Image Stitching Made Easy:

Transmission Electron Microscopy (TEM) has become increasingly vital in histology and its related fields, such as cytology and pathology. Traditionally, TEM was a very complex and expensive and significantly more challenging for low-atomic-weight materials such as pathology thin sections. However, the emergence of Low-Voltage Electron Microscopy (LVEM) has shifted this paradigm. By significantly reducing technical difficulties and costs, LVEM has established itself as a reliable and promising technique, especially for TEM of samples composed of low-atomic-weight elements. Its ability to provide high-contrast imaging without the need for heavy-metal staining, combined with its ease of use, has greatly expanded the application of LVEM in histological applications in both research and clinical diagnosis.

A particular challenge when studying biological samples in TEM, is imaging nanoscale features which span micrometers. One method is to capture the region at a low enough magnification to fit the entire region of interest in the image. This would however compromise the resolution of finer details. Image Stitching, the process of fusing multiple high-magnification image tiles from a larger sample area, is a preferred technique as it provides for higher detail over a wide area. This method provides a holistic view of the sample without sacrificing resolution.

This article provides a straightforward guide to image stitching using Fiji, a free and open-source software [1]. It aims to empower researchers and operators with the knowledge to harness the full potential of LVEM through effective image stitching techniques.

Transmission Electron Microscopy (TEM) plays a crucial role in histological studies, offering in-depth insights into the anatomy and mechanics of both intra- and inter-cellular structures. High-resolution TEM images are instrumental in enhancing our understanding of these complex biological systems. However, a common challenge in TEM is the limited field of view at high magnifications, which can restrict the comprehensiveness of the observed sample.

Image stitching emerges as a powerful solution to this issue. By combining multiple high-resolution images, it enables the creation of a seamless mosaic, effectively capturing a larger area of the sample without compromising detail. This technique broadens our comprehension by providing a more holistic view of the structural arrangement and interactions within the sample.

This guide is specifically designed for pathologists and researchers, although its application is not confined to these groups. It offers a straightforward, step-by-step approach, beginning with the image acquisition process and culminating in the creation of a smoothly fused, detailed image. Through this guide, users will be equipped with the knowledge and techniques necessary for effective image stitching, thus enhancing their capacity to extract valuable information from EM images.

The journey to producing a high-quality stitched image begins with acquiring excellent individual tile images.

Imaging the First Tile

1 - Select your Area of Interest

The first step in preparing for image stitching is to select an Area of Interest (AOI) for your study. Start by finding a large area (e.g. at 5,000x) that you plan to divide into the desired number of constituent tiles. Ensure saving an image of this entire AOI; this will be useful later for comparing with the final stitched image.

2 - Start with One Corner

Once your AOI is selected, begin the tile image capturing process by magnifying a corner of the AOI (e.g. at 20,000x). Obtain a well-focused image, and enhance its brightness and contrast to achieve optimal levels. Remember, TEM is an art, and mastering it requires patience, attention to detail, as well as some creativity, particularly at this initial step. Achieving and maintaining high quality and focus throughout the imaging process is essential for creating a smooth and seamless final stitched image.

3 - Save the First Tile

Once you are satisfied with the focus, brightness, and contrast of your first tile, save the image. Ensure using a consistent logical naming format throughout the imaging process (e.g. Image_001, Image_002, etc.). This practice is crucial for simplifying the subsequent stitching endeavor.

Info Bar: To Keep or Not to Keep?

• While the info bar is necessary to provide the image scale bar in later steps, its presence can complicate the stitching process. Therefore, a practical approach is to save the first image twice: first with the info bar and a second time without. The magnification level will remain constant throughout the imaging process. For all subsequent tiles, it is advised to continue saving with no info bar. After successfully creating the stitched image, we will reintegrate the scale bar onto the final composite, as detailed in section ‘Adding the Scale’.

Moving to the Next Tiles

Once you have saved the image of the first corner of your AOI, maneuver to the subsequent tile exclusively in either the X or Y direction. Since maneuvering only in a X/Y direction is very challenging using the joystick, it is recommended to use the keyboard and enter fixed increments into the LVEM25E software’s movement panel.

• The LVEM25E brings a significant advantage to histology applications, thanks to its integration of piezo motors. This technology allows for controlled and precise movement of the sample stage in consistent, equal steps. With LVEM25E, users can effortlessly achieve movement in a single direction over a specified distance—a level of control that manual navigation with a joystick simply cannot match. This precision ensures uniformity across all captured images, which is critical for producing high-quality, seamless stitched images.

Maintain a 15-20% overlap with the previous tile to ensure seamless stitching later.

For instance, suppose you aim to create a 3x3 mosaic image, which involves capturing nine images, starting from the top left corner.

Here’s how you would proceed in a snake pattern (Figure 1):

1. Save Tile 1.

2. Move to the right (X direction) ensuring some overlap with Tile 1, then save Tile 2.

3. Continue to the right (X direction) with overlap with Tile 2, then save Tile 3.

4. Shift downwards (Y direction) overlapping Tile 3, and save Tile 4.

5. Move to the left (X direction) overlapping Tile 4, then save Tile 5.

6. Continue this pattern until you reach and save Tile 9, always ensuring the necessary overlap with the previous tile.

Congratulations! You have now completed the image acquisition phase on the microscope. The next step is to proceed to the stitching part of the process on your computer.

• NOTE: It is recommended to transfer your images to a different computer for post-processing. It is advised to avoid installing any third-party software (such as Fiji) on the microscope’s PC.

We are using Fiji—a free, open-source, and user-friendly interface based on ImageJ [1, 2], for the purpose of this tutorial. You can download the software through this link: Fiji Downloads.

Image Stitching Procedure

Folder Setup

It is recommended to create a dedicated folder containing only the images that are going to be stitched. This organization ensures that the stitching software can easily access and import the correct image tiles without importing anything else, such as the large AOI image or the first tile containing the info bar.

To do this, simply create a new folder and transfer all the images intended for stitching into it.

Open the Software

Open Fiji. Under the ‘Plugins’ tab, scroll down and click on the 'Stitching' option where you will see three choices: Pairwise stitching, Grid/Collection stitching, and deprecated. For the purpose of this guide, we will only showcase the first two options.

Pairwise Stitching

As its name suggests, this mode is designed to stitch two adjacent tiles together at a time. Therefore, only use this mode when you are stitching just a few tiles. Otherwise, using the ‘Grid/Collection stitching mode’ is recommended.

1. Open all the images to be stitched in Fiji. You can do this by going to 'File' → 'Open' and selecting your images, or simply drag and drop them onto the Fiji main panel.

• Note: If there are no images open on Fiji, you will encounter an error upon selecting Pairwise stitching.

2. Under the 'Plugins' tab, select 'Stitching' and choose 'Pairwise stitching’.

A prompt window will ask you to select the first and second images. After doing so, click 'OK'.

3. A new window will pop open for selecting the parameters. The default settings usually work well for histology applications.

• Feel free to choose a proper naming for the image using the ‘Fused image name’.

• Click on ‘OK’ once happy with the parameters.

4. The software will perform some computation and display a fused image.

5. If you are satisfied with the alignment results, proceed to Step 6 to crop and save the fused image. Otherwise, if there are more tiles to be added, you can open them on Fiji and repeat Step 2. This process should be done one by one, i.e., stitching the last fused image with the next tile. Make sure you choose the newly stitched image as the First image (reference), and the next tile as the Second image:

6. Crop and Save

To crop the final fused image, select the rectangle tool from the Main Menu. Draw a rectangle around the desired area by clicking and dragging from corner to corner. Once the rectangle is drawn, right-click on the image and choose 'Duplicate.'

Save the final fused image through ‘File’ à ‘Save As’.

NOTE: To add a scale bar, review ‘Adding the scale bar’ section.

The Grid/Collection mode is a handy tool to stitch together multiple tiles of images to create a larger composite image.

Naming of the Individual Tiles

Unlike Pairwise stitching, the Grid/Collection mode eliminates the need to manually open individual tiles in the software. In this mode, the software locates and identifies the relevant images within a given folder based on their file names. Therefore, properly naming each individual tiles is crucial (unless using the ‘Unknown Positions’ option—explained later), as the software selects and arranges each individual tile by its name, following a designated layout (e.g., row-by-row, column-by-column, and snake pattern) and direction (left-to-right, bottom-to-top, etc.). If the images were not named in a consistent and logical order during microscopy, it is advised to rename them now before proceeding with the stitching process. To do so, simply copy them into a new folder and rename them accordingly. Having a proper and consistent naming system provides the software with clear indications about the positions of the tiles, thereby reducing computational effort.

• Note: The Grid/Collection stitching mode assumes that the tiles, set to be the building blocks of the stitched image, are sequentially numbered. Below are examples of several accepted naming formats for reference:

tile_1.tif, tile_002.tif, …, tile_049.tif

tile_001.png, tile_002.png, …, tile_049.png

tile_x01_y01.lsm, tile_x02_y01.lsm, …, tile_x10_y10.lsm

Intro to the Stitching Process

To access the Grid/Collection stitching plugin, go to 'Plugins' > 'Stitching' > 'Grid/Collection stitching.'

A new window opens up, allowing you to choose the stitching pattern and direction:

Open the drop-down menu in front of “Type” to see different options:

Below is a brief explanation of what each type does:

• Row-by-row, completes the stitching of one horizontal set (row) of images before going to the next row,

• Column-by-column, stitches a vertical set (column) of images before going to the next column,

• Snake by rows will finish the first row but doesn’t start the second row in the same direction, rather, moves to the next row on the top (or bottom) and moves in the opposite horizontal direction (imagine you finished reading a line from left to right, and will read the next line from right-to-left)

• Snake by columns, does similarly, but in the vertical direction,

• Filename defined position, will use the approximate positions as encoded in the tiles’ file names.

• Unknown positions, will try to determine the positions of tiles regardless of their naming system. Use this mode if you are clueless about where each tile is supposed to go. This mode will require more computational effort as the software needs to figure out the positions through an image processing effort.

• Positions from file, will use another file (metadata) to find the approximate position of the tiles.

In this training, we will only demonstrate two modes: 'Grid: Snake by rows' and 'Unknown position.' The other Grid options are very similar to ‘Snake by rows.’

Click on the ‘Order’ drop-down menu to select the direction onto which you wish the tiles to be stitched together. The accompanying 3x3 picture visually demonstrates this arrangement.

If you are uncertain about the approximate positions of the tiles, employ this mode. The software will attempt to calculate the stitching areas based on the contrast of the features along tile borders.

After choosing “OK”, a new window opens to enable you to make further adjustments and define the directory:

• Click on ‘Browse’ to select the folder that contains the image tiles that are going to be stitched together.

• Check the ‘Confirm files’ box. This will give you the option to select which tiles within the folder to be stitched.

• The ‘Computation parameters’ will give you the option to choose between using more RAM and be faster, or less RAM and be slower. If there are not other programs running in the background and you have a decent computer, go ahead and select the ‘Save computation time (but use more RAM) ‘option.

• You can leave the rest of the settings as is, and click ‘OK.’

• A new prompt asks you to check all the files intended for stitching. Uncheck them if they are not intended for stitching.

• After the computation takes place, the software will display the stitched image along with a log window. If you are satisfied with the results, save the image (‘File’ > ‘Save As’).

Use this mode (or any other mode except ‘Unknown positions’ if you know the approximate location of each tile.) Knowing the order in which the images were created during microscopy will determine the direction in which we want to stitch the tiles.

For the purpose of this tutorial, we imaged 25 tiles of a renal section with a 5x5 final mosaic in mind using an LVEM25. While imaging, we started from the top left corner and moved right. After saving the fifth tile, we moved down and then left. Therefore, for the stitching part, we select the ‘Right & Down’ order in the Grid/Collection stitching menu and click ‘OK’:

A new window opens up showing various parameters:

• Grid size: In this example, we input ‘5’ for both the X and Y grid size.

• Tile overlap percentage: The default overlap percentage is 20%. However, it really depends on how much overlap you had in mind during the imaging. The goal is to set the minimum overlap possible, as a higher overlap will result in a longer computation time.

Tile Overlap Percentage = (Overlap Width / Tile Width) x 100%

Example: Assuming our target tile width is 5 microns, but we capture images with tiles that are 8 microns wide (Figure 1). In such a scenario, the resulting overlap measures 3 microns (8 - 5 = 3), leading to the following overlap percentage calculation:

Overlap Percentage = (3 µm/ 8 µm) * 100% = 38%

• File names for tiles: Ensure that the file name and file type match the tile names in the directory folder. For example, if the images in the folder are labeled from ‘Image_0001.tiff’ to ‘Image_0030.tiff,’ use the appropriate file name format in the parameters window, which should be ‘Image_00{ii}.tiff.’ Note that we’ve retained two zeros before, and two ‘i’s inside the curly braces ‘00{ii}’, as there are 30 tiles (30 being a two-digit number). By inserting two 'i's inside the braces and leaving everything else before the braces, we keep the naming consistent between the original file names and the counter.

• Computation parameters: As described previously in the Unknown positions section, this option allows you to choose between a quicker rendering process that requires more RAM and a slower process that conserves RAM.

• Leave the rest of the parameters unchanged.

• Click ‘OK’ to let the software start stitching the tiles. Depending on the size and number of the tiles, this process might take tens of seconds.

Crop: After a fused image is produced and you are satisfied with the results, proceed to crop the image to remove any empty areas and achieve a complete image. To crop the image, select the rectangle tool on the main menu and draw a box around you area of interest. Then, right click anywhere on this area and select ‘Duplicate.’

Setting the correct scale for the fused image:

1. To correct the scale of the fused image, follow these steps:

a. Open one of the tiles containing an info bar.

b. Draw a horizontal line along the scale bar (press Shift while dragging the mouse).

c. Navigate to 'Analyze' -> 'Set Scale.'

d. Enter the known distance and unit and then click 'OK.'

To insert a scale bar into the fused image, navigate to 'Analyze' -> 'Tools' -> 'Scale bar.'

A new window will appear, allowing you to configure the appearance and placement of the scale bar (e.g., measurements, color, orientation).

Once you are satisfied with the settings, click 'OK' to add the scale bar to the image. Save the final result through 'File’ > ‘Save As’ > and select your preferred format. For optimal detail preservation, it is recommended to choose either TIFF or PNG formats.

So far, we have explored how to stitch numerous image tiles together to achieve a high-resolution image of a substantially wider area. However, seeing is believing, and visual examples often convey the effectiveness of these techniques more powerfully than words alone!

Below, we present an example showcasing a large pathological thin section (FOV ~20 µm). Initially, this section is imaged using very low magnification to encompass the entire area (Figure 3.a). It is followed by a fused image of the same area, composed of 25 high-resolution, high-magnification image tiles (Figure 3.b). Compare for yourself and witness the remarkable difference!

This guide, combining Fiji's image stitching with Delong Instruments Low-Voltage Electron Microscopes, offers a versatile tool for pathologists, histologists, and beyond. It simplifies the process of creating detailed composites from individual images, enhancing the analysis of biological specimens. To add more, the precision of LVEM25E's piezo motors enable capturing a series of images with consistent X/Y coordinates, thereby making the image stitching process more efficient and accurate. This guide is not just about mastering a technique; it's about expanding the horizons of scientific discovery, particularly in life sciences and other fields that require meticulous analysis of microstructures.

[1] Schindelin, J., Arganda-Carreras, I., Frise, E., Kaynig, V., Longair, M., Pietzsch, T., … Cardona, A. (2012). Fiji: an open-source platform for biological-image analysis. Nature Methods, 9(7), 676–682. doi:10.1038/nmeth.2019

[2] Ferreira T and Rasband WS. “ImageJ User Guide — IJ 1.46”, imagej.nih.gov/ij/docs/guide/, 2010—2012.

Acknowledgement

The author thanks and acknowledges the creators of Fiji, an open-source software. Special recognition is given to the developers of Fiji's 'Stitching' plugin, which was instrumental in the creation of this Image Stitching guide.

Emad Shahnam, MASc, is an Applications Engineer at Delong America. His experience, notably at the University of Waterloo's Center for Advanced Materials Joining (CAMJ), involved the development of nanocomposite filaments for water purification, the characterization of organic materials for electroclassification in food science, and the preparation and analysis of Pt/C catalysts for hydrogen fuel cells. At Delong America, Emad develops applications where Low Voltage Electron Microscopy can provide researchers with key insights into their samples.