Automated Collagen Segmentation from Masson's Trichrome Stained Images-Preliminary Results

摘要

We propose a new algorithm for automated collagen segmentation from Masson's trichrome stained images with two steps: color transfer from a known, optimum-stained image, followed by k-means clustering. The algorithm's output is scored by two proficient pathologists. Results show a good segmentation output (~60%) and strong agreement on pathologist's opinion on a good segmentation (Cohen's kappa = 0.8063).