4. Select the “Box training references” tab. Browse and find the
imported, pre-processed tomogram, as indicated by the “_preproc.hdf” file extension. Select “Launch”. This action will open
the particle list, options, and the main windows.
5. In the main window, middle-clicking the tomogram will bring
up a GUI where contrast, brightness, and magnification can be
adjusted. Typically, the auto contrast and a magnification of 1.0
works well. On the right side of the tomogram display window
is a slider for scrolling through tomogram slices. After boxing
particles for training, their location within the tomogram will
be displayed here.
6. Begin boxing “positive particles”, which contain the feature
(s) of interest (FOIs), by left-clicking FOIs in the “Main Window”. Use the slider to scroll through the tomogram and pick
particles from different Z-heights. The recommended minimum is 10 but more is better to a certain extent. We typically
choose between 20 and 50 particles. The most important
principle is to pick particles that include the variety of shapes,
sizes, angles, and cross-sections that fully represent the FOIs.
This is critical for segmenting the features fully in all dimensions because the software segments the FOIs within individual
two-dimensional slices of the tomographic data.
7. In the “Options” window, select the set “00” and rename it
“ptcls_good”. Be sure the check box next to the “ptcls_good”
set is ticked, and select “save” to output the particle stack file.
The filename suffix that displays should be “ptcls_good”.
8. Close “Main Window” and select the “Segment training references” tab. Browse for the “good” particles file and click
“Launch”. The purpose of this step is to use the pen tool to
“paint” over the features in each boxed particle. This step is
critical because the hand-segmented data is used during neural
network training to define which features it will consider for
segmentation. We find it is easier if you zoom in and typically
place the magnification at ~2.0 or higher for this step. The pen
size should be chosen based on the features being segmented
and on the fineness of the segmentation needed. Smaller pen
sizes give you finer control but take longer to paint large areas
of the tomogram (see Note 19). Paint by clicking the left
mouse button and dragging the cursor across the feature. Use
the arrows on the keyboard to move between particles. The
program autosaves for this step, so simply exit when done.
9. Again select “Box training references”, and click “Launch”. In
the “Options” window, create a new set and name it
“ptcls_bad”. Now that more than one particle set has been
created, newly boxed particles will be added to whichever set
is highlighted in the “options” window. Highlight the
Cryotomography of Neurons
41
imported, pre-processed tomogram, as indicated by the “_preproc.hdf” file extension. Select “Launch”. This action will open
the particle list, options, and the main windows.
5. In the main window, middle-clicking the tomogram will bring
up a GUI where contrast, brightness, and magnification can be
adjusted. Typically, the auto contrast and a magnification of 1.0
works well. On the right side of the tomogram display window
is a slider for scrolling through tomogram slices. After boxing
particles for training, their location within the tomogram will
be displayed here.
6. Begin boxing “positive particles”, which contain the feature
(s) of interest (FOIs), by left-clicking FOIs in the “Main Window”. Use the slider to scroll through the tomogram and pick
particles from different Z-heights. The recommended minimum is 10 but more is better to a certain extent. We typically
choose between 20 and 50 particles. The most important
principle is to pick particles that include the variety of shapes,
sizes, angles, and cross-sections that fully represent the FOIs.
This is critical for segmenting the features fully in all dimensions because the software segments the FOIs within individual
two-dimensional slices of the tomographic data.
7. In the “Options” window, select the set “00” and rename it
“ptcls_good”. Be sure the check box next to the “ptcls_good”
set is ticked, and select “save” to output the particle stack file.
The filename suffix that displays should be “ptcls_good”.
8. Close “Main Window” and select the “Segment training references” tab. Browse for the “good” particles file and click
“Launch”. The purpose of this step is to use the pen tool to
“paint” over the features in each boxed particle. This step is
critical because the hand-segmented data is used during neural
network training to define which features it will consider for
segmentation. We find it is easier if you zoom in and typically
place the magnification at ~2.0 or higher for this step. The pen
size should be chosen based on the features being segmented
and on the fineness of the segmentation needed. Smaller pen
sizes give you finer control but take longer to paint large areas
of the tomogram (see Note 19). Paint by clicking the left
mouse button and dragging the cursor across the feature. Use
the arrows on the keyboard to move between particles. The
program autosaves for this step, so simply exit when done.
9. Again select “Box training references”, and click “Launch”. In
the “Options” window, create a new set and name it
“ptcls_bad”. Now that more than one particle set has been
created, newly boxed particles will be added to whichever set
is highlighted in the “options” window. Highlight the
Cryotomography of Neurons
41