Surgery remains a primary tool to remove cancerous tumors; however, a major challenge of the procedure is incomplete surgical removal leading to recurrence, costly adjuvants and secondary surgeries. The current method to determine surgical success relies on histopathology, after the surgery is complete. We are developing methods to assess surgical success during surgery that would improve outcomes and reduce costs.
Specifically, we are developing Raman micro-spectroscopy (Raman) as a rapid, low-infrastructure method, capable of assessing excised whole tissues without any tissue preparation (sectioning and staining). Raman is a scanning laser-based technique that probes a plurality of tissue biochemical components including structural proteins (collagen, elastin, keratin), DNA content, and lipids (triolein, ceramide). When combined with machine learning to assist in the interpretation of Raman spectra and images, this approach has the potential to provide the surgeon with an automated and simple “Keep Cutting” or “Stop Cutting” guidance during surgery. Our vision for this image-guided approach is for Raman microscopy to provide biochemical images that machine learning algorithms will use to provide an automated diagnosis (Figure 1).
Figure 1. Raman microscopy provides image guidance to skin tumor resections
We have developed Raman methods to measure biophysical skin components and use those features to classify nonmelanoma skin cancer (basal cell carcinoma). We determined that Raman measures at least eight independent skin components including: collagen, elastin, triolein, cell nucleus, keratin, ceramide, melanin, and water (Feng 2017). In a study of 30 patients, we demonstrated that a predictive model trained on the biochemical components extracted from Raman images discriminates basal cell carcinoma from normal structures with a sensitivity of 90% with a specificity of 92% (Feng 2019). For this surgical guidance application, an ideal test would have a high positive predictive value (PPV = probability that a positive margin exist given the Raman test estimates a positive margin). Targeting high PPV, this model reached 93% PPV with a specificity of 97% and sensitivity of 52%. High PPV and specificity ensure that the surgeon would only be instructed to “keep cutting” when tumor is present (to >90% confidence). Our current efforts are focused on methods to speed this approach so that the entire process can be performed within minutes.
Figure 2. Example of Raman microscopy on a surgical skin specimen during Mohs micrographic surgery. LEFT: H&E stained skin surgical specimen showing reflectance confocal images where Raman spectra were acquired from various skin structures. RIGHT: Basis spectra of the eight biophysical components that Raman measures independently in human skin.
References
Feng, X., A.J. Moy, H.T.M. Nguyen, J. Zhang, M.C. Fox, K.R. Sebastian, J.S. Reichenberg, M.K. Markey, and J.W. Tunnell, Raman active components of skin cancer. Biomedical Optics Express, 2017. 8(6): p. 2835-2850.
Feng, X., M.C. Fox, J.S. Reichenberg, F.C.P.S. Lopes, K.R. Sebastian, M.K. Markey, and J.W. Tunnell, Biophysical basis of skin cancer margin assessment using Raman spectroscopy. Biomedical Optics Express, 2019. 10(1): p. 104-118.
