Evolutionary fuzzy modeling human diagnostic decisions

Area of Science:

• Artificial Intelligence
• Computational Biology
• Medical Informatics

Background:

• Fuzzy logic offers a framework for interpretable yet accurate computational systems.
• Constructing fuzzy systems (fuzzy modeling) is complex due to numerous parameters.
• Evolutionary computation, specifically cooperative coevolution, excels at optimizing complex search spaces.

Purpose of the Study:

• To introduce Fuzzy CoCo, a methodology combining fuzzy logic and evolutionary computation.
• To develop systems for accurately predicting human decision-making outcomes with understandable reasoning.
• To apply Fuzzy CoCo to breast cancer diagnostic problems for improved accuracy and interpretability.

Main Methods:

• Utilized cooperative coevolution, a type of evolutionary computation, for fuzzy system parameter identification.
• Applied the Fuzzy CoCo methodology to two breast cancer diagnostic datasets: WBCD and Catalonia mammography interpretation.
• Developed a web-based tool (COBRA) integrating an evolved Fuzzy CoCo system for radiologist assistance.

Main Results:

• Achieved high-performance and high-interpretability fuzzy systems for breast cancer diagnosis.
• Successfully modeled decision processes in both the WBCD and Catalonia mammography interpretation problems.
• Integrated an evolved system into COBRA, demonstrating practical application in aiding radiologists.

Conclusions:

• Fuzzy CoCo effectively constructs accurate and interpretable systems for complex decision-making processes.
• The methodology shows significant promise in enhancing medical diagnostic tools, particularly in mammography interpretation.
• The integration of Fuzzy CoCo into tools like COBRA can support clinical decision support systems.