| Lung cancer,as a common malignant tumor in the world today, has become the major reason of cancer leading to death. In the early period of lung cancer there are no or few symptoms, so it is difficult to find, when detected usually it has transferred to other part of the body. So the early diagnosis and treatment is a necessary method to improving the survival rate and reducing mortality of the patients with lung cancer, and it is also a difficult technology of lung disease detection. CT is widely used in diagnosis of lung cancer, but for doctors making correct diagnosis is not easy because of the complexity of CT and other uncertain factors.Although artificial neural network has had a wider application in the medical field, the design of the structure of the neural network and the initial value of the weight are still lack of support, and with slow convergence, classic BP algorithm easily fall into a local minima, all these problems need to solve. Genetic algorithm has a strong global search capability, and this can make up for the flaws of BP network. Many scholars have put forth a GA and BP network integrating algorithm, but the traditional GA-BP algorithm often optimize only one aspect of neural network-whether the weights or the structure. In this way the network may under a redundancy structure after only optimizing the weights' walue, and another method can not make sure to achieve the best weight value after only optimizing the structure. To solve this problem, this paper introduces Co-evolution Genetic Algorithm , structure and weights are at the same time encoded in the chromosome; And because GA has not a strong local search ability, the traditional GA-BP algorithm could not find the optimal solution as soon as possible when fastly converge to the neighborhood of the optimal solution, and which lead to so much times of iterations. So this paper introduces Simulated Annealing Algorithm, to ensure the local random search capabilities,which will accelerate convergence to the optimal solution. This paper also uses the algorithm of adaptive changing crossover probability and the mutation probability, in the initial period choose a larger mutation probability to maintain the diversity of the population, when network is close to the neighborhood of the optimal solution, reducing the mutation probability to obtain the optimal network design. This method avoids relying on the experience to decide the network structure, overcomes the randomness of the initialization of weights' walue and the oscillation in the process of determining the network structure, and effectively enhances the generalization ability of the neural network.In this paper, a large number of experimental data of lung cancer patients are analysed on the basis of comprehensive consideration of many complex factors, and an improved genetic algorithm optimizing neural network is used to establish the neural network model, and finally received the most closely 8 major risk factors associated with lung cancer, and the characteristics of lung CT images were analyzed and researched. The MATLAB simulation experiment proved that the accuracy rate of the improvement GA-BP algorithm is 97.5%, is higher than traditional GA-BP algorithm (94.2%) , and a little higher than the correct diagnosis rate (96.7%) of doctors, its convergence rate is also greatly increased. Useing this algorithm can help help doctors avoid the constraints of the knowledge and experience and subjectivity, and have an effective distinguish and identification of CT features with lung cancer and other lung-disease, to discover lung cancer as early as possible, and provide an effective way of decision in the early diagnosis of lung cancer. |
PDF Full Text Request