Breast cancer diagnosis from fluorescence spectroscopy using support vector machine

摘要

A novel support vector machine (SVM) classifier incorporating the complexity of fluorescent spectral data is designed to reliably differentiate normal and malignant human breast cancer tissues. Analysis has been carried out with parallel and perpendicularly polarized fluorescence data using 36 normal and 36 cancerous tissue samples. In order to incorporate the complexity of fluorescence spectral profile into a SVM design, the curvature of phase space trajectory is extracted as a useful complexity feature. We found that the fluorescence intensity peaks at 541nm-620nm as well as the complexity features at 621nm-700nm are important discriminating features. By incorporating both features in SVM design, we can improve both sensitivity and specificity of the classifier.