Biomedical Imaging

Learn about how TileDB handles biomedical imaging.
Under construction

This section is under construction and will appear in Academy very soon. Please stay tuned and contact us if you have any questions we can answer in the meantime.

TileDB provides support for importing, visualizing, analyzing, and exporting multi-resolution whole-slide microscopy images. TileDB’s BioImaging features include:

Ingestion

To get started with bioimages on TileDB, start up a TileDB notebook with the Genomics environment, and import the TileDB BioImage package:

import tiledb.bioimg
from tiledb.bioimg.converters.ome_tiff import OMETiffConverter
from tiledb.bioimg.converters.openslide import OpenSlideConverter

Each converter provides a to_tiledb function taking a source path and storage destination for the converted TileDB multi-resolution image group. For example:

OMETiffConverter.to_tiledb(src, dest, level_min=0)
OpenSlideConverter(src, dest, level_min=0)

Batched ingestion

TileDB includes functionality for large-scale ingestion from images stored in an Amazon S3 bucket or other supported object store.

import tiledb.bioimg

# `images` will be converted in bulk using
# `dest` as the base output path.
tiledb.cloud.bioimg.ingest(source=images, output=base_dest)

Chunked ingestion

Check out the demo notebook on TileDB Cloud.

Read image data

You may read image data with the TileDB Python Array API directly, or by using the OpenSlide Python-compatible API.

OpenSlide Python API

from tiledb.bioimg.openslide import TileDBOpenSlide

# Construct OpenSlide API wrapper
img = TileDBOpenSlide(tiledb_image_path)

# Read image data at `level`
#   from starting `(x0,y0)` location to
#   `(x + size_x, y + size_y)`
# Returns NumPy array
data = img.read_level((x0, y0), level, (size_x, size_y))

Visit the OpenSlide Python API documentation for all available commands.

Visualization

You can preview bioimaging assets within the built-in viewer in the TileDB user interface.

Storage format

TileDB-BioImg stores multi-resolution images as a group of arrays, with one dense array per resolution level. For example, given an input TIFF file with three resolution layer pages:

2132x2488
4266x4978
17068x19918

The following directory structure will be created:

Image.tdb      # TileDB group
  |- l_0.tdb   # TileDB array, 17068x19918
  |- l_1.tdb   # TileDB array, 4266x4978
  |- l_2.tdb   # TileDB array, 2132x2488

Image-level metadata is stored as TileDB group metadata, and individual level metadata is stored within the resolution-level array.

Utilities

Export to OME Formats

The TileDB BioImg library provides a utility function to export a TileDB BioImage to several OME formats.

# Export TileDB BioImg to an OME-TIFF file
OMETiffConverter.from_tiledb(tiledb_img_path, output_path)

# Export TileDB BioImg to an OME-Zarr directory
OMEZarrConverter.from_tiledb(tiledb_img_path, output_path)

Integrations

TileDB maintains integrations with several related tools and libraries.

Machine learning

The TileDB ML library includes support for loading images into PyTorch and TensorFlow.

Napari

TileDB maintains an integration with Napari via the napari-tiledb-bioimg plugin. This plugin supports visualization and analysis of TileDB BioImage datasets using Napari tools, with datasets loaded from TileDB and any supported storage system. (see the [demo video] on the repository README).

Install via the Napari plugin manager by searching for “TileDB” in the Napari plugin manager, or install with pip inside of a Napari Python environment:

pip install napari-tiledb-bioimg

Example of TileDB BioImage rendered in the Napari viewer from the Chan Zuckerberg Initiative. Example of TileDB BioImage rendered in the Napari viewer from the Chan Zuckerberg Initiative.

Something else?

Please get in touch if you are interested working with TileDB BioImages in other visualization software, or specific analysis tools.