Parallel Processing Strategies for Cell Motility and Shape Analysis - High-Throughput Image Reconstruction and Analysis - page 74

parent and child objects: single (no parent-no child), source (no parent-one
child), source-split (no parent-many children), sink (one parent-no child),
inner (one parent-one child), split (one parent-many children), sink-merge
(many parents-no child), merge (many parents-one child), merge-split (many
parents-many children) (Figure 3.16).
2. Segment Generation: Trajectory segments are formed by tracing the nodes
of the ObjectMatchGraph ObjectMatch Graph . A linked list of inner nodes
(objects/cells), starting with a source or split type node and ending with a
merge or sink type node, are identified and organized in a data structure
called SegmentList .
3. Segment Labeling: Extracted trajectory segments are labeled using a method
similar to connected component labeling. Each segment without a parent
is given a new label. Segments that have parents inherit the parents' la-
bels. In case of multiple parents, if the parents' labels are inconsistent, then
the smaller label (older trajectory) is kept and a flag is set indicating the
inconsistency. Trajectories are formed by joining segments sharing the same
label.
Cell Trajectory Validation and Filtering
Factors such as vague cell borders, DNA unwinding during cell division, noise,
fragmentation, illumination or focus change, clutter, and low contrast cause seg-
mentation and association problems, which result in missing or spurious trajectory
segments and false trajectory merges or splits. Filtering is performed at various
levels of processing to reduce the effects of such factors. At the image level ,small
objects that may be due to noise, background clutter, or imaging and segmenta-
tion artifacts are removed by using morphological operators. At the segmentation
level , use of coupling (Section 3.3.1) prevents false merges. At the correspondence
level , unfeasible cell-to-cell matches with distances above a threshold are elimi-
nated through gating. At the confidence level , matches with low confidence values
are pruned with absolute and relative prunings. A final validation and filtering
module analyzes trajectory segments using accumulated evidence such as temporal
persistence, size consistency, and area ratios; and to take appropriate action (i.e.,
removal or merge) for spurious or incorrectly splitted trajectory segments.
Figure 3.14(d) shows individual cell trajectories for a sample wound healing
image sequence. Segmentation of the cells [shown in Figure 3.14(b, c)] is accom-
plished by 4-level sets cell segmentation described in Section 3.3.1. Once trajecto-
ries are obtained various statistics can be obtained for individual or group of cells.
Besides trajectories, the tracking module supplies information on trajectory events
such as merges, splits, and disappearances. These events are useful for determin-
ing cell lineage and for computing mitosis (cell split) and apoptosis (cell death)
rates.
3.3.3 Distributed Cell Tracking on Cluster of Workstations
With the rapid increase in the amounts of data generated by biomedical applica-
tions, high-performance computing systems are expected to play an increased role.
Various high-performance computing strategies exist using systems ranging from