Detecting Silica-Coated Gold Nanostars within Surface-Enhanced Resonance Raman Spectroscopy Mapping via Semi-Supervised Framework Combining Feature Selection and Classification

Abstract

Raman Spectroscopy provides a non-invasive approach to study cells and tissues, and its ability to provide biochemical composition information of samples shows great importance for the research, diagnosis and treatment of cancer. However, conventional Raman Spectroscopy suffers from weak signal strength observed in many biological samples. Surface-Enhanced Resonance Raman Spectroscopy (SERRS) can overcome this disadvantage with the presence of roughened nano-dimensional noble-metal surfaces. In order to study the role of integrins in breast cancer invasiveness, gold nanostars were conjugated with cyclo-RGDf/k peptide for targeting integrins on breast cancer cells and high-speed Raman mapping was employed to assess the samples. Due to the high dimensionality of the datasets collected through SERRS, we have proposed a semi-supervised framework combining feature selection and classification techniques for nanostars detection and tested our method on a breast cancer cells. The results show the advantage of our framework over other data mining technique and potentially provide a new method for evaluating the role of integrins in tumor development. Also, the features selected can possibly be used for further studies on compositional changes observed during the process of breast cancer progression and metastasis.