Surface-Enhanced Raman Spectroscopy and Transfer Learning Toward Accurate Reconstruction of the Surgical Zone

Abstract

Raman spectroscopy is a photonic modality defined as the inelastic backscattering of excitation coherent laser light. It is particularly beneficial for rapid tissue diagnosis in sensitive intraoperative environments like those involving the brain, due to its nonionizing potential, point-scanning capability, and highly-specific spectral fingerprint signatures that can characterize tissue pathology [1]. While Raman scattering is an inherently weak process, Surface-Enhanced Raman Spectroscopy (SERS), which is based on the use of metal nanostructure surfaces to amplify Raman signals, has become a compelling method for achieving highly specific Raman spectra with detection sensitivity comparable to conventional modalities such as fluorescence [2]. A unique plasmonics-active nanoplatform, SERS gold nanostars (GNS) have previously been designed in our group to accumulate preferentially in brain tumors [2]. Raman detection, when combined with machine learning and robotics, stands to enhance the diagnosis of ambiguous tissue during tumor resection surgery, with the potential to improve extent-of-resection and rapidly reconstruct the dynamic surgical field. Here we demonstrate preliminary results from the use of a SERS-based robotics platform to efficiently recreate a tumor embedded in healthy tissue, which is modeled here as a GNS-infused phantom. Transfer learning, specifically through use of the open-source RRUFF mineral database, is employed here to address the dearth of collected biomedical Raman data [3].