|
|
| |
| |
| Live-Cell Imaging and Automation of Image Analysis |
| |
Technological advances in microscopic instrumentation and fluorescent labeling of proteins in vivo grant entirely novel insights into living cells and tissues. While better microscopes permit the acquisition of 3D time-lapse at movies increasing spatial and temporal resolution, the variety of fluorophores allows the monitoring of multiple biological processes in parallel. The explosion of data associated multi-dimensional imaging (Single time-lapse recordings can produce data in the triple-digit gigabyte range) pose novel computational challenges, both on the hardware and software side.
We are interested in studying animal development using 3D time-lapse microscopy and computer vision. Our principal goals are to develop protocols for live-cell imaging and software tools for the automated analysis high content microscopy data. Our computational pipeline comprises preprocessing, segmentation, feature extraction and classification (Figure 1).
Our system is currently directed at the phenotypic characterization of two biological processes in the model system Drosophila melanogaster; (1) Cell cycle progression of embryonic cells and (2) apoptosis and remodeling of muscle cells during metamorphosis.
|
| |
Figure 1 |
| |
Live Cell Imaging
We are focusing on two problems in developmental biology that we observe by live video microscopy. In embryos, we aim to track cells as they proliferate, measure DNA content and classify the phases of the cell cycle (Figure 2). The detection of abnormalities in cell cycle progression and mitotic mechanics will be useful for the characterization of gene function and genetic screens. Our second biological theme is the destruction and remodeling of muscles in metamorphosis (Figure 3). Studying the dynamics of muscle remodeling might prove a useful model muscle atrophy as the restructuring from larval to adult muscle is accompanied by alteration in size of the muscle fiber. These observations of real biological processes in tissues serve as vehicles for the development of our software tools. As we cannot easily obtain those data, we need to produce them ourselves. |
| |
Figure 2 |
| |
Scene Interpretation of Live Cell Movies using Computer Vision
The faithful segmentation (Figure 2) of object regions is a prerequisite for accurate feature detection and measurement. A good part of our research will be to design solutions that integrate preprocessing (deconvolution, noise reduction), segmentation and postprocessing such as morphological operators. However, after converting arrays of pixels into objects and extracting their feature values, we need to assign biological meaning to them. Object classification will help us to identify the actors in our movies. It is obvious that we need to apply statistical methods such as data mining and machine learning to filter out the interesting bits of knowledge in a sea of trivial information. |
| |
Figure 3 |
| |
Other Projects
- 4-dimensional imaging of cell cycle progression in Drosophila embryogenesis and quantification of cellular parameters
Ms. Puah Wee Choo
- Quantifying the performance of segmentation algorithms for 3D microscopy
Dr. Janos Kriston-Vizi
- Algorithm development for preprocessing and segmentation of 4D microscopic images
Dr. Chinta Rambabu
- Voxel to mesh conversion of microscopic 3D time-lapse image data
Dr. Chinta Rambabu
- Registration of 3-dimensional image stacks showing rapidly contracting muscles
Dr. Du Tiehua
- Tracking and Classification of dividing cells in 4D image datasets
Dr. Du Tiehua
- 5D live cell imaging of muscle destruction and remodeling during Drosophila metamorphosis
|
|
|
