1324 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 14, NO. 9, SEPTEMBER 2005 Automatic Ultrastructure Segmentation of Reconstructed CryoEM Maps of Icosahedral Viruses Zeyun Yu, Student Member, IEEE, and Chandrajit Bajaj, Member, IEEE Abstract—We present an automatic algorithm to segment all the local and global asymmetric units of a three-dimensional density map of icosahedral viruses. This approach is readily applicable to the structural analysis of a broad range of virus structures that are reconstructed using cryo-electron microscopy (cryo-EM) tech- nique. Our algorithm includes three major steps operating on the three dimensional density map: the detection of critical points of the volumetric density function, the detection of global and local symmetry axes, and, finally, the boundary segmentation of all the asymmetric units. We demonstrate the efficacy of our algorithm and report our results on several experimental volumetric datasets, consisting of both reconstructed cryo-EM molecular density maps taken from the European Bioinformatics Institute archive, as well our own synthetically generated (blurred) maps calculated from X-ray resolution molecular structural data taken from the Protein Data Bank. Index Terms—Cryo-electron microscopy (cryo-EM) maps, icosa- hedral virus, segmentation, structure analysis, symmetry detec- tion, three-dimensional (3-D) reconstruction. I. INTRODUCTION A N ANIMAL, plant, or bacterial virus, is a parasite on its host cells. Mature viruses, called virions, cannot “grow” in isolation, as they do not possess the requisite protein syn- thesis machinery. All virions must depend on their host cells to accomplish their biological processes and reproduce them- selves, by which they “consume” the cells’ nutrients, and disrupt the metabolism of the host cells to eventually kill the host cells [1]. Viruses or virions (we henceforth use this interchangeably) in their life cycle span can significantly influence and impact their local and the global environment. For example, an infesta- tion of the rice dwarf virus (RDV), a major pathogen of the rice plants in southeast Asia, would result in severe global economic consequences [2]. The Herpes simplex virus type 1 (HSV-1), a human pathogen, is responsible for various diseases including cold sores, blindness, and encephalitis [3]. Understanding the molecular structure of virions and its inherent geometric com- plexity (ultrastructure) is the first step toward an understanding Manuscript received December 2, 2004; revised May 19, 2005. This work was supported in part by the National Science Foundation-ITR under Grants ACI-022003 and EIA-0325550 and in part by National Insitutes of Health under Grants 0P20 RR020647 and R01 GM074258. The associate editor coordinating the review of this manuscript and approving it for publication was Dr. Erik Meijering. The authors are with the Computational Visualization Center, Department of Computer Sciences and The Institute of Computational Engineering and Sciences, The University of Texas at Austin, Austin, TX 78712 USA (e-mail: zeyun.yu@gmail.com; bajaj@cs.utexas.edu). Digital Object Identifier 10.1109/TIP.2005.852770 of the molecular processes of how these pathogens infect the host cells, and often necessary for devising means to suppress the consequences of viral infections. Many studies have revealed that most if not all functions of biological units including viruses can be traced back to their structures and compositions. Despite the large range of varia- tions, most viruses consist of two fundamental components: the genomic DNA/RNA and the coat proteins. The coat proteins form a shell (capsid) and play two major roles: to protect the fragile DNA/RNA and to help the viruses recognize the host cells. The coat proteins are very important for a virus to live and reproduce itself, and much attention has been drawn to study the three-dimensional (3-D) structures of these proteins. X-ray diffraction has been so far the most powerful technique to deter- mine the structures of virions (and many other individual pro- teins) at an atomic level. However, this method requires certain crystalline forms of the purified virus (crystallography) but un- fortunately many viruses are resistant to the formation of crys- tals big enough for the high-resolution image reconstruction through x-ray diffraction. Nuclear magnetic resonance (NMR) is another increasingly popular method used to determine the atomic structure of molecules. However, this method is limited to only relatively small molecules ( 50 K Dalton), which makes it hard to study even the smallest virus particles. To overcome the limitations of both X-ray crystallography and NMR methods, a noncrystallography technique using cryo- electron microscopy (cryo-EM) has been providing a powerful tool in revealing the “full picture” of large macromolecular com- plexes at subnanometer resolutions (5–10 ) [4]–[6]. By this technique, the biological particles (viruses, ribosome, etc.) are rapidly frozen in water to liquid nitrogen temperatures (about C). In this process, the frozen water does not form ice crystals but a thin layer containing the virus particles to ab- sorb and protect the native structure of particles from the high intensity electron radiation beams. However, the particle im- ages taken with transmission electron microscopes have a large amount of noise and very low contrast between the vitreous ice and the biological particles, so that thousands (or more) of iden- tical particles have to be purified, imaged and averaged in order to improve the signal-to-noise ratio for a reliable and reasonably high-resolution 3-D reconstruction [7]. This technique is com- monly known as single particle reconstruction [4]–[6]. The fun- damental theory of 3-D reconstruction from two-dimensional projections is basically the same as the well-known CT tech- nique as seen in medical imaging. However, in single particle reconstruction, the major difficulty arises from the unknown ori- entations of the particles that randomly “float” in the vitreous 1057-7149/$20.00 © 2005 IEEE