Learning a Cost Function for Interactive Microscope Image Segmentation

摘要: Quantitative analysis of microscopy images is increasingly important in clinical researchers' efforts to unravel the cellular and molecular determinants disease, for pathological tissue samples. Yet, manual segmentation measurement cells or other features remains norm many fields. We report on a new system that aims robust accurate semi-automated microscope images. A user interactively outlines one more examples target object training image. then learn cost function detecting objects same type, either different The incorporated into an active contour model, which can efficiently determine optimal boundaries by dynamic programming. validate our approach compare it some standard alternatives three types microscopic images: light blood cells, muscle sections, electron cross-sections axons their myelin sheaths.