden Markov ModelÀ» ÀÌ¿ëÇÑ ¿¬¼Ó½ÉÀ½ºÐ·ù¿¡ °üÇÑ ¿¬±¸

±èÈñ±Ù1), Á¤¿ëÁÖ*1)

1)°è¸í´ëÇб³ ÀüÀÚ°øÇаú

Abstract : Artificial neural networks(ANNs) have been widely used for the automatic classification of heart sound signals. However, in a recent study, we found that the heart sound signal classification using hidden Markov models (HMMs) produced much better results than the artificial neural networks. In both of the classification methods using HMMs and ANNs, the heart sound signals are usually manually segmented to obtain one cycle of the signal for the stable classification performance. However, it is not easy to segment the continuous heart sound signals in real environments. Although there have been some research efforts for the automatic segmentation, the segmentation errors were irrecoverable and as a consequence resulted in performance degradation. In this paper, we propose to modify the conventional HMM structure into an ergodic form where every state can be reached from any states. The proposed HMM structure is shown to be able to recognize the heart sound signals without any segmentation information. In experiments classifying the continuous heart sound signals, the proposed method performed successfully with classification accuracy about 99%..
keyword : automatic classification, Hidden Markov Model, heart sounds signals.

¿¬Á¦ºÐ·ù


÷ºÎÇÑ ÆÄÀÏ

hmm06_Chung.hwp

÷ºÎÇÑ e-posterÆÄÀÏ

Åõ°í³»¿ëÀ» »èÁ¦ÇϽ÷Á¸é "»èÁ¦"¹öÆ°À» ´©¸£½Ã°í ´Ù½Ã Á¢¼öÇÏ¿© Áֽñ⠹ٶø´Ï´Ù.

 ¾Ï È£

´ëÇÑÀÇ·áÁ¤º¸ÇÐȸ
Copyright (c) 2002 by The Korean Society of Medical Informatics