The Application Of SVM Classifier In The Earlier Diagnosis Of Liver Cancer

Hepatocellular carcinoma (HCC) is the second cause of cancer-related death worldwide, and the incidence rate of liver cancer has continuously increased, with approximately 750,000 new diagnosed cases each year. Especially in China, both the incidence and mortality rate of HCC have been ranked second among all cancers. Importantly, HCC mortality rate is similar to its incidence rate, indicating that most patients with liver cancer die from HCC. In clinical practice, liver cancers are usually diagnosed by detecting alpha-fetoprotein (AFP) and abdominal ultrasound. But there are two problems within the traditional diagnosis methods: The first, the patients with HCC have reached the advanced liver cancer when detecting the AFP abnormal. At this time, whether to have the surgery, radiotherapy, chemotherapy or any other treatment, the cure rate of patients with HCC is very low, and the cost is extremely high. The second, the AFP is not the unique marker in diagnosing HCC. And sometimes it’s invalid. Thus it may lead to the misdiagnosis, and delay the best treatment period of the patients. Therefore, in this paper, we seek to discover and extract hidden patterns and relationships among large number of biomarkers, then we create the classifier through using the biomarkers of HCC and their relation to predict HCC as soon as possible, and improve the accuracy of diagnosis of HCC. The research content includes the following aspects:(1) Analyze detection indicators of patients with earlier HCC and patients with liver disease, and these data are analyzed and preprocessing. Seen from the results of the analysis, the AFP are in the normal range of most patients with the earlier HCC, and is equal to patients of liver disease; the most other detection indicators are higher than the normal range. So their detection indicators exist cross sections, they can’t be distinguished by only indicators. In data preprocessing, association algorithm is utilized to extract specific indicators, using feature selection and principal component analysis (PCA) for data dimensionality reduction.(2) Support vector machines prediction model are built with these sample for discriminating patients with earlier HCC from patients with liver disease. Our HCC-SVM classifier integrated 16 specific biomarkers which are obtained by data preprocessing, these indicators have an important significance for the diagnosis of liver cancer; at the same time we use the grid-search and PSO algorithm to optimize parameters of the SVM model, the prediction accuracy respectively were 94.186% and 93.0233% of final classifier model.(3) Because basic PSO is easy to fall into local optimal, we use the adaptive mutation to optimize PSO, in order to achieve the global optimal solution and improve the accuracy of classifier, the prediction accuracy is 95.3488% in the end.