| Based on the 31Phosphorus Magnetic Resonance Spectroscopy, data samples of liver can be distinguished into three diagnostic types of hepatocellular carcinoma, hepatic cirrhosis and normal liver through pattern recognition. 31Phosphorus Magnetic Resonance Spectroscopy which is the only non-invasive method used in liver cells in vivo plays an important role in clinical applications, such as in non-invasive diagnosis, in treatment as well as in monitoring. But it is necessary to deal with the magnetic resonance spectroscopy before diagnosing its type. The treatment of the spectroscopy later is not only relaxed but also need accomplished manually by professional, and it is related with the skillful level of the operator. So it is necessary to research an objective method to evaluate spectrum. All the data for each 31phosphorus magnetic resonance spectroscopy are firstly obtained, and then pattern classifications are did on the data sets. But the correlation between each feature and high dimensionality challenge the pattern recognition methods. In order to solve the two problems above, optim features must be selected by genetic algorithm before classifications, and then linear classifier and quadratic classifier are used to classify data sets in pattern classification, moreover, k-fold cross-validation method is used to divide the data into training and testing subset between classification.Experiments show that the method which selects optimal feature subset using genetic algorithm and then classified by linear classifier improves the classification accuracy significantly, classification accuracy rate of the carcinoma samples, hepatic cirrhosis samples and normal liver samples are above 87.90%, 91.97% and 99.64%. Optimal features selected by genetic algorithm distribute round pivotal resonance peak that is nearly consistent with medical diagnosis, which can help clinicians to make medical diagnosis in 31phosphorus magnetic resonance spectroscopy in future. |
PDF Full Text Request