Cancer Diagnosis Based On Genetic Algorithm And Neural Network

Aimed at the problem that the recognition rate of cancer in 31phosphorus magnetic resonance spectroscopy is low, a recognition method based on genetic algorithm (GA) and neural network was presented. But the redundancy and small sample size challenge the pattern recognition methods. To avoid the curse of dimensionality of mass spectra, feature selection must be employed to reduce the dimensionality before classification and analysis. GA and t-test feature selection method were used to select optimal features form the 401 initial features. Then, BP (Back-propagation Neural Network, BP) and RBF (Radial Basis Function Neural Network, PNN) were used to classify samples based on these optimized features, and make comparison between results based on 20 optimal features and the all 401 features. The results of the experiment show that the method on GA can improve the recognition rate of cancer.Abnormalities of 31P-MRS were found in patients of hepatitis, liver cirrhosis, liver tumor, patients after liver transplantation. Evaluation of 31P-MRS is important in diagnosis and treatment of many hepatic diseases. As a non-invasive protocol for analyzing the energetic metabolism and biomedical changes in cellular level of living liver, 31P-MRS has a wide clinical application. Conventional methods to assay hepatic ATP require large tissue samples,making repeat measurements on the same animal or human impossible,and are unable to monitor the minimal changes in metabolism consistent with early or reversible cellular injury. Through the evaluation of the 31P-MR spectroscopy,we can distinguish three types of diagnosis: hepatocellular carcinoma, normal and cirrhosis. Back-propagation neural network (BP) and Radial Basis Function Neural Network (RBF) are applied to analyze 31P-MRS data, develop neural network models of 31P-MRS for the diagnostic classification of hepatocellular carcinoma to improve the recognition rate. The results suggest that BP models have better performance than RBF models. After application of neural network models, the diagnostic accuracy rate of hepatocellular carcinoma is improved greatly.