Application Of Back-Propagation Network Based On Genetic Algorithm In Medical Diagnosis

Along with exaltation of social progress and people's material culture living level, the civil health consciousness also is strengthened. The Nation value the citizen's healthy level highly, continuously perfect medical treatment guarantees system. There are a lot of talents in Jilin University, the healthy problem of the talents seem to be particularly important. In order to guarantee the health of students and the teaching staff, Jilin University signs the item "healthy engineering". "The University hospital management and the disease early-warning system" are under the guide of "healthy engineering", is aim at the special community, the educators and students in university.Because there are dissimilarities in public healthy cognition degree, cause some crowds who are placed in second healthy appearance but can not adopt valid measure to make he recover healthy appearance in time depravations of the condition, particularly at University, the teachers are corpus, their health will directly influence the teaching quality of University. Among them, some older teachers just concentrate on study business, devote their mind to teach but ignore their healthy appearance and make themselves placed in second healthy appearance but don't know, not only delay the treatment, but also bring loss for the nation, University and the families; Not only older teachers such, now some young teachers make themselves place in second healthy appearance but know nothing, just discover the disease at the time which the disease outbreak, make loss in material resources and financial powers. In diagnosis and decision of the medical treatment in recent years, rose to handle a great deal complicated medical information of connection with Artificial Neural Network, then analysis, reason logically, classify, predict etc. from it.One of the Artificial Intelligence Technology-Artificial Neural Network (ANN for short) is adopted by the author. ANN is a non-linear dynamic system that is rather complicated. It is made of large amount of, at the same time quite simple, processing units that connected together in some structure. ANN reflects some characteristic of man's mind. But the function of mind is simulated more simply. The author uses BP algorithm, one of classic algorithms in ANN to design the system of the disease diagnosis model which is carried out by program on traditional computer system than ANN computer.At the realm of medical science, the second index is complex, the relation of the terminal index and the second index is not linear mostly, because of restriction of funds and research time, the tests are mostly difficult to observe the clinical terminal point, in the meantime the second index information source is neither complete nor implying appearance, and usually meet indeterminate information, if the second index and the terminal index are not linear, then can hardly predict the terminal index through the second index, even predict and usually make the analytical result be conflicting or can follow without the reason, this is because the traditional covariance analysis adopt empirical formula or Mathematical Statistics method.It is actually a problem of system identify that predict the terminal index through the second index, identify relation of the system exportation (the second index) and the importation (the terminal index). The ANN is good at withdrawing the relation of the system exportation and the importation from distortion, damage and the complicated background information, so this research hopes to make use of ANN, build up predicting model, and expect to provide a kind of viable diagnosis method through medical diagnosis.The chapter 1 introduced writing background and purpose of this thesis and involve of item. The chapter 2 introduced basic concept and theories of ANN, and research on medical knowledge. Follow on the chapter 3 introduced BP algorithm, one of classic algorithms in ANN. The chapter 4 applies BP algorithm in the disease diagnosis. We take the coronary disease for the example, concretely introduce and analyzed the coronary disease. We give process of the experiment and the result analysis.Each coin has two sides, BP algorithm itself has some deficiencies: 1 Lower speed of convergence of BP algorithm makes thousands of iterative times during a training process.2.From the point of mathematic, it is a non-linear gradient optimized process, so it is inevitable the ANN will be trapped in local minimum easily during the optimized process, and then hard to find best global result when the BP algorithm is used;3.Now there is no fixed theory to set the number of hidden layers of ANN and number of neural nodes of each hidden layer, sometimes it is all by experience .According to this, the author introduces a improved algorithm of the disease diagnosis model in chapter 5, uses the Generation Algorithm (GA for short) to optimize some parameters of ANN. GA is a highly parallel random, self-adapted searching method which simulates natural selection and natural heredity of biology. It has connotative parallel ability and the power of effective utilization, which can reflect the large space in searching set only need searching less structure. GA can find the best result make use of the fitness information of the population and selection crossover and mutation operator, so it can promote the function of the disease diagnosis model.The experiment result of the chapter 5 compare with the one of the chapter 4, diagnose diagnostic descend slightly, but both diagnostic speed and correctness have a very obvious improvement. This also explained diagnostic accuracy and diagnostic speed to be contradictory bodies. The price of diagnostic accuracy exaltation is to sacrifice some diagnostic speed. It is the same that promote of diagnostic speed would lower diagnostic accuracy. If want to raise the accuracy of the diagnosis model, we have to sacrifice some diagnostic speed and increase times of diagnostic training.Because of complexity and Repetitiveness of disease symptom, we have to withdraw the most advantageous symptoms which can judge the diagnostic result. So the chapter 6 discusses about Knowledge Representation Systems in Intelligent Medical Diagnosis System. Rough Sets is the important tool of knowledge detection. Constructs the arithmetic based on the knowledge representation reduction of the decision tables, and forms the reduction of decision representation from Rough Sets. Then, an example is given to analyze the process of knowledge reduction and a decision scheme is produced based on data analysis and deduction. The example is coronary heart disease patients' cases, which present some amazing medical knowledge. By analyzing the intelligent decision procedure based on Rough Sets and comparing with the parameters obtained by Medical Diagnosis system, the sub coronary heart disease symptoms are extracted, while those symptoms are dominating symptoms for majority coronary heart disease. Finally gives the whole data mining arithmetic.