Hycones: a hybrid approach to designing decision support systems

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Medical Informatics

Background:

• Hybrid connectionist expert systems offer a powerful approach to complex classification problems.
• Integrating symbolic reasoning (frames) with neural networks (connectionism) can enhance system performance.
• Existing systems may not fully leverage the combined strengths of different neural network architectures for medical diagnostics.

Purpose of the Study:

• To introduce Hycones II, a tool for constructing hybrid connectionist expert systems.
• To evaluate the performance of three distinct neural network models within a hybrid framework for classification.
• To compare the efficacy of Combinatorial Neural Model (CNM), Fuzzy ARTMAP, and Semantic ART (SMART) models in medical diagnostic tasks.

Main Methods:

• Hycones II integrates a frame-based system with three neural network models: CNM, Fuzzy ARTMAP, and SMART (a CNM and Fuzzy ARTMAP combination).
• The study involves applying these hybrid models to solve diagnostic problems in two distinct medical domains.
• Performance comparison focuses on the accuracy and effectiveness of each model in classification.

Main Results:

• The comparative analysis will highlight the strengths and weaknesses of CNM, Fuzzy ARTMAP, and SMART in medical diagnosis.
• Results will indicate which hybrid model demonstrates superior performance for the selected diagnostic tasks.
• The study provides empirical data on the utility of different neural network architectures in a hybrid expert system.

Conclusions:

• Hycones II facilitates the development of sophisticated hybrid expert systems for medical classification.
• The choice of neural network model significantly impacts diagnostic problem-solving performance.
• The SMART model, combining CNM and Fuzzy ARTMAP, shows potential for enhanced diagnostic accuracy in hybrid systems.