February 15, 2013
Here are some movies of HL60 cells moving as seen by Bessel beam 3-d microscopy. Movies were made with Chimera. Data from Lillian Fritz-Laylin, Jenny Hsiao, and Dyche Mullins at UCSF.
Cell 7 movie (11-07-12) top view. Used deconvolved files (*.deconv.am). Cell goes partly out of view at top of slab. Deconvolution seems to create high density where cell is cutoff, possibly by including density values of 100 from outside slab.
Cell 7 movie (11-07-12) side view. Used deconvolved files (*.deconv.am).
Gray level cell 7 movie (11-07-12) made by Janelia Farm group, probably with Amira. Chimera can make this type of movie but I did not try it.
Cell 6 movie (11-07-12) with moving parts colored red, non-moving blue. Coloring is done by measuring the correlation between the image at the current time and the image at the next time over a sliding window 5 by 5 by 5 grid points in size. Colors were blue for correlation 0 (or less), white for correlation .5, and red (actually salmon) for correlation 1. Colors were linearly interpolated at other correlation values. Oddly the cell seems stuck and does not move from its original position. Used deconvolved files (*.deconv.am).
Cell 9 movie (11-08-12) two channels, channel 1 shown transparent blue, channel 2 shown opaque pink. Did not use deconvolved files (*.deconv.am) because they have varying normalization at different times. Used "translated" .tif files. Filopodia are to some extend smeared out and not seen in non-deconlved data. Microscope stage moved to keep cell in view. Cell reverses direction of motion. Cell goes partially out of slab field of view.
|Cell 6 stereo movie (11-07-12). Left and right eye movies can be played in a stereo movie player, such as Bino. It can be viewed with LCD shutter glasses or other stereo viewing techniques such as cross-eye or wall-eye stereo pairs, or red-blue glasses. Images in this page show a cross-eye pair (right eye image on left, left eye image on right).|
I reduced the size of the data sets so that they can be viewed and played back in real time. The original data was approximately 250 Mbytes per frame with 150 to 200 frames so one time series (with one channel) was is about 40 to 50 Gbytes. At typical disk read speeds of 10-20 seconds per Gbyte this takes a long time (10 minutes) just to read the data off disk and does not fit in memory on typical desktop computers.
I reduced the data size about 100-fold using the observation that the cells only occupied about 1% of the 3d grid (based on an actual measurement of cell 7 in Chimera). I thresholded the data, setting all values below a chosen low density level to that level, then compressed the data using the Chimera "cmap" format based on the HDF5 file format. The thresholding is needed otherwise the random noise does not allow significant compression. I also reduced the 32-bit float data values to 16-bit integer to save a factor of two more in file size. This causes no noticable loss in precision as the microscope CCD certainly does not exceed this dynamic range. This reduced the data to typically 0.25 Gbytes which reads quickly and fits in memory.
|Edge view of several frames of cell 7 (11-07-12) in different colors showing imaged slab. Green box was used to crop data to reduce file size for quicker display in Chimera.||Top view of same cell 7 frames showing that cell is partly out of view (clipped) at the top.|
Here is the Chimera script compress.py I used to convert the data sets to compressed cmap format. All time frames are placed in a single .cmap file for one channel of one time series. Here is an example command to run this script from a terminal window.
chimera --nogui --nostatus --script "compress.py /Volumes/UCSFdata/JFRC_2012_NOV/11-07-12/cell10_zp4um/*.am cell10.cmap 10,231,0,470,571,180 90 int16"
Converting a data set took about 1 hour on a Mac desktop.
The compressing script allow cropping to a subregion. I inspected several frames in a time series to see if a cell stayed within a space much smaller than the data grid. If so I used Chimera subregion selection to find the reduced bounds. This does not significantly reduce file size, but reduces the memory and time consumed when individual frames are decompressed in computer memory. Here's some info on the bounds I chose.
Chimera commands to make movies looked like
movie record supersample 3 vseries play #0 wait 150 movie encode ~/Desktop/cell6.mp4 framerate 8 quality high roundtrip true
and were executed by opening a command file such as record.cmd. The first command starts capturing images for a movie. The second starts playing the image series, the third waits for 150 frames to elapse, and the last command writes an H.264 format movie for playback at 8 frames per second. The "roundtrip true" option has the movie run the captured sequence forward and then backward.movie record supersample 3 ; vseries play #0 ; wait 150 ; movie encode ~/Desktop/cell6.mp4 framerate 8 quality high roundtrip true