Chemistry Reference
In-Depth Information
particle tracking experiments, allowing the QDs to internalize into the cells after
binding.
Appendix 6.B: Software and Image Processing
6.B.1
Single Particle Tracking
Data for single particle tracking (SPT) were acquired using wide-
eld detection
as above. Video rate series consisted of 900 frames of 20-ms exposure at 33 frames/s.
QD
EGF labeling was performed in the presence of a kinase inhibitor, PD153035
that inhibits the activation and dimerization of the EGFR receptor [25]. A focal plane
with relatively low curvature in the area at the top of the cell was chosen for imaging.
We devised an automated method for selecting QDs for tracking and analysis that
t
high intensity regions to a Gaussian pro
le approximating the point spread function
(PSF) of themicroscope (see text). The quality of the
t was used to determine if a spot
was chosen for analysis. This allowed batch processing of many acquired series. The
low labeling and the selection of in-focus QDs yielded 1
-
10 tracked QDs per video
series. The tracking routines were written in DIPimage (TU Delft, www.qi.tnw.
tudelft.nl/DIPimage) a toolbox for Matlab (The Mathworks, Massachusetts, USA).
The tracking was performed of
ine after acquisition.
-
6.B.2
Real Time Optically-sectioned Imaging with the PAM
In the LCoS-based PAM, the conjugate and the non-conjugate images are recorded
simultaneously on the same CCD camera side-by-side. To yield a properly optically
sectioned image, subtraction of these images is required. Before acquisition, the two
images are registered by optimizing the two-dimensional correlation coef
cient by a
search procedure using a
fixed step size. The registration parameters are translation
(using a step size of 0.1 pixels), rotation (step size 0.001 radians), and magni
cation
(step size 0.001). During acquisition, a background image is
first subtracted from the
image pair. Then, the non-conjugate image is transformed to overlap the conjugate
image by applying the established registration parameters. For subpixel transforma-
tions, bilinear interpolation is used. The images are then subtracted using a
weighting factor which is dependent on the duty cycle of the pattern used for
acquisition [13]. The
final image is scaled, and offset removed. If desired, a Gaussian
filter with an adjustable sigma is applied. In order to perform the transformation and
subtraction with suf
cient speed, it is carried out on a NVidia Quadro FX 4400 GPU
board using the BrookGPUGPU library [22] and DirectX 9 runtime. The GPU board
was coupled to an Intel Xeon 3.2GHz processor. By of
oading the image processing
to the GPU, the computation time is less than the fastest possible PAMexposure time
required for a full scan (
16ms). Interaction with the PAM is thus real-time, i.e.
optically-sectioned images are displayed at video rate on the screen.
