Optimizing Image-Data Size

Optimizing image size offers significant benefits

The SLICE system enables rapid, high-resolution, three-dimensional imaging of intact specimens, including cleared organs and whole rodent-brain preparations. With the SLICE system, you can conduct whole-brain imaging or acquire images from large organ surveys in a matter of hours. This capability presents a significant computing and data management challenge; a single whole-brain acquisition from a SLICE system may be hundreds of gigabytes or more in size, and a typical experiment can generate terabytes of data in a single day, depending on the number of channels, bit depth, and spatial resolution employed. The result is that storage requirements can exceed the practical limits of institutional infrastructure, such that analysis pipelines spend more time reading and writing data than performing actual computation.

Fortunately, the information density of biological images is rarely uniform across spatial scales. Many of the structures of greatest scientific interest, such as cell bodies, projection pathways, and anatomical regions, are large relative to the imaging voxel size. Reducing the spatial resolution of stored data to match the scale of the structures being analyzed can dramatically reduce file size with minimal or no impact on analytical outcomes.

Data-size reduction strategies, including spatial downsampling, bit-depth adjustment, and JPEG 2000 compression, can be used to dramatically reduce dataset footprints without compromising scientific validity. Data size reduction is not a compromise; when properly validated, reduced-resolution datasets can support the same statistical conclusions as full-resolution data, while enabling faster analysis, reduced infrastructure costs, and more efficient collaboration. The guidance presented here is applicable to all MBF Bioscience analysis software and is particularly relevant for managing experiments that include large numbers of samples. The table below summarizes data-size optimization strategies.

Reduced data storage requirements

The strategies described here can be used to optimize the size of image data obtained using the SLICE system to create dramatically smaller files that retain the image data you need for your research. There are compelling advantages to optimizing your image-data size, including the following.

• More efficient use of expensive storage resources.

• Ability to archive more image data in the same physical storage space.

• Reduced costs for long-term data retention.

• When working with large cohorts, the storage savings from data-size optimization become increasingly important. Consider this practical example:

  For archiving images of 10 brains from an experiment, storing them at 1/4 resolution instead of full resolution could reduce your storage requirements by a factor of 16 (4x in each of three dimensions). This dramatic reduction makes long-term data retention far more practical.

Faster data transfer

Smaller file sizes translate directly to faster transfer times when moving data between storage systems and analysis computers.

Strategy	Workflow Stage	Downsampling?	Reversible?	Recommended Use	BrightSLICE software
Pixel binning	Image capture	Yes (XY)	No	Pixel binning combines adjacent camera pixels into a single pixel in the image data, effectively reducing spatial resolution in the XY plane. 10x detection objective: 2×2 pixel binning is recommended as default; 3×3 for large structures 5x detection objective: 3x3 pixel binning is sufficient for most uses
XY Save resolution	Post image capture	Yes (XY)	No*	After acquiring images, you can downsample the data by choosing a reduced "XY Save resolution" in the final step of image stitching. Choosing 1/2 resolution is effectively the same as acquiring images with 2x2 pixel binning, but without the advantages of faster imaging and smaller raw data size. *Not reversible, however, you can retain the original (raw) image tiles by choosing a different save location or filename.
Image resize	Post image capture	Yes (XY)	No*	Downsample existing image or subvolume files. This is equivalent to downsampling during image stitching using the XY save resolution dropdown menu, but it offers additional customization options. *Not reversible, however, you can retain the original (raw) images by choosing a different save location or filename.
Bit-depth adjustment	Image capture	No	No	Reducing bit depth to 8-bit may be useful for morphological analyses in which fine gradations in fluorescence intensity are not as critical as they are for quantitative analyses.
Z-step size	Image capture	No (axial sampling)	No	When structure size allows, capturing image stacks with a larger z-step size (distance between image planes) can reduce the overall size of the dataset. The optimal z-step size should be based on the detection-objective depth of field and dimensions of the structure(s) of interest.	or
Compression Ratio	Save/Export	No	Yes	15–20:1 compression is recommended for SLICE data. This ratio offers ~93% file-size reduction compared to no compression. Validate for quantitative intensity measurements.
JPEG 2000 file format	Save/Export	No	No*	Substantially smaller than TIFF with no loss of spatial resolution. Available in all MBF Bioscience software and our free MicroFile+ tool. MBF JPEG 2000 is the recommended file format for use with MBF Bioscience software JPEG 2000 format is the recommended file format for use with software from other sources. *Not reversible, however, you can retain the original (raw) images by choosing a different save location or filename.

Controls in BrightSLICE for optimizing data size

BrightSLICE software provides multiple points in the imaging and post-processing workflow at which data size can be reduced. These controls, described below, span hardware-level camera configuration during acquisition, z-step planning, Image stitcher resolution settings, and file format and compression choices.

Pixel binning for downsampling

Pixel binning during image acquisition is the most upstream data-reduction option in BrightSLICE software. It occurs directly at the sensor level before image data is written to disk. Binning combines adjacent camera pixels into a single pixel in the image data; it accelerates image-acquisition speed by reducing the volume of data that must be offloaded from the camera between frames and dramatically reduces image file size.

In most cases, we recommend 2×2 binning as a starting point. For imaging larger structures, 3×3 binning may provide sufficient resolution, additional improvements in image-acquisition speed, and a 9-fold decrease in image-data size.

• Given that the lateral resolution of SLICE is 0.8–1.0 µm with the 10x detection objective, 2×2 binning results in an effective pixel size of 520 nm that preserves most structural details while improving imaging speed and reducing data size.

• When equipped with the 5x detection objective, the SLICE system provides 1.5-2 µm lateral resolution (depending on sample clearing); 2x2 binning yields an effective pixel size of 1 µm, again roughly satisfying the Nyquist-Shannon sampling criteria.

• Using 3×3 binning (effective pixel size of 780 nm) can further increase speed and decrease file size, but may compromise your ability to capture the smallest structural features.

Binning Mode	Relative # Pixels	Data Size Reduction
1x1	100%	—
2x2	25%	4x smaller
3x3	11.1%	9x smaller

To set pixel binning, go to the Camera Settings panel, click More Settings and choose the Binning and Cropping tab. Select the desired pixel-binning setting from the Binning dropdown menu.

Downsampling using the XY save resolution setting during image stitching

The BrightSLICEImage Stitcher and other post-processing tools benefit significantly from working with downsampled data. Image stitching is a computationally intensive operation, and processing at a reduced resolution can substantially decrease both compute time and peak memory usage, particularly for large, whole-organ, image data sets.

To downsample using the XY save resolution setting:

1. Control downsampling in the Image Stitcher, by choosing to Preview images before stitching in the Preview options window.

2. In the Tissue Scan Preview dialog box opens, choose the amount of downsampling from the X, Y save resolution dropdown menu.

   In practice, 1/2 and 1/4 resolution are the most commonly used levels for analysis workflows targeting cell-body-scale structures. For visualization-only purposes or atlas registration, 1/8 or 1/16 resolution may be sufficient.

This constitutes true spatial downsampling of acquired image data. Reducing the XY save resolution combines signal from adjacent pixels to produce a lower-resolution stitched output image. The result is a smaller file that is well-suited for analysis workflows where the target structures are sufficiently large to be detected at reduced resolution.

As with all spatial downsampling, reducing XY resolution involves a trade-off between file size and image detail. The appropriate setting depends on the experimental objectives and should be validated against full-resolution data. Once a resolution level is validated for a given experimental paradigm, it can be applied consistently to all subsequent samples in the cohort.

Image Resize

The Image Resize tool in BrightSLICE performs true downsampling on images that have already been acquired; it is the same spatial reduction operation as choosing a reduced XY save resolution setting during image stitching. Previously saved images or subvolumes extracted from larger images can be downsampled using this tool.

Find Image Resize on the image ribbon.

For downsampling portions of images (subvolumes), we recommend that you extract and save subvolumes at full resolution and then use Image Resize to save the image files at the desired downsampled resolution for analysis. Find the Subvolume tool in the 3D Visualization window.

Bit-depth reduction during imaging

SLICE image acquisition produces 16-bit images by default, which can represent 65,536 discrete fluorescence-intensity values. For many biological samples and downstream analyses, this dynamic range exceeds what is scientifically necessary. Converting 16-bit images to 8-bit, with 256 fluorescence-intensity levels reduces file size by 50% with no change in spatial resolution. The trade-off is that subtle differences in fluorescence intensity will be eliminated. This may be acceptable for visualization and morphological analysis but could affect quantitative fluorescence measurements. It is important to validate bit-depth reduction in the context of the specific quantitative requirements of each experiment.

To change the bit depth adjust the Camera Settings before beginning image acquisition.

Click More Settings and go to the Binning and Cropping tab, then adjust the Bit Depth using the dropdown menu.

Z-step size (Distance between image planes during imaging)

The Z-step size, which is the axial distance between image planes, is not a downsampling operation in the strict sense, but increasing it is a practical and often underutilized method for reducing total image-acquisition time and image-data size. Because the total number of image planes is determined by the tissue depth divided by the Z-step, even modest increases in step size can meaningfully reduce file size and imaging time for thick specimens.

The appropriate Z step is affected by the optical properties of the microscope objective and the size of the structures you are imaging. The axial depth of field of the objective sets the lower bound on the meaningful Z resolution: sampling more finely than the depth of field does not contribute additional information.

For large structures, a Z-step that is two to five times the diffraction-limited axial resolution of the microscope objective is often sufficient. The guiding principle is to image the structure with adequate axial sampling to support accurate detection of structures of interest and morphological analysis, without redundant oversampling that inflates file size without scientific benefit.

To set the Z-step size:

• In the Tissue scan workflow, go to step 3. Z settings

• When acquiring an image stack from the current field of view, enter a value in µm for the Distance between images; alternatively click Options for more Z-step controls.

Compression Ratio

File compression settings have a direct and significant impact on image file size. We carefully examined the effects of SLICEdata compression and identified an optimal compression range of 15–20:1, which reduces file size approximately 93% while preserving scientific validity (see our white paper, A Practical Guide to Selecting Compression Levels for 3D Light Sheet Fluorescent Microscopy Data (PDF)). We recommend routinely applying 15–20:1 compression for both working files and long-term archiving. More aggressive compression ratios are possible but should be validated for each data type and analysis workflow by comparing to uncompressed or lightly compressed versions of the same data to confirm that fluorescence-intensity values are not meaningfully affected.

Set Compression Ratio and File Format in any image file saving dialog, for example by going to File > Save As or at the end of the Tissue Scan workflow.

File Format: JPX vs. TIFF

The file format used to save image data can also have a profound impact on data size. BrightSLICE software supports saving images in both JPX (JPEG 2000 Extended) and TIFF formats. For most workflows, JPX is strongly preferred over TIFF. JPX files are substantially smaller than their TIFF equivalents for comparable image content, due to the inherent efficiency of the JPEG 2000 wavelet-based compression architecture. For large volumetric datasets, the difference in file size between formats can be dramatic. JPX compression is also available in all other MBF Bioscience commercial software, as well as in our free MicroFile+ image-converter tool.

Unless a downstream analysis tool requires TIFF input specifically, JPX should be used as the default output format.

Set File Format and Compression Ratio in any image file saving dialog, for example by going to File > Save As or at the end of the Tissue Scan workflow.