| Autor: |
Arakawa T; Department of Mechanical Systems Engineering, Aichi University of Technology, Gamagori, Aichi, 443-0047, Japan. arakawa-toshiya@aut.ac.jp., Tanave A; Transdisciplinary Research Integration Center, Research Organization of Information and Systems, Minato-ku, Tokyo, 105-0001, Japan.; Mouse Genomics Resource Laboratory, National Institute of Genetics (NIG), Mishima, Shizuoka, 411-8540, Japan., Takahashi A; Mouse Genomics Resource Laboratory, National Institute of Genetics (NIG), Mishima, Shizuoka, 411-8540, Japan.; Laboratory of Behavioral Neuroendocrinology, University of Tsukuba, Tsukuba, Ibaraki, 305-8572, Japan., Kakihara S; Formerly National Graduate, Institute for Policy Studies, Minato-ku, Tokyo, 106-8677, Japan., Koide T; Transdisciplinary Research Integration Center, Research Organization of Information and Systems, Minato-ku, Tokyo, 105-0001, Japan.; Mouse Genomics Resource Laboratory, National Institute of Genetics (NIG), Mishima, Shizuoka, 411-8540, Japan.; Department of Genetics, SOKENDAI, Mishima, Shizuoka, 411-8540, Japan., Tsuchiya T; National Graduate Institute for Policy Studies, Minato-ku, Tokyo, 106-8677, Japan.; The Institute of Statistical Mathematics, Tachikawa, Tokyo, 190-8562, Japan. |
| Abstrakt: |
Recent innovations in sensing and Information and Communication Technology (ICT) have enabled researchers in animal behavior to collect an enormous amount of data. Consequently, the development of an automated system to substitute for some of the observations and analyses that are performed currently by expert researchers is becoming a crucial issue so that the vast amount of accumulated data can be processed efficiently. For this purpose, we introduce a process for the automated classification of the social interactive status of two mice in a square field on the basis of a Hidden Markov model (HMM). We developed two models: one for the classification of two states, namely, indifference and interaction, and the other for three states, namely, indifference, sniffing, and following. The HMM was trained with data from 50 pairs of mice as provided by expert human observers. We measured the performance of the HMM by determining its rate of concordance with human observation. We found that sniffing behavior was segmented well by the HMM; however, following behavior was not segmented well by the HMM in terms of percentage concordance. We also developed software called DuoMouse, an automated system for the classification of social interactive behavior of mice, that was based on the HMM. Finally, we compared two implementations of the HMM that were based on a histogram and a Gaussian mixture model. |
