About Montaging

Montaging allows one to capture an array of multiple, overlapping frames automatically using either electronic image shift or stage movement. Each such frame is called a piece. Pieces are numbered and referred to by their position in X and Y within the montaged image (numbered from 1) and by their section number or Z value (numbered from 0). For example, in the first set of images making up a 2 by 3 montage, piece 2, 2 is the middle piece in the right column, and the Z value would be 0. All pieces in a montaged image have the same Z value.

Montaging can be initiated in several ways: by selecting New Montage or Montage Parameters from the File menu, by selecting Montage or Prescan from the Camera menu, or by pressing the Start or Prescan buttons in the Montage Control panel . Whichever way is used, you will then encounter a series of dialog boxes: the Montage Setup dialog box for defining the montage size, File Properties dialog for specifying how images are stored, and the Save As dialog box to specify the output file. Montaging can also be restarted on an existing file simply by reopening that file. In this case, you will enter the Montage Setup dialog box to see the parameters governing the montage, but you will not be able to change most parameters.

You should always calibrate the image shift at the given magnification before starting to acquire montages for a tilt series. See the Image Shift command in the Calibrate menu.

When montaging is started, the Montage Control panel opens up. At that point, a montage can be acquired by pressing the Start button there or the Montage button in the Camera & Script Control panel, or by using the Ctrl M hotkey or the entry in the Camera menu. Images are always acquired with the 'Record' camera parameters.

The Prescan option, available in control panels and the Camera menu, will acquire binned down images at all piece positions and use them to compose an overview image. These images are acquired relatively quickly and are not saved to file, so this is a good way to find out what features are included in a montaged area without taking actual montages.

To acquire a montage, the program shifts to each of the frames and acquires and saves an image. As it goes, it composes an 'overview' image, a single, binned down image of the entire montaged area. This image is left in buffer B. The different pieces may not be shifted into perfect registration when composing the overview, so this image may show some sharp transitions between the pieces. This should not be a cause for concern. There is an option in the Montage Control panel to have pieces shifted into register instead.

You can use the right mouse button to impose an alignment shift on an overview image to reposition the field of view.

As it acquires frames, the program also uses cross-correlation to measure the misregistration between each pair of overlapping pieces. When the whole montage has been acquired, it uses these errors in registration to determine how to shift all of the pieces so as to minimize the error. The program informs you in the Log window of the error before and after shifting pieces into best registration. The error before shifting can become high (more than 10 pixels) if there is drift during the acquisition or if image shift is not well calibrated. The error after shifting should be low (under 0.5 pixel) at low tilt angles, but may become relatively high (1-4 pixels) at high tilt because image distortions prevent pieces from fitting together well. A high error can also occur if there is an error in the correlation between some pair of pieces. The latter problem can be corrected afterwards in Midas. Another possible source for a relatively high error is distortion in the images. It is possible to correct this in IMOD also by measuring the image distortion field.

After acquiring a montage, the program also composes a 'center' image and leaves it in Buffer A. For a montage with an odd number of pieces in X and Y, this would just be the center piece, but in other cases it composes this image from halves of two overlapping pieces or from quarters of four overlapping pieces. The misregistrations between pieces are taken into account in fitting these pieces together, so the center image should not show sharp transitions and should be suitable for aligning from one tilt to the next.

You can reconstruct the overview and center pieces for a stored montage by simply selecting the Read command from the File menu. The program will go through the same sequence of operations using each piece from the file that it does with newly acquired pieces, leaving a center image in buffer A and an overview in buffer B and in the Read buffer.

If the range of image shift required to make a montage is large, you can use stage movements instead. This is specified by an option in the Montage Setup dialog box. Montages constructed with stage movement are useful for getting an image of a large area but will not fit together very well unless the option is selected to have pieces shifted into register. It may also be necessary to select the option to allow for very sloppy montages by correlating larger areas in the overlap zones. If there are grid bars in the overlap zones between images, the correlations can easily fail and either of these options may make things worse.

When montages are acquired with stage movement, it is possible to resume a montage that is stopped for some reason. To resume a montage, just start the montage by any of the usual means. The program will ask if you want to resume it or redo it completely. By default, it will restart where it left off. However, if you see that some images are bad and want to have them reacquired, select the Overwrite Pieces command in the File menu. Enter the number of pieces to be replaced.

Acquiring High-Quality Montages of Large Areas

Montages of large areas generally need to be taken with stage movement. If you are acquiring such montages for image analysis or 3D reconstruction rather than just for an overview or mapping in the Navigator, then you will want high-quality images. The Montage Setup dialog box allows you to enable a collection of settings for a high-quality stage montage. For a substantial data collection effort, you may need to experiment with these settings to find which ones you need.

Focusing is important for large montages. There is an option to run the autofocus routine on every frame. However, this is probably not necessary in many cases. For example, if the sample is locally tilted by 1°, then the Z height and thus the focus will change by only 0.18 µm over a distance of 10 µm. Thus, there is also an option to acquire the montage in successive square blocks of images, focusing only in the center of each block. On an FEI scope, this should save significant amounts of time. On a JEOL, the time savings may be less because of the slower stage and it may result in more variable overlaps between adjacent pieces.

If there is enough drift after a stage movement to impair the quality of the images, you may need to increase the delay between stage movement and image acquisition.  In addition, if you are focusing for each piece, you can select an option to have it repeat the autofocus if the drift is higher than a selected limit.

To deal with the problems of persistent stage drift over long periods of time as well as irregularities in stage movement, the program can realign to an existing piece with image shift before acquiring each piece.  This procedure has proven capable of producing a stage montage with very little variability in the overlaps between pieces, which allows the overlap to be kept at 10% or even reduced to as low as 5%.  In addition to dealing with drift, this realignment can overcome the effects of cyclical variations in compustage movement, documented by Pulokas et al., 1999.  Over a certain range of image sizes (around 4 microns), these effects can be large enough to cause inadequate overlap between adjacent pieces unless the nominal overlap is set very high (20-25%).  This option can thus be useful not only for very large montages but also for modest-sized montages with piece sizes in this range.  It is recommended that the option be used for acquiring a medium-magnification map upon which supermontages are to be placed, because it is important for such maps to have well-aligned pieces.  Additionally, the Navigator can use information about the stage positions of the individual pieces to achieve more precise positioning when moving to or realigning to a position marked on a map acquired with this method.  Thus, it can be advantageous to use this method for other medium-magnification maps, provided that the realignment works reliably with your specimen and that the increased acquisition time is acceptable.

With image shift realignment, pieces can be acquired in two different sequences.  One is the ordinary sequence, in columns starting at the lower left and finishing at the upper right.  This sequence is appropriate as long as the entire area being montaged contains image features suitable for correlation.  The other sequence starts at the center and works outward.  This sequence should be chosen if the edges of the area contain problematic features such as empty resin or grid bars.  With such a sequence, alignment failures on peripheral pieces will not throw off the alignment of the more central pieces where the material of interest is.  This sequence does require that the program realign to a lower-magnification image between the first and second halves of the montage.  For more details, see the description of these options in the help for the Montage Setup dialog box.

There is an option to skip the correlations needed to shift pieces into register. These correlations are of no real value if the data are to be processed elsewhere and can consume significant time in a large acquisition.

In stage montages, the effects of backlash are minimized by always moving the stage to a piece position from the same direction. For each column of pieces to be acquired, there is a backlash correction movement in the opposite direction. The amount of this correction is set by the property StageMontageBacklash, which defaults to 5 µm on an FEI scope, 10 µm on a JEOL, and 20 µm on the Hitachi HT7700. With JEOL stages, a value of 20-25 µm may be needed to reduce variability in the overlaps. To determine the needed value, take a small montage (e.g., 5 x 5) and process it in IMOD as described below. After running blendmont, open the file in midas and compare the overlaps between the first and second piece in each column with the overlaps between pieces higher in Y. Much lower overlap between the first and second piece in each column indicate more backlash correction is needed, but a point will be reached where additional correction does not improve the overlaps.

Regions of a plastic section that have not previously been exposed will undergo rapid changes under the beam that may impair the quality of the first image taken there. Thus, areas to be imaged should be pre-exposed. Fortunately the required dose is much less than for tomography, around 5-10 electrons per square Angstrom. This can be achieved by irradiating the whole area for a few minutes with a bright beam at low magnification.

Processing a Montage in IMOD

If you have acquired a tilt series, it is ready to process with eTomo in IMOD, and the Tomography Guide describes the steps that are special for working with a montage. Otherwise, if you want to work with the montaged images, you should process it with the interface for aligning serial sections and blending montages in eTomo in order to get a single blended image out of each montage. This is preferable to saving the overview image, because the blending program in IMOD, Blendmont, does a better job than SerialEM of shifting the pieces into optimal positions and it can produce seamless transitions between pieces. This interface is available in IMOD version 4.7 or higher.

If you just want to blend some isolated images (i.e., not serial sections to align), the procedure is fairly simple.  You may be able to get through it without reading more detailed instructions by following these steps:

  1. Place the file to be blended into a directory for the computations.  If you want to reuse a directory where you have blended before and you do not care to revisit the earlier blend, remove any existing '.ess' file from the directory.
  2. Open a terminal and cd to that directory.
  3. Start etomo
  4. Select Align Serial Sections / Blend Montages.
  5. Press the file chooser icon on the Stack line and select the image file.
  6. Switch to Montage and press OK.
  7. Select Treat as very sloppy montage if the montage was taken with stage movement instead of image shift
  8. Select Robust fitting if it is a low magnification map with grid bars or extensive regions off the edge of the grid.
  9. Press Make Blended Stack.  This will create a file named with the root name of the original image file plus '_preblend.mrc'.
  10. When it is done, press Open Blended Stack.

 For more complicated situations, or if that fails, here are some general instructions for processing the file in IMOD, which refer you to the more extensive help available there.

The montage is stored as individual frames in the MRC file. The MRC file has the coordinates of the frames in its extended header, unless the coordinates are too large to be stored there, in which case they are stored in a 'metadata autodoc' file whose name is the name of the image file with the extension '.mdoc'. If there is a '.mdoc' file associated with your image file, be sure to keep it together with the image file for processing.

If your file is named 'filename.st' and the coordinates are stored only in the '.mdoc' file, first extract the list of piece coordinates into a file 'filename.pl' with:

extractpieces -mdoc filename.st filename.pl

If you run 'header' on the image file in IMOD, the image size will show up as the size of the pieces and the number of pieces will be given for the number of sections. There is a program, 'montagesize', that can be run on a file to find out the actual size of the montage and the number of sections. You can run this as

montagesize filename.st

or, if the coordinates had to be extracted from an 'mdoc' file,

montagesize filename.st filename.pl



and select the topic 'Serial Section Alignment and Montage Blending'.  Read the sections 'Getting Started' and 'Blending Montaged Images' and follow the directions there.

The blending operation can be partially parallelized by splitting it into chunks. Run 'imodhelp splitblend' for details.  This cannot be done through the eTomo interface, but you can start the final blend, immediately kill it, then run splitblend on the file 'blend.com'.