|Author||Stephan Preibisch & David Hörl|
|Maintainer||Stephan Preibisch & David Hörl|
Introduction & Overview
The BigStitcher is a software package that allows simple and efficient alignment of multi-tile and multi-angle image datasets, for example acquired by lightsheet, widefield or confocal microscopes. The software supports images of almost arbitrary size ranging from very small images up to volumes in the range of many terabytes, which are for example produced when acquiring cleared tissue samples with lightsheet microscopy.
As its predecessor, the Image Stitching, the BigStitcher can run in fully automatically mode but is additionally able to guide the user through the alignment process by interactively showing intermediate results with the option for user interaction, which includes interactive setup of approximate tile locations. The BigStitcher is fully integrated with the BigDataViewer and is thus able to interactively display and process the input images independent of their size. An improved global optimization is now able to efficiently align sparsely connected datasets, where image content is separated by large areas of almost constant background.
Specifically intended for lightsheet acquisitions of cleared samples (e.g. by CLARITY), the BigStitcher offers several features:
- Registration of multi-view multi-tile acquisitions, where each view of the sample that is acquired from different orientations may consist of many individual image tiles.
- Integrated downstream processing of the data such fusion and deconvolution of the aligned data at different resolution levels (full or reduced resolutions) and for selected areas (full image or selected parts)
- Since cleared samples are typically acquired by a lightsheet microscope that can illuminate from different directions, the BigStitcher supports a pre-selection of the best illumination direction at every image block in the sample
- Support of non-regular acquisition grids, which includes 'intelligent acquisitions' where some of the image blocks can be missing if only background is present
- Support of image data acquired at different resolutions to combine overview scans with high resolution acquisitions of specific areas of interest
A beta-Version of BigStitcher is available via the Fiji Updater. Just go to , click ▶
Manage update sites and select
BigStitcher in the list. After applying the changes and restarting Fiji, BigStitcher will be available under .
The source code is available on GitHub, please also report feature requests & bugs there.
The workflow of the BigStitcher consists of multiple steps covering import, visualisation, processing, and output that are explained in full detail under the respective pages linked below:
Opening / Importing an Image Dataset
Since image data is produced in a variety of formats, the first step is to assemble all images into a dataset and load the necessary metadata or manually define it if it is missing. The BigStitcher is integrated with the BigDataViewer and thus shares the same data representation, which is stored as a XML file on disk.
Once a dataset was defined and the XML file was saved, you can quickly load any dataset:
The BigStitcher can run in two modes depending on whether you want to align tiled images imaged from the same direction (Stitching mode) or align tiled acquisitions from different angles or time points (MultiView mode). The following sections give an overview of the UI in both modes:
Ideally, the location of tiles should have been loaded from metadata while importing a dataset. With the BigStitcher you can, however, also move views to a regular grid manually, load view locations from a tile configuration file or manually translate single views.
Many lightsheet microscopes offer illumination of the sample from multiple directions. Since typically one illumination direction produces the best image for a given point in the sample, you might want to discard images illuminated from other directions.
Stitching consists of aligning multiple overlapping image tiles using a translation model. If your dataset consists of tiled acquisitions from multiple angles, the stitching should be done for all angles independently before aligning the angles using Multiview Reconstruction:
The basic stitching pipeline consists of 3 steps:
Additionally, we offer the possibility to refine the alignment with a more complex affine transformation model using the iterative Closest Point (ICP) algorithm:
In addition to the standard stitching pipeline, we also offer advanced functionality, such as different algorithms for determining pairwise shifts:
Once you have aligned the views imaged from the same angles to each other, you can then proceed to align those groups of views to each other, using Multiview-Reconstruction.
This process consists of finding interest points such as nuclei or fiducial beads in the single views, matching corresponding interest points in other views and then aligning the views based on the corresponding interest points.
In order to fuse the aligned data into a "classical image" and for example save it as TIFF, it is necessary to define the area that should be fused. This is done by defining a Bounding Box. By default, the software will always pre-compute two Bounding Boxes covering (a) the area covered by all views and (b) the area covered by the currently selected views. This means that only if you want to specify a certain (sub)area, for example because there are a lot of empty pixels around your imaged object you need to define a bounding box. To really minimize the fused area, might also want to first re-orient your sample first. More details on Bounding Boxes can be found here:
Once you decided to fuse everything or defined a specific bounding box, you can start the fusion process. If you want to check how the fused area will look like, you can do a quick preview by for example choosing . This will immediately pop up an ImageJ Virtual Stack (browsing is somewhat slow) containing a fusion of all selected views 4x downsampled (multiple channels or illuminations will be fused together). ▶ ▶