Nonlinear dimensionality reduction in the analysis of high dimensional medical data
WENG Shifeng, ZHANG Changshui, ZHANG Xuegong (Department of Automation, Tsinghua University, Beijing 100084, China)
It was difficult to find the intrinsic structure of high dimensional medical data by traditional technologies. The new method named Isomap was applied to two classic medical datasets, lung cancer gene expression data and breast cancer pathological data. The intrinsic dimensionalities of these two datasets were found to be less than three, so they could be visualized in low dimensional space. Comparison of a standard linear dimensionality reduction method, PCA, with Isomap showed that Isomap gave better performance when calculating the within class distance. Therefore, nonlinear dimensionality reduction technology has potential in the analysis of high dimensional medical data.