September 4, 2009Kaela Leonard
| Kaela Leonard - Statistical Dynamics of Flowing Red Blood Cells by Morphological Image Processing | | | | Statistical Dynamics of Flowing Red Blood Cells by Morphological Image Processing
John M. Higgins, David T. Eddington, Sangeeta N. Bhatia, L. Mahadevan
PLOS Computational Biology, vol. 5 issue 2, 2009
Presented by: Kaela M. Leonard
Friday, September 4th
Summary:
This article aims to study the dynamics of individual cells to explore the process by which cells in blood vessels coagulate. Coagulation occurs when blood cells form a clot. Within a blood vessel this is known as thrombosis and is the body’s means of repairing a damaged blood vessel. If too much coagulation occurs this can lead to an embolism, which is when the clotting cells become disengaged from the vessel and travel to another site in the body where they block the blood vessel. This process is not very well understood in terms of how the characteristics of the blood and the individual blood cells affect the likelihood of coagulation. For their experiments this group designed a microdevice that had a channel of 12 microns wide so that single cell tracking would be possible. They then used computational image processing (code written in MATLAB) to track the movements of the blood cells under various conditions.
The researchers found that the flow of blood can be characterized as a diffusion process with a diffusion coefficient of D. This coefficient is altered by changes in the bulk velocity of the blood or the stiffness level of the individual cells. This lead to the conclusion that those with high blood velocity or stiff cells are more likely to experience coagulation at a flow rate that normal blood wouldn’t. To further make this point they used the analogy that “hot” blood is less likely to coagulate than “cold” blood.
Outline:
1. Abstract
2. Introduction
a. Red blood cell facts
i. Radius of 4 microns, thickness of 1-2 microns
ii. Very small equilibrium diffusivity (.1 square microns per second)
iii. Essential to study the dynamics of individual cells as well as cells in a group to properly understand the process of coagulation or thrombosis
b. Article aims to provide quantitative evidence of heterogeneity in velocity and density
i. This heterogeneity could play a role in coagulation and thrombosis
c. Computational Image Processing
i. Used to segment and track individual cells
ii. Videos all captured at high spatial and temporal resolution in microdevice
iii. Cell trajectories (25 million steps) measured over 500,000 images
iv. Figure 1 shows the experimental setup as well as sample particle tracking images- shows not just the particle trajectory but also the relative velocity (needs scale bar)
v. Figure 2 shows the segmentation process. Thought this was great image that really showed how the software works and how it recognizes cells. A little confused about the dimensions though because the width of the channels appears to be a lot more than 12 microns.
d. Microfluidic Device
i. 12 micron dimension confines cell movement in one direction
ii. Primary advantage is ability to see cells easily any increase in channel size would make particle tracking difficult
iii. Parameters that were chosen are relevant to common physiological parameters
iv. Data collected from middle 5th of long 250 micron channel
3. Methods
a. Video Acquisition
i. Captured under controlled oxygen conditions
ii. EDTA containers
iii. Hematocrit levels between 18-38%
iv. Videos captured at 60 frames per second, resolution of 6 pixels per micron (we should get this second number for our publications)
b. Image Segmentation (of particular interest given current involvement with automeasurement program provided by Zeiss)
i. Figure 2 shows great images of how this segmentation process occurs
ii. Code written in MATLAB
iii. Algorithm uses “marker-controlled watershed segmentation”
iv. Annular cells- fillable cells not touching border
v. This section was very confusing. Could anyone make sense out of what they were saying?
c. Tracking Between Frames
i. For subsequent images, cells were ranked by changes in size, shape and location for likelihood of coming from a given parent cell
ii. Maximum changes in x,y position calculated based on flow rate
iii. Maximum changes in area, perimeters and shape determined by tracking using Adobe Photoshop as a check on the MATLAB code
iv. Median inter-frame displacement calculated and any cells greater than 5 times this were excluded and program was re-run
v. This was redone excluding cells greater than 2 times the median inter-frame displacement
d. Assessment of Calculated Cell Velocity
i. Inaccuracies come from: error in cell location and error in tracking event between two cells How else could they have minimized this?
ii. Cell location- chose centroid as the single pixel to identify a cell by
1. Encountered the same problems I have with varying image intensity due to lighting and the focal plane
2. Found that the centroid could vary 5 pixel lengths in any direction
iii. False positive tracking event
1. Optimized segmentation algorithm
2. Applied heuristics
a. Evaluated in two passes- wide tolerance follow by a much tighter tolerance
3. Velocities compared to results from a random composite video such that any tracking in this video would be a false positive
a. Video rarely yielded more than 10 tracking events
b. To be conservative they used 20 as the number for false positive tracking events and excluded video from data set if it included less than 20 events
4. Also looked at velocity compared to known bulk velocity to notice any discrepancies
iv. Noise
1. Noise is more prevalent in shorter trajectories
2. Will average out over longer trajectories
3. What does this mean experiment-wise?
e. Measurement of 2D Cell Density
i. Projected density measured by thresholding grayscale intensity images and combining them with foreground cell markers
ii. Density measurement stable over time with variation coefficient of 10-25%
4. Results
a. Figure 3a- Needs better explanation. I can’t tell what the difference is supposed to be between the three graphs
i. Supposedly shows that the effective diffusion constant is much larger than the equilibrium one
ii. Mean square displacement isotropic at short times and anisotropic at long times
b. Figure 3b- no consistent effect on cell dynamics- this can be easily seen because the error bars are huge
c. Figure 3c- not sure how this figure lead them to the conclusion that “typical flow is diffusive”
d. Diffusion process typically has characteristic length scale corresponding to distance cell travels before interaction
i. In system this can be determined by cell size, cell separation or cell distance from boundary
ii. Length scale will change based on density, cell geometry and cell size
iii. When boundary is infinitely far away the length scale scales with the radius of the particle this won’t work for this system because the boundary is too close
iv. Remaining possibilities are:
1. Distance between cells (~3microns)
2. Height of channel (12microns)
a. Used to calculate local shear and effective diffusivity
v. Seem to switch back to calculating the characteristic length scale based on distance between cells later
vi. Found this section extremely confusing- maybe needed more information or better explanation
e. Figure 4a- shows diffusion coefficient as a function of velocity. Found at long time scales they can characterize this as a diffusive process
f. Investigated the effect cell shape/stiffness on diffusive process
i. Used sickle cell patients because cells become stiff deoxygenated environment (this can increase risk of vaso-occlusive events)
ii. Figure 4b shows that stiff cells have smaller diffusion coefficient at same bulk flow rate meaning the dimensionless constant C is smaller for deoxygenated cells
iii. This fits with the diffusion coefficient being inversely proportional to frictional drag and stiffer cells having a higher frictional drag
5. Discussion
a. Changes in bulk velocity or stiffness will change D and control velocity fluctuations
b. Velocity fluctuations may be shown as suspension temperature (why?)
i. Slower flow and stiffer particles will have higher suspension temperature
c. Figure 5- Odd way to show a probability graph… can’t make too much sense of it?
i. Longer tails than Maxwell-Boltzmann distribution?? How are they seeing this?
d. Can use analogy of “hot” blood being less likely to coagulate than “cold” blood
e. Virchow’s Triad characteristics: stasis, endothelial dysfunction and hypercoagulability their results offer explanation of why blood with stiff cells and small flow rate with coagulate at a flow rate when normal blood would. What does this mean for people?
f. Indentified “random, walk-like” behavior for particles in pressure drive suspension
g. Velocity quantified in terms of: blood flow rate, shape, stiffness
h. Next step: “well-defined microscopic mechanism”
Recommendation:
Accept with minor revision on the wording and explanation of certain sections. Figures were large and for the most part pretty good. Would have been nice to see a better explanation of cell image segmentation because this is the section that is of particular interest to me.
Definitions:
1. Non-brownian: Brownian motion is the random movement of particles while suspended in a fluid. Red blood cells are known as non-brownian because they do not experience random movement in their suspending fluid.
2. Thermal diffusivity: Ratio of thermal conductivity to volumetric heat capacity often given as k/cp
3. Coagulation: Process by which red blood form clots in the human body. This process is known as being complex.
4. Thrombosis: The formation of a specific blood clot that occurs in a blood vessel, which can prohibit or weaken the flow of blood through that vessel. If too much clotting occurs and the clot breaks free of the vessel, it is known as embolism.
5. Embolism: Is when one blood clot migrates from one body site to another vessel and causes a blockage in that vessel.
6. Heterogeneity: Descriptor used to talk about a system, which has many items consisting of differing variations. This is the opposite of a homogeneous system.
7. Fahraeus effect: ? Couldn’t find a good definition for this
8. Isotropic: Same in all directions
9. Anisotropic: Not the same in all directions
10. Vaso-occlusive: Resulting from the bringing together of two sides of blood vessel
11. Maxwell-Boltzmann distribution
12. “Marker-controlled watershed segmentation”: http://www.mathworks.com/products/image/demos.html?file=/products/demos/shipping/images/ipexwatershed.html
| | | | Journal Club Home |
|
|