Toward Normative Expert Systems: Part II Probability-Based Representations for Efficient Knowledge Acquisition and Inference

Abstract:

We address practical issues concerning the construction and use of decision-theoretic or normative expert systems for diagnosis. In particular, we examine Pathfinder, a normative expert system that assists surgical pathologists with the diagnosis of lymph-node diseases, and discuss the representation of dependencies among pieces of evidence within this system. We describe the belief network, a graphical representation of probabilistic dependencies. We see how Pathfinder uses a belief network to construct differential diagnosis efficiently, even when there are dependencies among pieces of evidence. In addition, we introduce an extension of the belief-network representation called a similarity network, a tool for constructing large and complex belief networks. The representation allows a user to construct independent belief networks for subsets of a given domain. A valid belief network for the entire domain can then be constructed from the individual belief networks. We also introduce the partition, a graphical representation that facilitates the assessment of probabilities associated with a belief network. We show that the similarity-network and partition representations made practical the construction of Pathfinder.

Key-Words

Expert Systems - Diagnosis - Probability Theory - Decision Theory - Artificial Intelligence - Belief Networks - Similarity Networks - Partitions - Pathology

• REFERENCES

• 1 Howard RA, Matheson JE. Influence diagrams. In: Howard RA, Matheson JE. eds. Readings on the Principles and Applications of Decision Analysis. Volume 2. Menlo Park: Strategic Decisions Group; 1981: 721-62.
• 2 Pearl J. Reverend Bayes on inference engines: A distributed hierarchical approach. In: Proceedings AAAI-82 Second National Conference on Artificial Intelligence. Pittsburgh. Menlo Park: AAAI Press; 1982: 133-6.
• 3 Heckerman DE, Horvitz EJ, Nathwani BN. eds. Pathfinder Research Directions . Technical Report KSL-89-64. Stanford: Computer Science Group, Section on Medical Informatics, Stanford University; 1985
• 4 Heckerman DE, Horvitz EJ, Nathwani BN. Update on the Pathfinder project. In: Proceedings of the Thirteenth Symposium on Computer Applications in Medical Care. Silver Spring: IEEE Computer Society Press; 1989: 203-7.
• 5 Heckerman DE. Probabilistic Similarity Networks . Cambridge: MIT Press; 1991
• 6 Good IJ. A causal calculus (I). Br J Philos Sc. 1961. 11: 305-18 Also in: Good IJ. ed. Good Thinking: The Foundations of Probability and its Applications. Minneapolis: University of Minnesota Press; 1983
• 7 Good IJ. A causal calculus (II). Br J Philos Sc. 1961. 12: 43-51 Also in: Good IJ. ed. Good Thinking: The Foundations of Probability and its Applications. Minneapolis: University of Minnesota Press; 1983
• 8 Rousseau WF. A Method for Computing Probabilities in Complex Situations. Technical Report 6252-2. Stanford: Center for Systems Research, Stanford University; 1968
• 9 Pearl J. ed. Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference. San Mateo: Morgan Kaufmann; 1988
• 10 Verma T, Pearl JH. Causal networks: Semantics and expressiveness. In: Proceedings of the Fourth Workshop on Uncertainty in Artificial Intelligence. Mountain View: Assoc for Uncertainty in Artificial Intelligence; 1988: 352-9 Also in: Shachter R, Levitt T, Kanal L, Lemmer J. eds. Uncertainty in Artificial Intelligence 4. New York: North-Holland Publ Comp; 1990: 69-76.
• 11 Geiger D, Pearl J. On the logic of causal models. In: Proceedings of the Fourth Workshop on Uncertainty in Artificial Intelligence. Mountain View: Assoc for Uncertainty in Artificial Intelligence; 1988: 136-47 Also in: Shachter R, Levitt T, Kanal L, Lemmer J. eds. Uncertainty in Artificial Intelligence 4. New York: North-Holland Publ Comp; 1990: 3-14.
• 12 Lauritzen SL, Spigelhalter DJ. Local computations with probabilities on graphical structures and their application to expert systems. J Royal Statist Soc B 1988; 50: 157-224.
• 13 Cooper GF. NESTOR: A Computer-based Medical Diagnostic Aid that Integrates Causal and Probabilistic Knowledge . Stanford: Medical Computer Science Group, Stanford University; 1984
• 14 Howard RA. Knowledge maps. Management Science 1989; 35: 903-22.
• 15 Schachter RD. Evaluating influence diagrams. Operations Research 1986; 34: 871-82.
• 16 Pearl J. Fusion, propagation, and structuring in belief networks. Artificial Intelligence 1986; 29: 241-88.
• 17 Henrion M. Propagation of uncertainty by probabilistic logic sampling in Bayes’ networks. In: Proceedings of the Second Workshop on Uncertainty in Artificial Intelligence. Mountain View: Assoc for Uncertainty in Artificial Intelligence; 1986. Also in: Shachter R, Levitt T, Kanal L, Lemmer J. eds. Uncertainty in Artificial Intelligence 2. New York: North-Holland Publ Comp; 1988: 149-64.
• 18 Cooper GF. Bayesian belief-network inference using recursive decomposition. Technical Report KSL-90-5. Stanford: Medical Computer Science Group, Section on Medical Informatics, Stanford University; 1990
• 19 Geiger D, Verma T, Pearl J. Identifying independence in Bayesian networks. Networks 1990; 20: 507-34.
• 20 Suermondt HJ, Cooper GF, Heckerman DE. A combination of cutset conditioning with clique-tree propagation in the pathfinder system. In: Proceedings of the Sixth Workshop on Uncertainty in Artificial Intelligence. Mountain View: Assoc for Uncertainty in Artif Intell; 1990: 273-9.
• 21 Heckerman DE. Formalizing heuristic methods for reasoning with uncertainty. Technical Report KSL-88-07. Stanford: Medical Computer Science Group, Section on Medical Informatics, Stanford University; 1987
• 22 Heckerman DE. Probabilistic similarity networks. Networks 1990; 20: 607-36.
• 23 Heckerman DE, Nathwani B. An evaluation of the diagnostic accuracy of Pathfinder. Comput Biomed Res. 1992. 25: 56-74.