Método de detecção de câncer de ovário utilizando padrões proteômicos, análise de componentes independentes e máquina de vetores de suporte

Abstract

varian cancer is a kind of cancer gynecologic origin more difficult to be diagnosed, because the majority of malignant ovarian tumors is manifested only in the advanced stage of the disease, thus decreasing the chance of cure. Only about 20% of ovarian cancers are diagnosed early. Transvaginal ultrasonography is the most widely used diagnostic method for differential diagnosis, but is not yet effective for the early diagnosis. This paper proposes a CAD (Computer-Aided Diagnosis) method for early detection of ovarian cancer with the aim of helping other existing methods, using the Independent Component Analysis (ICA) technique, for feature extraction of proteomic signals using the FastICA algorithm, the technique of Maximum Relevance and Minimum Redundancy (mRMR), to decrease the computational cost and reduced dimensionality of the features matriz, it is through the selection of the most significant features among all extracted by ICA. After extraction and selection of features used the Support Vector Machine (SVM), to classify the samples between the presence or absence of cancer. The method was tested in the database SELDITOF proteomic patterns, containing 253 samples at low resolution (15,154 points), being 162 cancer and 91 benign. Based on the tests, the best performance was obtained with a vector of 10 features, resulting in an average accuracy of 98.80%, with 95.65% specificity and 100% sensitivity.