Automatic segmentation of histological structures in normal and neoplastic mammary gland tissue sections

摘要

In this paper we present a scheme for real time segmentation of histological structures in microscopic images of normal and neoplastic mammary gland sections. Paraffin embedded or frozen tissue blocks are sliced, and sections are stained with hematoxylin and eosin (H&E). The sections are then imaged using conventional bright field microscopy. The background of the images is corrected by arithmetic manipulation using a "phantom". Then we use the fast marching method with a speed function that depends on the brightness gradient of the imageto obtain a preliminary approximation to the boundaries of the structures of interest within a region of interest (ROI) of the entire section manually selected b the user. We use the result of the fast marching method as the initial condition for the level set motion equation. We run this last method for a few steps and obtain the final result of the segmentation. These results can be connected from section to section to build a three-dimensional reconstruction of the entire tissue block that we are stud ing.