Human glioma tumor diagnosis using visible resonance Raman spectroscopy and deep learning

Abstract

Convolutional neural network (CNN) based deep learning is used to analyze spectral data collected by visible resonance Raman (VRR) spectroscopy to distinguish human glioma tumors from healthy brain tissues using binary classification and identify the cancer grades of the glioma tumors using multi-class classification. Classification was performed using both raw spectral data and baseline-subtracted data for comparison. The classification using both datasets yielded high accuracy, with the results obtained from baseline subtracted spectra slightly better than that obtained from raw spectra. The study showed VRR combined with deep learning provides a robust molecular diagnostic tool for accurately distinguishing glioma tumors from normal tissues and glioma tumor tissues at different cancer grades. Deep learning aided VRR technique may be used for in-situ intraoperative diagnosis of brain cancer. It may help a surgeon to identify cancer margins and even cancer grades during surgery.