Visual Analysis on Red Blood Cells
Dong Hyun Jeong
The
In biological insight, all the red blood cells located in
human body can be tracked to find the problems or happens when some of diseases
exist on their body. In such case, the tracked blood cells can have significant
changes in terms of the number of blood cells, velocity, acceleration, etc. As
a preliminary procedure, biologists take pictures of the blood cells. From the
taken pictures, several approaches can be made to analyze the data and find the
significant factors. But analyzing thousands of time-series images are tedious
and time-consuming job. To remove the difficulties, we have designed visual
analyzing tool, called RBCVis. As an initial step, several different image
processing techniques are applied in order to find and track the red blood
cells. Also their diameter is analyzed. In a second step, all the analyzed
datasets are used in the designed visual application. With using the
application, user can easily track the red blood cells and blood vessels. Also
to increase visual perception while analyzing the datasets, HSB color space is
used to directly map with parametric information.
3Tracking RBCs
To track
RBCs, image stabilization technique and temporal matched filter are used. The
motion caused when the animal breathing itself has to be removed as a
preprocessing procedure. Even though the datasets with a very large degree of
motion are discarded that results in a significant blurring, most images still
have some movement in images. First, removing the movement has been performed by
recovering the 2D translation from the reference frame based on the following
observation. As a second step procedure, Temporal Matched Filter is used to
track red blood cells. In general, a matched filter describes the appearance of
a searched signal. The kernels of the matched filter have been designed to
detect the temporal changes of the intensity in given datasets. Also to detect
the cells traveling at different speed thus occupying a range of number of
frames (F), a set of matched filters are designed. Through analyzing the datasets
with designed kernels and matched filters, the best matched kernel and filter
have been selected. Also all the tracked datasets have been analyzed further with
finding the motion behavior of red blood cells. To understand this methodology,
contact Dr. Min Shin, CS, UNCC.
System Layout
Tracking
red blood cells and vessels are performed in order to find important features
and characteristics when injection has been applied to animals. Hence, red
blood cells or vessels tracking are performed before and after injecting a
medicine to animals. Because of the procedure, two different time-series datasets
have to be analyzed having a comparative analysis form. From the tracked datasets,
the application has to be designed on finding important features or unveiling
hidden features. Also time-consumption
has to be considered minimizing it when analyzing the datasets.
To
support comparative analysis on red blood cells, RBCVis designed having a
capability of comparative analysis by showing more than two grouped datasets at
the same time in each individual window. Each window has its own setting,
dataset and interactive layout. Therefore user easily can compare and analyze
more than two datasets at the time by interactively changing the given
attributes
The overall layout with supporting interactive analysis. Two different datasets including interactive windows are shown.
Interactive
analysis supports easily finding important features and showing the detail
information of the selected object or region. Most analyses of large and
complex data have an exploratory component. This principle has been
encapsulated in the dictum, 'Focus + Context' from information visualization
[Pir01]. When analyzing the datasets, user might be get interested on seeing
specific regions. Zooming capability is necessary feature increasing the high
interactivity with the datasets. RBCVis is designed based on Pad [Per93] and
Pad++ [Bed94, Fur95] technique which supports zooming capabilities including
panning function. Simply user can
zoom in or out by clicking left or right mouse button on an interested region.
The
basic layout of tracked red blood cells and vessels and the view when zooming
is applied.
Finding the
number of red blood cells located in each vessel
In order to find the number of red blood cells located in
each blood vessel, we have used the solution by Philippe Reverdy [Bou87, Wal99].
The solution compute the sum of the angles made between the test point and each
pair of points making up the polygon. If this sum is 2pi then the point is an
interior point, if 0 then the point is an exterior point. This also works for
polygons with holes given the polygon is defined with a path made up of
coincident edges into and out of the hole as is common practice in many CAD
packages.
Polygon Interior Testing. By geometry, a 2D point lies on a polygon if the angle between it and all of the outward normal vectors on the polygon's edges are greater than 180 degrees.
Direct and Indirect selection
To
support interactive visual analysis, all the red blood cells and vessels are
managed as objects. Therefore, each individual object is selectable.
RBCVis
has two different selection mechanisms: direct and indirect selection. Each
individual object is selectable by directly dragging the mouse over the
displayed object. Instead of using direct selection, each object can be selected
indirectly with considering its detail attributes. This feature is useful when
user only focused on a certain object having higher velocity or larger diameter
size.
Directly
selected blood vessels (left) and red blood cells (right).
To
support indirect selection technique, RBCVis has a selection window. In the
selectable window, user can simply select individual red blood cells and
vessels. To support individual cell and vessel selection, all the information
have been managed as an object. Additionally, the red blood cells are managed
two different ways such as an individual object or a group. The group means
that a partial number of red blood cells are tracked as the same cell in
captured time-series images. This feature is necessary because all the red
blood cells have to be tracked in order to find their attributes such as
velocity, acieration, etc. to analyze and find the differences on different
datasets. Hence, all the red blood cells are managed individually or as a
group. When user chooses interesting data by clicking the checkboxes, it
directly affects the displayed layout by changing the color information. To
show the color information, all the attributes of red blood cells and vessels
are managed with referencing the defined color mapping. About six different
information (red blood cells’ velocity and acceleration, blood vessels’
velocity, acceleration, diameter, and number of red blood cells existed in
them) can be directly mapped with color information to represent the data with
a certain kind of color.
Selection
window having two pages; RBCs (left) and Vessels (right).
When
direct or indirect selection has been made, the detail information of selected
object is displayed in the selection window.
Color mapping
As
mentioned, there are several attributes are used to show as individual object’s
features. To increase human’s visual perception which helps distinguishing more
than two different datasets, color mapping method is used. The basic color
mapping function is defined with HSB color space model [Hoff01]. When colors
are displayed on a computer monitor, they are usually defined in the RGB (red,
green and blue) color space. A way of making the same colors is considering red,
green and blue as the X, Y and Z axes. Another way of making the same colors is
to use their hue (X axis), their saturation (Y axis) and their brightness (Z
axis). This is called the HSB color space. Generally
Color space model and a designed color selection window used for defining the color mapping function.
Each
attribute has its own color mapping function. Based on the defined color
mapping method, each object can be color coded to increase visual perception.
Color mapped images. Mapping with velocity
of red blood cells (left) and diameter of blood vessels (right).
Closed Vessel Region
Tracked
datasets consist of positional information of red blood cells and blood vessels.
Especially the blood vessels are tracked with considering the centerline of
each vessel and its relative left and right location. Visualizing the blood
vessels with given tracked datasets, it can create a vessel outlines. But analyzing
the datasets is not as much efficient. In human visual processing, human can
process the limited amount of information at a time [Healy, Tre98]. Also a
limited set of visual properties such as color, shapes, layout, etc. are
detected instantly by the human’s low-level visual system. To increase the human’s
visual perception - preattentive, blood vessel information has been designed
showing as closed vessel region display.
Closed
vessel region with diameter color mapping method (left) and with the number of
red blood cells located in each blood vessel (right)
Evaluation
With the designed application,
we did a simple evaluation by conducting two people. They have enough knowledge
about red blood cell tracking methodology and its major focusing on finding
important features. Simply we showed RBCVis to them and give short explanation
how to use it. After 5 minutes, we have requested two tasks. One is finding the
highest velocity of red blood cell and its relative blood vessel number in
which the red blood cell located. The other is finding highly compacted vessel
region with red blood cells and its diameter. As a comparative analysis, we
also asked finding the same information using Microsoft Excel having the same
datasets.
In
result, they only spent less than 1 minute (average 20 second) to finish up the
given tasks and find the requested object. Even if they have spent almost the
same amount of time on finding the specific object, they mentioned that they
can find easily a certain object and its relative information easily when using
Microsoft Excel, but it is difficult to understanding its spatial layout what
it looks like and where the object is exactly located.
Future Work
As a
future work, we are going to integrate all different color values showing the
several features at the same time. Even though there is a color attribute
overlapping problem can be existed, having different color mapping method can
remove such problem. Also to minimize the overlapping problem, changing
brightness can be used. Furthermore, the used color mapping method has to be
evaluated carefully when all different information has been displayed.
Also
not just displaying original tracked information, pixel-based representation
method or other visualization techniques to represent the datasets as
abstracted manner might be useful to highlight important features. The
possibility of applying those techniques has to be carefully tested.
References
[Dob02]
J. G. G. Dobbe, M. R. Hardeman, G. J. Streekstra, J. Strackee, C. Ince and C.
A. Grimbergen, “Analyzing Red Blood Cell-Deformability Distributions,” Blood
Cells, Molecules, and Diseases, Vol. 28, Issue 3 , May 2002, Pages 373-384.
[Per93]
Perlin, K., and Fox, D., "Pad: An Alternative Approach to the Computer
Interface," Proc. ACM SIGGRAPH '93,
[Bed94]
Bederson, B.B., and Hollan, J.D., "Pad++: A Zooming Graphical Interface
for Exploring Alternate Interface Physics," Proc. ACM UIST '94,
[Fur95]
G. Furnas and B.B. Bederson. Scale Space Diagrams: Understanding Multiscale
Interfaces. CHI'95, pp. 234-241 (1995).
[Hoff01]
Gernot Hoffmann, “Color Order Systems: RGB / HLS / HSB”, http://www.fho-emden.de/~hoffmann/hlscone03052001.pdf
[Dong04]
Gang Dong; Damiano, E.; Smith, M.L.;
[Pir01]
P. Pirolli, S. Card, and M. Van Der Wege. Visual Information Foraging in a
Focus + Context Visualization. Proceedings
of CHI 2001, pp. 506-513 (2001).
[Healy]
Christopher G. Healey, Perception in Visualization,
http://www.csc.ncsu.edu/faculty/healey/PP/
[Tre98]
Treisman, Anne M., and Nancy G.Kanwisher, "Perceiving visually presented
objects: recognition, awareness, and modularity," Current Opinion in
Neurobiology, 8 (1998), pp 218-226
[Bou87]
Paul Bourke , “Determining if a point lies on the interior of a polygon,” November
1987, http://astronomy.swin.edu.au/~pbourke/geometry/insidepoly/
[Wal99]Robert
J. Walker Jack Snoeyink, “Practical Point-in-Polygon Tests Using CSG
Representations of Polygons,” Technical Report TR-99-12, http://www.cs.ubc.ca/labs/imager/tr/pdf/walker.1999a.pdf