Study On Artificial Neural Networks Method In Multianalysis Of Medical Data

The common limitions of some traditional statistics methods are: D require to know the arithmetic models of the objects beforehand, but they are always uncertain, changing and have complex nonlinearity. Some of them are known but difficult to describe with mathematics, some are unknown nonlinearity. This makes it difficult to modeling. require the variables to be normal and the covariates to be independent. Artificial neural network (ANN) is a nonlinear dynamic system, it is a information-deal-with system invoked by biology neural network for its structure, function and some basic characters, but being abstracted and simplified. ANN has distributed storage form and parallel disposing form of information. Introducing ANN into the analysis of medical dada, combining with statistics methods, and solving the problems above better are the main aim of the study. On the basis of study on ANN theory and realization method, simulation study on theoretical data and application study on factual data have been done. Results of simulation study indicate:ID the weights and biases of ADLLNE are equal or very near to the coefficients of GLM, the only difference is the method to acquire them. as GLM, ADL1NE can not overcome multi-collinearity, the ridge regression and incomplete principal component analysis can give reasonable results when multi-collinearity exists. the nonlinearity disposing ability of multi-layer network is not special to the nonlinear transfer functions of networks and the nonlinearity of data. Results of application study indicate: D for two category discriminant problem, if the transfer function and structure can be selected properly, ANN can give the same results as Logistic regression, but ANN with other transfer function may give better predictive results.the predictive results of multi-layer network is greater than Logistic regression and single-layer network on sensitivity, specificity and area under ROC curve. with CSR(Cox-snell residual) and x/3 as the predictive indexes of BP neural network and Cox proportional hazards regression respectively, pair-t-test of 20 pairs of concordance indexes indicates: BP neural network has higher predicative concordance. the result of multi-layer BP neural network to predict the recurrence time is ideally. Artificial neural network has powerflil ability of nonlinearity disposing, and has liffle limitions for the data, so should be popularized in multianalysis of medical data.