Hybrid fuzzy logic committee neural networks for classification in medical decision support systems

Recent emphasis on healthcare cost containment has created a growing need for developing automated diagnostic systems to aid the physician. Such systems should be reliable and robust. Neural networks have been used to provide advice to the physician. However, the reliability and robustness of a neural network depend on various factors such as the architecture, the initial weights used in training and the training data. In order to improve the reliability and robustness, we have developed hybrid intelligent systems with committee neural networks and fuzzy logic. Our concept of committee networks involves training and evaluating several networks and recruiting five networks into the committee based on initial testing. A majority opinion of the member networks of the committee is used to arrive at the final decision. The system reliability is improved since several networks in the committee are used in to make the decision. We have developed and evaluated hybrid intelligent systems with fuzzy logic and committee networks for distinguishing spondyloarthropathy of the knee from rheumatoid arthritis. Testing revealed that majority opinion of the committee well classified the arthritis.