Measurement of Nanoparticle Uptake by Alveolar Macrophages - Safety of Nanomaterials along Their Lifecycle - page 145

20 µ m 20 µ m

(a) (b)

20 µ m 20 µ m

(c) (d)

FIGURE 7.6 A moving macrophage ingesting settled particles. Four different positions
(a-d) are shown at time intervals of 2 min each. Arrows track the center of the macrophage,
whose contour is filled and overlaid in black.

A major issue during program development was to properly identify the objects
prior to their binarization. We found that the well-known procedures for object edge
detection such as Laplace or Marr-Hildreth operators delivered only poor results, as
they were unable to draw coherent outer contours due to the varying contrast around
macrophages. Therefore, a new algorithm needed to be developed, which expanded
the possibilities and functions of digital image processing. To identify the contour of
a macrophage (an object with more than 400 pixels) we developed a new algorithm.
This algorithm uses the center point of the macrophage area from where it automati-
cally constructs a set of radii. The intersection of each radius with a so far identified
object edge defines a set of reference points (Figure 7.7a). The edge of the object will
then be completed between the reference points. Therefore, a new search algorithm
was developed: starting simultaneously at two neighboring reference points, the
algorithm wanders alongside the edge of the object and completes the outer contour
upon collision. This algorithm considers the grayscale difference between the actual
pixel and its neighbors as well as the grayscale difference between both reference
points. This mode of operation allows us to draw complex contours and is visualized
in Figure 7.7c and d. It can be seen that the “clipping” of image data and the detec-
tion of island-like structures by the Laplace method are overcome by the algorithm.
Eventually, results from the algorithm based on reference points and Laplace filtering
were combined to compute the final contour, which was then used to calculate the
perimeter from single pixels (Figure 7.7b).