The added value of Artificial Organisms in the analysis of medical data: six years of experience

摘要

The authors introduce the notion of Artificial Organisms (AO), computerized systems coupling fuzzy logic, artificial neural networks and evolutionary algorithms with the aim to optimize the classifying capability of artificial neural networks in complex medical problems. This article analyze the power of AO as data mining tools applied on databases collected in several medical intervention domains, with the aim of verify their real effectiveness, compared with the regression models and the standard ANN models, in processing such kind of complex real-world data. These tools have been applied by the authors in independent applications on nineteen analyses carried out on thirteen databases collected in five main medical domains. The performance obtained by the ANNs coupled with AO shows a systematic significant increase in terms of the overall accuracy, compared with both the traditional statistic methods (Linear Discriminant Analysis and Logistic Regression) and the Standard ANNs. The added value of AO over standard ANNs consisted in an average increase in overall accuracy of about 10% (from 77.59% to 88.07% respectively), bringing to a 16 % difference versus classical statistics models. This increase in predictive capacity was obtained with an average 50% intelligent reduction of the input variables. The impact of AO approach in processing such kind of complex real-world data, is discussed emphasizing how the implementation of clinical diagnostic tools based on AO methods can be considered an added value for physicians in their clinical practice.