Ovarian cancer is the fifth most deadly cancer among women in North America. Because this type of cancer is often diagnosed late, cytoreductive surgery is often the first therapeutic step. Currently, visual inspection of the surgical cavity is the only technique used to detect residual tumors. Therefore, there is a need for the development of new imaging techniques that can detect cancer tissue with high specificity and sensitivity during cytoreductive procedures. To address this unmet clinical need, we developed an intraoperative wide-field Raman spectroscopy (RS) imaging system to be used alongside tissue classification models trained to recognize cancer tissue using artificial intelligence techniques. The system can sequentially acquire up to 5 Raman bands in imaging mode over a macroscopic tissue area of more than 1-centimeter diameter. Preliminary analyses are presented demonstrating the ability of the system to recover the main Raman tissue bands in synthetic and biologic tissue. Two types of tissues in a biological sample can also be differentiated by the system. Moreover, cancer detection models are produced using a single-point RS probe based on ex vivo human measurements collected from 20 ovarian cancer patients. Using supervised machine learning techniques, it is demonstrated the model can detect tissue containing epithelial cancer cells with an accuracy higher than 90%. Based on this dataset, multivariate statistical analyzes were performed demonstrating the 5 features contributing the most to the classification. These studies pave the way to the development of a new generation wide-field Raman spectroscopy techniques for macroscopic tissue characterization during surgery.

Date Published: 4 March 2019
PDF
Proc. SPIE 10890, Label-free Biomedical Imaging and Sensing (LBIS) 2019, 108901O (4 March 2019); doi: 10.1117/12.2509773

Published in SPIE Proceedings Vol. 10890:
Label-free Biomedical Imaging and Sensing (LBIS) 2019
Natan T. Shaked; Oliver Hayden, Editor(s)