Files in this item



application/pdf9136689.pdf (5MB)Restricted to U of Illinois
(no description provided)PDF


Title:Multiresolution, adaptive methods and classification for image analysis of DNA autoradiographs
Author(s):Palaniappan, Kannappan
Doctoral Committee Chair(s):Huang, Thomas S.
Department / Program:Electrical and Computer Engineering
Discipline:Electrical Engineering
Degree Granting Institution:University of Illinois at Urbana-Champaign
Subject(s):Biology, Molecular
Engineering, Electronics and Electrical
Computer Science
Abstract:The analysis of autoradiograph images generated by the multiplex DNA sequencing method is considered. An overall approach to obtaining sequence data from autoradiograph images is outlined, and specific approaches for segmentation of the image into sets of four lanes, obtaining one-dimensional profiles of the lanes, detecting peaks in the one-dimensional profiles using a multiscale approach and aligning profiles across lanes using a geometric correction method are discussed. The extraction of various shape features and profile-related features is shown and their utility for classification is evaluated. For reconstructing the ordered sequence, several methods for classification of the feature set including statistical pattern recognition, expert system and adaptive network synthesis are evaluated. The major contributions include a new method of multiresolution signal analysis which is shown to be robust in decomposing the profile into a set of peaks. The method relies on the Laplacian of a Gaussian operator to determine a set of initial parameters. An important property of the operator that is proven and exploited in designing an algorithm is that the location of the peaks is invariant. The initial parameters are used as the starting conditions to solve the nonlinear maximum likelihood equations using the Expectation-Maximization algorithm. A variety of similarity criteria including several new information theoretic-based measures are evaluated for image matching and registration. Although the performance of the entropy based measures is statistically as good or better than correlation type measures they usually require more computation. A new adaptive moment-based method for local estimation and correction of image distortion is developed. The method extracts various knot points whose location are sensitive to the edges in the image. The number of knot points used determines the order of the interpolation. A classification algorithm combining statistical methods with a rule-based approach is proposed for reconstructing the sequence from the profiles. Theoretical results for classification using multithreshold networks are presented and offer an area for further investigation.
Issue Date:1991
Rights Information:Copyright 1991 Palaniappan, Kannappan
Date Available in IDEALS:2011-05-07
Identifier in Online Catalog:AAI9136689
OCLC Identifier:(UMI)AAI9136689

This item appears in the following Collection(s)

Item Statistics