Interactive statistical system and method for predicting expert decisions

• US 4,872,122 A
• Filed: 06/19/1987
• Issued: 10/03/1989
• Est. Priority Date: 06/19/1987
• Status: Expired due to Fees

• Current Assignee: University of Pennsylvania
• Original Assignee: University of Pennsylvania
• Inventors: Whittington, Richard, Altschuler, Martin D.
• Primary Examiner(s): Lall, Parshotam S.
• Assistant Examiner(s): Teska, Kevin J.
• Application Number:
  US07/065,304
• Time in Patent Office: 837 Days
• Field of Search: 364/300, 364/513, 364/415, 364/554, 364/413.02, 364/200
• US Class Current: 356/432
• CPC Class Codes: G06N 5/04 Inference or reasoning models
• CPC Class Codes: G09B 7/04 characterised by modifying the teaching programme in response to a wrong answer, e.g. repeating the question, supplying a further explanation
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems
• CPC Class Codes: Y10S 706/924 Medical

Abstract

A system and method for selecting from a set of output actions for combinatory situations such as medical diagnosis defined by a plurality of input parameters. A decision making structure is established by an expert having a plurality of linked nodes. Random values of the input parameters are generated, which includes biasing the random values by a function of a preceding response. The steps are repeated until a sufficient number of responses having a predetermined statistical significance is achieved for each node. After this set up and teaching, the accuracy of the system'"'"'s predictions may be verified against actual responses by the expert. The decision making structure may be interrogated by entering a situation to be analyzed. The path the expert would take through the decision making structure is predicted by determining the probability of the expert'"'"'s response at each node in the structure to arrive at an output action. A plurality of structures may be established, one for each individual expert, allowing interrogation of any or all of the decision making structures.

Claims

1. A method for generating an expert system comprising the steps of:

(a) establishing a set of input parameters having input values which define a case situation and establishing a set of output actions in which the relationship between each input parameter and each output action defines a combinatory situation;

(b) establishing a decision making structure formed of linked nodes with specified nodes defining questions;

(c) generating random values to serve as input values defining a first case situation for a first node;

(d) responding by the expert to the questions posed at the first node;

(e) generating a new set of random values to serve as new input values defining a new case situation for the first node;

(f) repeating steps (d) and (e) until a sufficient number of responses having a predetermined statistical significance is achieved for the first node;

(g) generating in turn for each node subsequent to the first node new random values defining a further new case situation biased by a function of a response from a preceding node;

(h) responding by the expert to the questions posed at each subsequent node;

(i) generating a new set of random values as in step (g);

(j) repeating steps (h) and (i) until a sufficient of responses having a predetermined statistical significance is achieved for each of the subsequent nodes.

2. The method of claim 1 in which step (g) includes generating random values which are consistent with the preceding responses.

3. The method of claim 1 in which step (g) includes generating non-biased random values for a first node in the linked node structure.

4. The method of claim 1 in which step (c) includes grouping the random values into a set of values defined as a simulated situation to be responded to in step (d).

5. The method of claim 1 in which step (d) includes responding to the nodes based upon the random values of the input parameters.

6. The method of claim 1 in which there is provided the further step of generating one set of random values, responding to the set of random values at each node, independently predicting the response to the random values of the input parameters based on step (e), and comparing the response to the set of random values to the predicted response.

7. The method of claim 1 in which step (b) includes establishing a plurality of decision making strctures each for an individual expert and in which steps (c)-(j) follow for each expert'"'"'s decision making structure.

8. The method of claim 1 in which the situation is a medical condition of a patient, the input parameters are physical conditions of the patient and the output actions are possible treatments in response to the medical condition.

9. The method of claim 8 in which there is provided the further step of comparing the output actions of each decision making structure.

10. The method of claim 1 in which step (g) includes generating a biased random value by establishing chi square significance.

11. The method of claim 10 in which there is provided the further step of generating a table of Bayesian probabilities to predict the response for any subsequent situation based on the responses in step (e).

12. The method of claim 11 in which there is provided a further step of interrogating the decision making structure by entering a situation to be analyzed, predicting which path the expert would take through the decision making structure and determining the probability of the expert'"'"'s response at each node in the decision making structure to arrive at an output action.

13. The method of claim 12 in which there is provided the further step of determining for each group of output actions the probability of selection by the expert of those output actions.

14. A method for selecting from a set of output actions, designated by an expert, for combinatory situations defined by a plurality of input parameters having input values which define a case situation comprising the steps of:

(a) establishing a decision making structure having a plurality of linked nodes extending from a first node and ending in a terminal node with each node defining a question;

(b) entering the input values of the input parameters defining a case situation to be analyzed;

(c) determining which input values of the case situation are significant from node to node and predicting which responses the expert would make for the question posed at the first node based on the values of the significant parameters and then predicting the response for the questions posed at each subsequent node, said subsequent nodes forming a path based on the previous responses through the decision making structure;

(d) repeating step (c) through the linkages of the nodes until a predicted response is an output action at a terminal node of the decision-making structure; and

(e) determining the probability that the expert would have selected the output action predicted in step (d).

15. The method of claim 14 in which step (d) further includes determining at a terminal node for all output actions the probability of selection by the expert of each output action from the set of output actions.

16. The method of claim 14 in which steps (b) and (c) include generating the probability that a given path through the linked nodes of the decision-making structure would be chosen by the expert in determining the selection of the output action.

17. A system for selecting from a set of output actions, designated by an expert, for combinatory situations defined by a plurality of input parameters having input values which define a case situation, comprising

means for establishing a decision making structure having a plurality of linked nodes extending from a first node and ending in a terminal node with each node defining a question;

means for entering input values of the input parameters defining a case situation to be analyzed;

means for determining which input values of the case situation are significant from node to node and means for predicting which responses the expert would make for the question posed at the first node based on the values of the significant parameters;

means for predicting the response for the questions posed at each subsequent node, said subsequent nodes forming a path based on the previous responses through the decision making structure;

means for determining which input values of the case situation are significant from node to node and predicting which responses the expert would make for the question posed at each of the nodes subsequent to said first node based on the values of the significant parameters until a predicted response is an output action at a terminal node of the decision making structure;

means for determining the probability that the expert would have selected the output action predicted.

18. The system of claim 17 in which there is further provided

means for interrogating the decision making structure by entering a situation to be analyzed;

means for predicting which path the expert would take through the decision making structure; and

,

means for determining the probability of the expert'"'"'s response at each node in the decision making structure to arrive at an output action.