A connectionist approach to the recognition of trends in time-ordered medical parameters.
暂无分享,去创建一个
Recognition of trends that emerge over time is a capability that successful medical systems must possess. In this paper the network architecture introduced by Elman (CRL, Technical Report 9901, Center for Research in Language, University of California, San Diego, CA, 1988) for predicting successive elements of a sequence is reviewed. It is proposed that a similar architecture can be utilized in a medical domain. The simple example of a diabetic patient receiving a continuous insulin infusion, for whom only the current serum glucose and infusion status is known, is suggested as a test problem. Two networks are described, that when presented with the serum glucose and pump settings at time steps t and t + 1, are capable of predicting the serum glucose, and suggesting a pump setting at time t + 2. Possible applications, and the limitations of this model are then discussed.
[1] James L. McClelland,et al. Learning Subsequential Structure in Simple Recurrent Networks , 1988, NIPS.
[2] Terrence J. Sejnowski,et al. NETtalk: a parallel network that learns to read aloud , 1988 .
[3] M. G. Kahn,et al. Model-based interpretation of time-ordered medical data , 1989 .
[4] Jeffrey L. Elman,et al. Finding Structure in Time , 1990, Cogn. Sci..
[5] Geoffrey E. Hinton,et al. Learning representations by back-propagating errors , 1986, Nature.