Geometric context driven inference for high throughput cryogenic electron tomography
- Cryogenic Electron Tomography (Cryo-ET) has gained increasing interest in recent years due to its ability to image whole cells and subcellular structures in 3D at nanometer resolution in their native environment. However, due to dose restrictions and the inability to acquire high tilt angle images, the reconstructed volumes are noisy and have missing information. In order to overcome these limitations and fulfill the promise of this method, it is necessary to image numerous instances of the same underlying object and average them, requiring a high throughput pipeline. Furthermore, recent advances in microscope automation have increased the data generation capacity of this method by many times, placing additional strain on the postprocessing portion of the electron tomography pipeline. Currently, the bottlenecks in this pipeline are a set of image inference tasks which require manual intervention by an expert due to weak and unreliable local image features. In this thesis we propose the use of geometric context in a structured probabilistic models framework to overcome the low reliability of local features and achieve automation and high throughput for two of the bottleneck tasks---precision registration of 2D images and 3D segmentation of whole cells. The central idea in our approach is to overcome the uncertainty from unreliable features by exploiting their mutual geometric and spatial relationships in varying degrees of locality to classify them more accurately. Structured probabilistic models provide a framework for encoding a diverse set of geometric relationships, as well as a substantial body of efficient yet effective approximate inference algorithms. In the first problem of precision registration of 2D images, the features are a set of gold markers which can be difficult to distinguish at high tilt angles. Precision alignment of the images requires the successful tracking of these markers throughout the series of images. We track markers jointly as a group, using their geometric relationships. Therefore the geometric relationship of interest for overcoming the unreliable features in this case is the pattern formed by the gold markers. We encode the relative geometric arrangement of pairs of markers as pairwise factors in a Conditional random field (CRF) framework, and use loopy belief propagation to find the most likely correspondence of markers between images. This approach, called RAPTOR (Robust Alignment and Projection estimation for TOmographic Reconstruction) has resulted in successful automatic full precision alignment of electron tomography tilt series. The second problem of 3D segmentation of whole cells is challenging due to uncertain boundary characteristics. Intensity and intensity gradients based methods easily confuse many non boundary pixels as boundaries, and therefore precision extraction of the cell boundary is difficult, manual and time intensive. We present an efficient recursive algorithm called BLASTED (Boundary Localization using Adaptive Shape and TExture Discovery) to automatically extract the cell boundary using another CRF framework in which boundary points and shape are jointly inferred with the help of a learned boundary feature detector and shape evolution model. The algorithm learns the texture of the boundary region progressively, and uses a global shape model and shape-dependent features to propose candidate boundary points on a slice of the membrane. It then updates the shape of that slice by accepting the appropriate candidate points using local spatial clustering, the global shape model, and trained boosted texture classifiers. This method has successfully segmented numerous datasets starting from one hand labelled slice each, reducing the processing time from days to hours.
|Type of resource
|electronic; electronic resource; remote
|1 online resource.
|Moussavi, Farshid, (Engineer)
|Stanford University, Department of Electrical Engineering
|Horowitz, Mark (Mark Alan)
|Horowitz, Mark (Mark Alan)
|Statement of responsibility
|Submitted to the Department of Electrical Engineering.
|Thesis (Ph.D.)--Stanford University, 2010.
- © 2010 by Farshid Moussavi
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...