1 / 36

SWENG 597C Mid-Term: A Neural Expert System with Automated Extraction of Fuzzy If-Then Rules and Its Application to Med

SWENG 597C Mid-Term: A Neural Expert System with Automated Extraction of Fuzzy If-Then Rules and Its Application to Medical Diagnosis. By Matt Hallahan. Overview . Introduction Application to Medical Diagnosis Fuzzy Neural Expert System with Automated Extraction of Fuzzy If-Then Rules

jada
Download Presentation

SWENG 597C Mid-Term: A Neural Expert System with Automated Extraction of Fuzzy If-Then Rules and Its Application to Med

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. SWENG 597C Mid-Term: A Neural Expert System with Automated Extraction of Fuzzy If-Then Rules and Its Application to Medical Diagnosis By Matt Hallahan Matt Hallahan

  2. Overview • Introduction • Application to Medical Diagnosis • Fuzzy Neural Expert System with Automated Extraction of Fuzzy If-Then Rules • Automated Extraction of Fuzzy If-Then Rules from Trained Neural Networks Matt Hallahan

  3. Introduction • FNES – fuzzy-neural expert system • A new method for giving confidence measurements for all inferences and explanations to neural expert systems • Purpose: To compare diagnostic capabilities of neural network approach to static approach and evaluate performance of fuzzy If-then rules from a neural-network knowledge base Matt Hallahan

  4. Application to Medical Diagnosis • Used a real medical database containing the results of nine biochemical tests involving the likes of primary hepatoma, liver cirrhosis, and hepatobiliary disorders • 536 patients clinically admitted to university-based hospital randomly assigned data; 373 given training data and 163 given test data Matt Hallahan

  5. Application to Medical Diagnosis • Fuzzy cell groups with three input cells used to represent degree of abnormality of each biochemical item and four output cells for the hepatobiliary disorders • 50,000 iterations of Pocket Algorithm performed for each output cell Matt Hallahan

  6. Application to Medical Diagnosis • Fuzzy neural network correctly diagnosed 75% of test data and 100% of training data; whereas linear discriminant analysis correctly diagnosed 65% of test data and 68.4% of training data • 48 general fuzzy If-then rules extracted – 12 rules for confirming diseases and 36 rules for excluding diseases; total accuracy of rules for confirming diseases was 87.7% Matt Hallahan

  7. Fuzzy Neural Expert System with Automated Extraction of Fuzzy If-Then Rules • Distributed Neural Network • Fuzzy Neural Network Matt Hallahan

  8. Distributed Neural Network • Figure 1 – a schematic diagram of a fuzzy neural system with automated extraction of fuzzy IF-THEN rules Matt Hallahan

  9. Distributed Neural Network • Backpropagation: p input cells, q intermediate cells, r output cells; connections run from every input cell to every intermediate cell and from every intermediate cell to every output cell • Same cells and connections ala backpropagation, plus direct connections from input to output cells • Each connection has weight wij that roughly corresponds to the influence of cell uj on cell ui • Weights of connections from input layer to intermediate layer generated using random number generator • Cell activations discrete, taking values of +1, 0, or -1 Matt Hallahan

  10. Distributed Neural Network • Value of cell Io is always +1 and is connected to every other cell except for input cells • A set of equations used to activate the input cells Ii(i = 1, 2, . . ., p), intermediate cells Hj (j = 1, 2, . . ., q), and the output cell Ok (k = 1, 2,. . ., r) Matt Hallahan

  11. Distributed Neural Network Matt Hallahan

  12. Distributed Neural Network • Figure 2 – a distributed neural network Matt Hallahan

  13. Fuzzy Neural Network • Need to interpret subjective input data with non-Boolean quantitative/qualitative meaning to handle various fuzziness in input layer of the distributed neural network • Fuzziness can be incorporated into training data by using only Boolean inputs; allowing them to be processed by Pocket Algorithm Matt Hallahan

  14. Fuzzy Neural Network • Figure 3 – a neural network with fuzzy cell groups and crisp cell groups Matt Hallahan

  15. Fuzzy Neural Network • Truthfulness of fuzzy information and crisp information represented by fuzzy cell groups and crisp cell groups respectively • Fuzzy cell groups consist of m input cells which have the level set representation using binary m-dimensional vector, each taking values in {+1,-1} • Crisp cell groups consist of m input cells which take on two values {(+1, +1, . . ., +1), (-1, -1, . . .,-1)} Matt Hallahan

  16. Automated Extraction of Fuzzy If-Then Rules from Trained Neural Networks • Providing linguistic relative importance for each proposition makes each fuzzy If-then rule have more flexible expression, facilitating automated extraction of fuzzy If-then rules from a trained neural network and enhancement of presentation capability and flexibility Matt Hallahan

  17. Automated If-Then Rule Extraction Algorithm • Step 1: Extraction of framework of fuzzy If-then rules • select propositions in an antecedent (If) of a rule Matt Hallahan

  18. Automated If-Then Rule Extraction Algorithm • Step 2: Assignment of linguistic truth value to each extracted rule • Each fuzzy If-then rule selected in step 1 is given a linguistic truth value determined by relative amount of weighted sum of output cells and used to indicate the degree of certainty to drawing a conclusion Matt Hallahan

  19. Automated If-Then Rule Extraction Algorithm • Step 3: Assignment of linguistic relative importance to each proposition • Linguistic relative importance assigned to each proposition of antecedent in fuzzy If-then rules and represents the degree of effect of each proposition on consequence Matt Hallahan

  20. Algorithm to extract framework of fuzzy If-Then rules • Extraction of dispensable propositions on cell groups in an antecedent is required for the extraction of framework of fuzzy If-then rules • Assume each cell consists of three input cells • Fuzzy cell group takes on three values in {(+1,-1,-1),(+1,+1,-1),(+1,+1,+1)} • Crisp cell group takes on two values in {(+1,+1,+1),(-1,-1,-1)} Matt Hallahan

  21. Algorithm to extract framework of fuzzy If-Then rules • Activations of intermediate cell Hj and output cell Ok Matt Hallahan

  22. Algorithm to extract framework of fuzzy If-Then rules • Step I: Select one output cell Ok Matt Hallahan

  23. Algorithm to extract framework of fuzzy If-Then rules • Step II: Select one cell group; if selected cell group is a fuzzy cell group, set values of cell group in (+1,-1,-1) / (+1,+1,-1) / (+1,+1,+1,); if selected cell group is crisp cell group, set the values of cell group in (+1,+1,+1) / (-1,-1,-1); set value of cell groups which were not selected to (0,0,0) Matt Hallahan

  24. Algorithm to extract framework of fuzzy If-Then rules • Step III: Forward search - Determine all values of intermediate cells Hj by using values of cell groups from Step II and the Hj activation conditions; determine value of output cell Ok using Ok activation conditions; if value of Ok is +1/-1, go to Step V; if value of Ok is 0, go to Step VI Matt Hallahan

  25. Algorithm to extract framework of fuzzy If-Then rules • Step IV: Backward search - Let v* be max value of |vkj| (absolute value of the weight of the connections between the output cell Ok & intermediate cell Hj, whose activate value is 0); let u* be max value of |uki| (absolute value of the weight of the connections between output cell Ok & input cell Ii, whose activate value is 0); if u* >= v* or values of all intermediate cells are determined, go to Step IV-1; otherwise, go to Step IV-2 Matt Hallahan

  26. Algorithm to extract framework of fuzzy If-Then rules • Step IV-1: For input cell Ii, which is incident to uki (|uki| = u*); if input cell Ii is included in fuzzy cell group, go to Step IV-1-F; if input cell Ii is included in crisp cell group, go to Step IV-1-C Matt Hallahan

  27. Algorithm to extract framework of fuzzy If-Then rules • Step IV-1-F: If SOk >= 0, select one pattern of the fuzzy cell group which has the max value of SOk among the cell values; if SOk < 0, select one pattern which has min value of SOk; go to Step V Matt Hallahan

  28. Algorithm to extract framework of fuzzy If-Then rules • Step IV-1-C: If SOk >= 0, select one pattern of crisp cell group which has max value of SOk among the cell values; if SOk < 0, select one pattern which has min value of SOk; go to Step V Matt Hallahan

  29. Algorithm to extract framework of fuzzy If-Then rules • Step IV-2: Let w* be max value of |wji| (absolute value of the weight of connections between intermediate cell Hj adjacent to vkj (|vkj| = v*)); if input cell Ii is included in fuzzy cell group, go to Step IV-2-F; in crisp cell group, go to Step IV-2-C Matt Hallahan

  30. Algorithm to extract framework of fuzzy If-Then rules • Step IV-2-F: If SHj >= 0, select one pattern of fuzzy cell group which has max value of SHj among the cell values; if SHj < 0, select one pattern which has min value of SHj; go to Step V Matt Hallahan

  31. Algorithm to extract framework of fuzzy If-Then rules • Step IV-2-C: If SHj >= 0 select one pattern of the crisp cell group which has the max value of SHj in the cell values; if SHj < 0, select one pattern which has min value of SHj; go to Step V Matt Hallahan

  32. Algorithm to extract framework of fuzzy If-Then rules • Step V: Extraction of framework of If-then Rules - if value of Ok is determined, extract input items corresponding to a determined cell group as proposition in an antecedent; if value of Ok is +1, consequence is set to "Ok is True"; if value of Ok is -1, consequence set to "Ok is False"; if multiple frameworks of If-then rules with same antecedent & consequence extracted, adopt one of them Matt Hallahan

  33. Algorithm to extract framework of fuzzy If-Then rules • Step VI: Termination condition of extraction algorithm for each output cell - for output cell Ok, if there are any cell groups which not selected yet, or for selected cell groups there are any patterns which are not selected yet, go to Step II; otherwise, go to Step VII Matt Hallahan

  34. Algorithm to extract framework of fuzzy If-Then rules • Step VII: Termination condition of whole extraction algorithm - repeat Steps II-VI until termination condition of extraction algorithm for each output cell satisfied; if there are any output cell Ok which not selected yet in Step I, go to Step I; otherwise, stop whole extraction algorithm Matt Hallahan

  35. Conclusion • Lots of potential for the neural network knowledge base approach, for generating practical knowledge bases from various databases among other things • With enhancement of interpretation capability of real data and embodiment of implicit/subjective knowledge, significant reduction of man power for knowledge acquisition in expert system development possible Matt Hallahan

  36. References • Hayashi, Yoichi. A Neural Expert System with Automated Extraction of Fuzzy If-Then Rules and its Application to Medical Diagnosis. In: Proceedings of the 1990 conference on Advances in neural information processing systems 3; 1990 Nov 26-29; Denver, CO. San Francisco, (CA): Morgan Kaufmann Publishers Inc.,1990 p. 578-584. Matt Hallahan

More Related