Discriminación de patologías tumorales en tejidos cancerígenos mediante espectroscopía de imagen

Abstract

Multi-spectral scatter visualization of tissue ultrastructure in situ can provide a unique tool for guiding surgical tumor resection. The variations in scattering parameters, i.e. scattering power, scattering amplitude and average scattered intensity, across different tissue types has been analyzed. Since scatter changes are subtle, tissue sub-type identification requires multiparametric analysis of optical data to help in tumor delineation. The proposed methodology has been validated on tissue types observed across pancreatic tumor samples that were pathologically classified under three major groups (epithelium, fibrosis and necrosis) with their corresponding subtypes. This methodology combines a statistical pre-processing of the scattering parameters and an ensemble segmentation method. The latter merges the predictions of k-nearest neighbors (kNN) and Artificial Neural Network (ANN) algorithms for tissue type classification. The classification accuracy inside some predefined regions of interest was determined. Results show a strong correlation between the automated and expert-based classifications.