Towards Recovery of 3D Chromosome Structure Abstract 1 The objectives of this work include automatic recovery and visualization of a 3D chromosome structure from a sequence of 2D tomographic reconstruction images taken through the nucleus of a cell. Structure is very important for biologists as it affects chromosome functions, behavior of the cell and its state. Chromosome analysis is significant in the detection of diseases and in monitoring environmental gene mutations. The algorithm incorporates thresholding based on a histogram analysis with a polyline splitting algorithm, contour extraction via active contours, and detection of the 3D chromosome structure by establishing corresponding regions throughout the slices. Visualization using point cloud meshing generates a 3D surface. The 3D triangular mesh of the chromosomes provides surface detail and allows a user to interactively analyze chromosomes using visualization software. 1. Introduction 1.1 Motivation Tracking and visualizing chromosomes gives biologists valuable information regarding their three- dimensional (3D) structure and behavior. Previously, segmentation of banded chromosomes frozen in metaphase of mitosis was important for classification especially in the karyotyping process. This process facilitates the classification and detection of chromosomal abnormalities such as Kleinfelter’s, Down’s, and Turner’s syndrome. 3D visualization of the chromosome can be useful for biologists in the following ways: (1) identifyng the space occupied by the chromosome within the cell, (2) visualizing 1 This work was performed under the auspices of the U.S. Department of Energy by University of California Lawrence Livermore National Laboratory under contract number W-7405-Eng- 48. UCRL-PROC-203893 specific structures along the contour such as “constrict points,” and binding sites with other intercellular molecules such as proteins, enzymes, and other organelles, (3) using the visualization to accurately classify the chromosomes, (4) detecting anomalies, such as chromosomal disorders, and (5) helping to identify the behavior of the organelle over time (sometimes called 4D reconstruction, with time as the the fourth dimension). 1.2 Previous Work Previous research of chromosomes in 2D images was primarily focused on abnormality detection and classification of chromosomes. In chromosome classification (Karyotyping), one of the main efforts includes the problem of separation of partially occluded chromosomes. Lerner et al. [1] proposed classification based on skeleton points, and local feature extraction for classification purposes (CPOOS – Classification-Driven Partially Occluded Object Segmentation method). Shi et al. also used local features such as cut points, skeleton points, junction points, and ravine points to separate touching chromosomes using Parallel Mesh algorithm [3]. Lerner et al. [9] trained Multilayer Perceptron (MLP) Neural Networks to classify chromosomes and used a “knock out” technique as well as Principle Component Analysis (PCA) for feature selection. Vidal et al. used syntactic/structural pattern recognition algorithms such as Error-Correcting Grammatical Interface (ECGI) and MLP to classify chromosomes by formulating rule- based string representation of the features extracted [2]. Keller et al. presented a fuzzy logic system in addition to neural network based classification system to deal with ambiguities during the classification process [11]. 3D visualization enables scientists to discern occluding chromosomes for further classification better than 2D image analysis. There have also been works on visualization and 3D reconstruction of large and small biological objects based on various imaging Sabarish Babu, Pao-Chuan Liao, Min C. Shin Dept of Computer Science UNC Charlotte Charlotte, NC 28223 {sbabu, pliao, mcshin}@uncc.edu Leonid V. Tsap Electronics Engineering Dept Lawrence Livermore National Laboratory Livermore, CA 94551 tsap@llnl.gov Proceedings of the 2004 IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops (CVPRW’04) 1063-6919/04 $ 20.00 IEEE