The major goals for these software projects are to design, develop and implement programs for semiautomated segmentation and classification of 3D volumes, and to develop software for the visualization of three-dimensional structure. In the previous years of this project we have developed the program suites, Synu and Ducky, for performing sophisticated three-dimensional visualizations. These programs have been found to be extremely useful by biologists and have been distributed to many investigators throughout the US and internationally. During this year we have focused on improvements in software for segmentation and classification as well as programs for combining volume datasets for use in performing serial tomography and for merging volume correlated datasets acquired from the light and microscope (Perkins et al., 1997). Manual segmentation-tracing programs: In many light and electron microscopic studies, the objects of interest for three-dimensional reconstruction or analysis are delineated by membranes. These membranes are often extremely difficult to define using image processing techniques. Generally it has been found necessary to specify the location of the membranes using an interactive manual tracing method. The program XVOXTRACE developed at NCMIR is specifically designed for manually contouring features of structures in tomographic volumes by tracing with a mouse on planes of the volume along any one of the three major axes. Additional guides are provided to achieve accurate tracing. Thus, as the contours are traced, they can be viewed simultaneously superimposed on selected images from the original tilt series. Pairs of tilt-series images with the contours superimposed may also be displayed as anaglyphic or stereoscopic views. In addition, the accuracy of tracing can be evaluated by examining the contours superimposed on renderings of the volume viewed from an arbitrarily selected orientation. We have developing an improved version of this program which will include stereoscopic volume rendering as well as enhanced contour editing facilities such as automatic contour resplining. Capabilities for panning and zooming on a view of a selected region of the specimen in the volume slice, volume rendering, and tilt displays will also be provided. Semiautomated particle detection: Several on-going NCMIR research projects are interested in defining the three-dimensional distribution of particle-like structures such as synaptic vesicles. In collaboration with Dr. Ioana Martin (Dept. Computer Science, Univ. Indiana, South Bend) we have been developing semiautomated tool for detecting and defining particles. An initial tool was developed based on methods used by Dr. Martin to detect virus particles in two-dimensional electron micrographs. The tool was evaluated to detection of synaptic vesicles in tomographic reconstructions. Although partly successful, we expect that algorithms, currently under development, using three-dimensional information will be considerably more accurate. Measurements based on contours: XDEND is a graphics-based analysis tool developed at NCMIR to obtain measurements of the length, surface area and volume of objects defined as sets of contours such as those produced with XVOXTRACE. In addition, contours can also be derived directly from data volumes in those cases in which the structure can be segmented using simple built-in thresholding. Components of the structure may be resectioned into contours in the transverse, sagittal, or coronal axes to obtain the best orientation for length measurements. The length measurements are based on linear segments interconnecting the centroids of the contours. The program automatically detects branching segments. Objects may be viewed in perspective as contour outlines, centroids or skeletons. XDEND can also be used to reclassify and edit objects after recontouring. We are currently improving this program by providing resectioning along an arbitrary axis, and more accurate skeletonization methods. Software for serial tomography and volume merging We previously developed and implemented a procedure for combining serial section techniques with tomographic reconstruction (Soto et al., 1994). For a fixed tilt range and tilt increment, the resolution of the volume reconstructed by tomography decreases as the thickness of the object under reconstruction is increased. Thus, the resolution obtainable in reconstructions of very thick specimens is relatively poor. To overcome this limitation, we combined tomography with the serial section reconstruction technique to reconstruct large structures with adequate resolution. A series of consecutive sections is cut with a thickness sufficient to achieve the resolution desired in the reconstruction. Each section is reconstructed using tomography. Subsequently, the resulting volumes in the series are aligned and merged into a single volume. We recently developed two programs, MOG and TRANSMOG that are useful in the alignment of serial volumes. Fiducial marks are assigned to features in the raw or segmented sections from the consecutive volumes using the NCMIR fiducial marking tool, FIDO. MOG computes the polynomial function coefficients necessary to align the two sections, including a correction for in-plane warpage that may have occurred between the two volumes. The translations computed by MOG are then applied by TRANSMOG to the remaining sections in the volume. These have programs are also useful for aligning and scaling correlated three-dimensional light and electron microscope data and have been used for this purpose on a collaborative study examining alterations in cardiac muscle in a model of heart failure. Further, this software has also been found useful for assessing changes in structures due to shrinkage. The NCMIR program MOG can be used to obtain numeric estimates of changes in single sections resulting from beam exposure or specimen preparation.
Showing the most recent 10 out of 384 publications
