Registration and segmentation based analysis of microscopy images

Kevin S Lorenz, Purdue University

Abstract

Optical microscopy exhibits many challenges for digital image analysis. In general, microscopy volumes are inherently anisotropic, suffer from decreasing contrast with tissue depth, lack object edge detail, and characteristically have low signal levels. Image analysis is motivated by the desire to quantify biological characteristics such as cell count and tissue volume. This thesis describes methods for an integrated approach to segmentation and registration of intravital microscopy image sets. Image data are acquired using one of two distinct techniques. One data set type consists of a series of images corresponding to focal planes looking deeper in the tissue (three dimensional data), and a second type consists of a series of images corresponding to a sequence of time instances imaging a single focal plane (time-series data). Analysis is performed via a combination of segmentation and registration techniques, but is complicated by factors such as live specimen motion during image acquisition. In particular, we describe a method that utilizes image enhancement, spatial filtering, rigid and non-rigid registration, and temporal filtering. Experimental results indicate our methods are promising based on the analysis of several sets of liver, kidney, lung, and salivary gland images. By lacking ground truth data to evaluate accuracy of results, analysis consistency is evaluated using flipped image volumes and reverse image acquisition. Registration validation is performed using motion vector analysis, a non-traditional application of a technique borrowed from motion estimation.

Degree

Ph.D.

Advisors

Salama, Purdue University.

Subject Area

Electrical engineering|Medical imaging

COinS