Automated Alignment and Segmentation for Electron Tomography

点击翻译按钮获取中文摘要

0241182Ji The current project addresses two procedural bottlenecks that limit high throughput electron tomography: image alignment and volume segmentation. Presently, image alignment requires the use of fiducial markers and is relatively labor-intensive and error prone. An information theoretic procedure is proposed that performs automatic image alignment without fiducial marks. The proposed solution exploits the rich information content in tomographic images and explicitly accounts for the unique transformation between images. Volume segmentation is one of the most crucial, yet labor-intensive, time-consuming, and subjective steps in electron tomography. Automated segmentation of tubular objects can be very challenging because of extremely low object contrast relative to the surroundings, a highly uneven and irregular surface topology, and significant variations in cellular attachments. The proposed segmentation approach consists of dividing the problem into three major tasks: 1) tubular structure enhancement; 2) tubular structure detection; and 3) tubular surface morphology reconstruction and detection of end structures. Towards the first task, a model-based filter is proposed that will enhance cylindrical structures while de-emphasizing the irrelevant structures. Towards the second task, a robust feature detection technique is proposed to localize the tubular portion of fibers. Towards the third task, a statistical local region growing technique is proposed that will grow the detected underlying cylinder in all directions to produce the surface morphology and internal discontinuities of the actual structure. The proposed methods will have wide applicability to electron tomography and other forms of medical imaging. The alignment methods are completely general and the prevalence of microtubules and similar tubular or fibrous motifs in cellular and medical imaging ensures wide applicability of the segmentation methods. With appropriate modifications, the proposed segmentation methods could be adapted to membrane and vesicle geometries. By addressing the two bottleneck steps, the proposed methods have the potential to transform electron tomography into a powerful, routine tool for research and diagnostic investigations. These methods also have potential application to other medical imaging projects. Thus, this project is designed to enable electron tomography to realize its full potential for analysis of subcellular structure and function in the post-genomic era. This grant is made under the Joint DMS/NIGMS Initiative to Support Research Grants in the Area of Mathematical Biology. This is a joint competition sponsored by the Division of Mathematical Sciences (DMS) at the National Science Foundation and the National Institute of General Medical Sciences (NIGMS) at the National Institutes of Health.