Automated segmentation based upon remitted scatter spectra from pathologically distinct tumor regions

Abstract

Multi-spectral scatter visualization of tissue ultra-structure in situ can provide a unique tool for guiding surgical resection, but since changes are subtle and the data is multi-parametric, an automated methodology was sought to interpret these data, in order to classify their tissue sub-type. Tissue types observed across AsPC-1 pancreatic tumor samples were pathologically classified under three major groups (epithelium, fibrosis and necrosis) and the variations in scattering parameters, i.e. scattering power, scattering amplitude and average scattered intensity, across these groups were analyzed. The proposed scheme uses statistical pre-processing of the scattering parameter images to create additional data features followed by a k-nearest neighbors (kNN) based algorithm for tissue type classification. The classification accuracy inside some predefined regions of interest was determined and the mean region values of scattering parameters turned out to be stronger data sets for classification, rather than the individual pixel values. This presumably indicates that pixel-to-pixel variations in the remitted spectra need to be minimized for reliable classification approaches. Results show a strong correlation between the automated and expert-based classification within the predefined regions of interest.