Estimation from noisy images with the EM-algorithm
by
A. M. Nikiforov
SUNY at Stonybrook, NY, USA
Coauthors: M.B. Malyutov (Northeastern University, Boston, MA, USA), R. Mirchev, D.E. Golan (Harvard Medical School, Boston, MA, USA)

The paper deals with a series of noisy images generated from a moving object or several objects without a good model of the motion. An estimation procedure for the object location, shape and some other parameters is proposed that is based on the EM-algorithm. The output of this algorithm is the sequence of centroids locations which enables the tracking of trajectories of objects

The problem discussed in this paper stems from the study of motion of band 3 protein molecule over a red blood cell membrane. Parameters of this motion are of large biological interest. A gold or latex bead of size from 5 to 40 nm is attached to the molecule, and plane images (typically 128 by 80 pixels) are recorded at speed up to 10, 000 per second by a Kodak camera to track molecules. Enhanced optical microscopy applied for imaging objects that are smaller than the resolution limit of visible light provides images of beads ("bead spots"), that are often very noisy, have complex shape and have sizes from 100 to 500 nm, that exceed the desired accuracy of the bead position estimate by at least an order of magnitude.

Heuristic threshold- or moment-based algorithms often used by experimenters, besides being inaccurate in strong noise, are heavily biased when bead spots overlap and/or when a spot is too large or bead is too close to the image edge to fit a spot into the image frame due to the limited view area of the microscope.

We propose here a procedure based on the EM-algorithm that estimates the bead position more accurately and seems to be one of the best possible, if the motion modeling is not clear.It was tested by simulation. We hope to apply it to the real data from the experimental work described before in the near future. The main idea of the approach is to estimate parameters relevant to the bead spot generation using a series of images, and then to solve the inverse problem by establishing the most probable center coordinate for each image along with other characteristics

Date received: October 16, 2000

Copyright © 2000 by the author(s). The author(s) of this document and the organizers of the conference have granted their consent to include this abstract in Atlas Conferences Inc. Document # cafr-64.