Metabolic profiling research of primary liver tumor based on multiple discriminant analysis

Abstract

Objective: To compare the diagnosis performance of discriminant analysis methods through application on clinical serum samples. Methods: Linear discriminant analysis (LDA), quadratic discriminant analysis (QDA) and logistic discriminant analysis (LogDA) were applied to metabolic profiling analysis deriving from clinical serum samples of 109 healthy controls, 87 patients with benign liver tumor and 31 patients with malignant liver tumor. The diagnosis performance of these three methods was compared in discrimination of healthy controls and patients with liver tumor(benign tumor and malignant tumor) and in discrimination of patients with benign tumor and those with malignant tumor. Results: Based on current clinical metabolic profiling data, the effectiveness of all these three methods worked better in discrimination of healthy controls and patients with liver tumor than in discrimination of patients with benign tumor and those with malignant tumor. The overall performance of QDA was superior to LDA and LogDA, with the precision of 87.67% in discrimination of healthy controls and patients with liver tumor and the precision of 67.80% in discrimination of patients with benign tumor and those with malignant tumor. Conclusion: QDA outperforms LDA and LogDA in processing primary liver tumor metabolic data.