Automated ensemble segmentation of epithelial proliferation, necrosis, and fibrosis using scatter tumor imaging

Abstract

Conventional imaging systems used today in surgical settings rely on contrast enhancement based on color and intensity and they are not sensitive to morphology changes at the microscopic level. Elastic light scattering spectroscopy has been shown to distinguish ultra-structural changes in tissue. Therefore, it could provide this intrinsic contrast being enormously useful in guiding complex surgical interventions. Scatter parameters associated with epithelial proliferation, necrosis and fibrosis in pancreatic tumors were previously estimated in a quantitative manner. Subtle variations were encountered across the distinct diagnostic categories. This work proposes an automated methodology to correlate these variations with their corresponding tumor morphologies. A new approach based on the aggregation of the predictions of K-nearest neighbors (kNN) algorithm and Artificial Neural Networks (ANNs) has been developed. The major benefit obtained from the combination of the distinct classifiers is a significant increase in the number of pixel localizations whose corresponding tissue type is reliably assured. Pseudo-color diagnosis images are provided showing a strong correlation with sample segmentations performed by a veterinary pathologist.