Application Of Protein Mass Spetrum Data And Neural Network In Diagnosis Of Ovarian Cancer

Ovarian cancer, oophoroma, is one of hackneyed cancer in gynecological cancers. Ovarian cancer has some salient features between other cancers, which are the symptoms of ovarian cancer are not obvious and are not easy to examine in the early stage. Most patients were in the terminal stage of ovarian cancer once they were diagnosed with ovarian cancer without any examinations in early stage. It is very dangerous, because terminal ovarian cancer has a high mortality. But ovarian cancer is examined in early stage, and through some treatment, the mortality rate of it is lower. So the features make people to find good methods to examine ovarian in early stage. This paper is based on mass spectrometric data, combining with the technologies in pattern recognition to examination ovarian cancer.Proteomics which was attention by many scholars, becomes another hot spot in life science after genetics. Mass theory and mass spectrometry promoted the development of proteomics, and become the powerful tools in some study of proteomics. Genetic algorithm and neural network are the methods in optimization and pattern recognition respectively, which are widely applicable in many directions. This paper did some researches including mass spectrometric data preprocessing, features selection using genetic algorithm, pattern classification using neural network.The SELDI-TOF MS data from the data base build by NCI-FDA would be used as the raw data. A preprocessing course was made at first in this paper, which included data loading, noise reduction, baseline correction and normalization. Through the preprocessing course, the raw data would be better to classify. After the preprocessing, genetic algorithm would be use to extract and select features. I this paper, 4 features were selected as the best features. And 5 features were extracted from the areas of feature triangle and perimeter of the star plot, which were made from the star plots of the 4 features mentioned. Lastly, the neural network was build, and trained by 5 features of all samples. After training, we pick a set of mass spectrometric data randomly to detect the neural network, and find that it had a good result in classification.