Knowledge refinement of an expert system using a symbolic-connectionist approach
In this paper we have shown how to solve a real connectionist learning problem with a symbolic interpretation for the refinement of scoring systems and risk evaluation systems. This is a significant issue not easily manageable by classical symbolic methods  that are specially oriented to static domains. This has lead us to implement a symbolic-connectionist approach which combines the efficiency of connectionist learning with the comprehensibility of symbolic methods.
The network structure combines a linear part that does the same additions as physicians do to determine the weighted contributions of input variables and a non-linear transformation that determine the punctuation levels. The output units represent the scoring levels and they can be interpreted in a symbolic way. The system is currently validated with more real data, and it is being used to refine the scoring and risk evaluation systems of other hospitals by training the net with a set of local patients.
KeywordsMachine Learning Expert Systems in Medicine Knowledge Refinement Neural Networks
Unable to display preview. Download preview PDF.
- 1.P. Clark and T. Niblett. The CN2 induction algorithm. Machine Learning, 1988.Google Scholar
- 2.Edwall F.C., The knee society total knee arthroplasty roentgenographic evaluation and scoring system. Clinical Orthopaedics, (248):9–12, 1989.Google Scholar
- 3.L.M. Fu. Knowledge-based connectionism for revising domain theories. IEEE Transactions on Systems, Man and Cybernetics, 23(1):173–182, 1993.Google Scholar
- 4.J. Heras and R.P. Otero. TKR-tool: An expert system for total knee replacement management. Artificial Intelligence in Medicine Europe. Lecture Notes in Artificial Intelligence, (934):444–446, 1995.Google Scholar
- 5.Insall J.N., Dorr D.D., Scott R.D., and Scott W.N., Rationale of the knee society clinical rating system. Clinical Orthopaedics, (248):13–14, 1989.Google Scholar
- 6.R. P. Otero, D. Lorenzo, and P. Cabalar. Applying induction in temporal expert systems. IJCAI-Workshop on Data Engineering for Inductive Learning, 1995.Google Scholar
- 7.D.E Rumelhart, G.E. Hinton, and R.J. Willians. Learning internal representations by error propagation. Parallel Distributed Processing, Explorations in the Microestructure of Cognition, MIT Press, Cambridge, 1:318–362, 1986.Google Scholar
- 8.J. Santos, R.P. Otero, and J. Mira. Nettool: A hydrid connectionist-symbolic development environment. From Natural to Artificial Neural Computation, J. Mira and F. Sandoval (Eds.), Lecture Notes in Computer Science, (930):658–665, 1995.Google Scholar
- 9.G. Towell and J.W. Shavlik. Refining symbolic knowledge using neural networks. In Machine Learning, A Multistrategy Approach, volume 4, chapter 15, pages 405–429. Morgan Kauffman Publishers, 1994.Google Scholar