Multispectral tissue analysis and classification towards enabling automated robotic surgery

Accurate optical characterization of different tissue types is an important tool for potentially guiding surgeons and enabling automated robotic surgery. Multispectral imaging and analysis have been used in the literature to detect spectral variations in tissue reflectance that may be visible to the naked eye. Using this technique, hidden structures can be visualized and analyzed for effective tissue classification. Here, we investigated the feasibility of automated tissue classification using multispectral tissue analysis. Broadband reflectance spectra (200-1050 nm) were collected from nine different ex vivo porcine tissues types using an optical fiber-probe based spectrometer system. We created a mathematical model to train and distinguish different tissue types based upon analysis of the observed spectra using total principal component regression (TPCR). Compared to other reported methods, our technique is computationally inexpensive and suitable for real-time implementation. Each of the 92 spectra was cross-referenced against the nine tissue types. Preliminary results show a mean detection rate of 91.3%, with detection rates of 100% and 70.0% (inner and outer kidney), 100% and 100% (inner and outer liver), 100% (outer stomach), and 90.9%, 100%, 70.0%, 85.7% (four different inner stomach areas, respectively). We conclude that automated tissue differentiation using our multispectral tissue analysis method is feasible in multiple ex vivo tissue specimens. Although measurements were performed using ex vivo tissues, these results suggest that real-time, in vivo tissue identification during surgery may be possible.