wsidicom

wsidicom is a Python package for reading DICOM WSI. The aims with the project are:

• Easy to use interface for reading and writing WSI DICOM images and annotations either from file or through DICOMWeb.
• Support the latest and upcoming DICOM standards.
• Platform independent installation via PyPI.

Installing wsidicom

wsidicom is available on PyPI:

pip install wsidicom

And through conda:

conda install -c conda-forge wsidicom

Important note

Please note that this is an early release and the API is not frozen yet. Function names and functionality is prone to change.

Requirements

wsidicom uses pydicom, numpy, Pillow, marshmallow, fsspec, universal-pathlib, and dicomweb-client. Imagecodecs, pylibjpeg-rle, pyjpegls, and pylibjpeg-openjpeg can be installed as optionals to support additional transfer syntaxes.

Limitations

• Levels are required to have (close to) 2 factor scale and same tile size.

• Only 8 bits per sample is supported for color images, and 8 and 16 bits for grayscale images.

• Without optional dependencies, the following transfer syntaxes are supported:

  • JPEGBaseline8Bit
  • JPEG2000
  • JPEG2000Lossless
  • HTJPEG2000
  • HTJPEG2000Lossless
  • HTJPEG2000RPCLLossless
  • ImplicitVRLittleEndian
  • ExplicitVRLittleEndian
  • ExplicitVRBigEndian

• With imagecodecs, the following transfer syntaxes are additionally supported:

  • JPEGExtended12Bit
  • JPEGLossless
  • JPEGLosslessSV1
  • JPEGLSLossless
  • JPEGLSNearLossless
  • RLELossless

• With pylibjpeg-rle RLELossless is additionally supported.

• With pyjpegls JPEGLSLossless and JPEGLSNearLossless is additionally supported.

• Optical path identifiers needs to be unique across instances.

• Only one pyramid (i.e. offset from slide corner) per frame of reference is supported.

Basic usage

Load a WSI dataset from files in folder.

from wsidicom import WsiDicom
slide = WsiDicom.open("path_to_folder")

The files argument accepts either a path to a folder with DICOM WSI-files or a sequence of paths to DICOM WSI-files.

Load a WSI dataset from remote url using fsspec.

from wsidicom import WsiDicom
slide = WsiDicom.open("s3://bucket/key", file_options={"s3": "anon": True})

Or load a WSI dataset from opened streams.

from wsidicom import WsiDicom

slide = WsiDicom.open_streams([file_stream_1, file_stream_2, ... ])

Or load a WSI dataset from a DICOMDIR.

from wsidicom import WsiDicom

slide = WsiDicom.open_dicomdir("path_to_dicom_dir")

Or load a WSI dataset from DICOMWeb.

from wsidicom import WsiDicom, WsiDicomWebClient
from requests.auth import HTTPBasicAuth

auth = HTTPBasicAuth('username', 'password')
client = WsiDicomWebClient.create_client(
    'dicom_web_hostname',
    '/qido',
    '/wado,
    auth
)
slide = WsiDicom.open_web(
    client,
    "study uid to open",
    "series uid to open" or ["series uid 1 to open", "series uid 2 to open"]
)

Alternatively, if you have already created an instance of dicomweb_client.DICOMwebClient, that may be used to create the WsiDicomWebClient like so:

dicomweb_client = DICOMwebClient("url")
client = WsiDicomWebClient(dicomweb_client)

Then proceed to call WsiDicom.open_web() with this as in the first example.

Use as a context manager.

from wsidicom import WsiDicom
with WsiDicom.open("path_to_folder") as slide:
    ...

Read a 200x200 px region starting from px 1000, 1000 at level 6.

region = slide.read_region((1000, 1000), 6, (200, 200))

Read a 2000x2000 px region starting from px 1000, 1000 at level 4 using 4 threads.

region = slide.read_region((1000, 1000), 6, (200, 200), threads=4)

Read 3x3 mm region starting at 0, 0 mm at level 6.

region_mm = slide.read_region_mm((0, 0), 6, (3, 3))

Read 3x3 mm region starting at 0, 0 mm with pixel spacing 0.01 mm/px.

region_mpp = slide.read_region_mpp((0, 0), 0.01, (3, 3))

Read a thumbnail of the whole slide with maximum dimensions 200x200 px.

thumbnail = slide.read_thumbnail((200, 200))

Read an overview image (if available).

overview = slide.read_overview()

Read a label image (if available).

label = slide.read_label()

Read (decoded) tile from position 1, 1 in level 6.

tile = slide.read_tile(6, (1, 1))

Read (encoded) tile from position 1, 1 in level 6.

tile_bytes = slide.read_encoded_tile(6, (1, 1))

Close files

slide.close()

API differences between WsiDicom and OpenSlide

The WsiDicom API is similar to OpenSlide, but with some important differences:

• In WsiDicom, the open-method (i.e. WsiDicom.open()) is used to open a folder with DICOM WSI files, while in OpenSlide a file is opened with the __init__-method (e.g. OpenSlide()).

• In WsiDicom the location parameter in read_region is relative to the specified level, while in OpenSlide it is relative to the base level.

• In WsiDicom the level parameter in read_region is the pyramid index, i.e. level 2 always the level with quarter the size of the base level. In OpenSlide it is the index in the list of available levels, and if pyramid levels are missing these will not correspond to pyramid indices.

Conversion between OpenSlide location and level parameters to WsiDicom can be performed:

with WsiDicom.open("path_to_folder") as wsi:
    level = wsi.levels[openslide_level_index]
    x = openslide_x // 2**(level.level)
    y = openslide_y // 2**(level.level)

Metadata

WsiDicom parses the DICOM metadata in the opened image into easy-to-use dataclasses, see wsidicom\metadata.

with WsiDicom.open("path_to_folder") as wsi:
    metadata = wsi.metadata

The obtained WsiMetadata has child dataclass properties the resembelse the DICOM WSI modules (compare with the VL Whole Slide Microscopy Image CIOD):

• study: The study the slide is part of (study identifiers, study date and time, etc.).
• series: The series the slide is part of.
• patient: Patient information (name, identifier, etc.).
• equipment: Scanner information information.
• optical_paths: List of optical path descriptions used for imaging the slide.
• slide: Slide information, including slide identifier, stainings done on the slide, and samples placed on the slide, see details in Slide information
• label: Slide label information, such as label text.
• image: Image information, including acquisition datetime, pixel spacing, focus method, etc.
• frame_of_reference_uid: The unique identifier for the frame of reference for the image.
• dimension_organization_uids: List of dimension organization uids.

Note that not all DICOM attributes are represented in the defined metadata model. Instead the full ´pydicom´ Datasets can be accessed per level, for example:

with WsiDicom.open("path_to_folder") as wsi:
    wsi.levels.base_level.datasets[0]

If you encounter that some important and/or useful attribute is missing from the model, please make an issue (see Contributing).

Slide information

The Slide information model models the Specimen module has the following properties:

• identifier: Identifier for the slide.
• stainings: List of stainings done on the slide. Note that the model assumes that the same stainings have been done on all the samples on the slide.
• samples: List of samples placed on the slide.

Note that that while the parsing of slide information is designed to be as flexible and permissive as possible, some datasets contains non-standard compliant Specimen modules that are (at least currently) not possible to parse. In such cases the stainings and samples property will be set to None. If you have a dataset with a Specimen module that you think should be parsable, please make an issue (see Contributing).

SlideSample

Each sample is model with the SlideSample dataclass, which represents an item in the DICOM Specimen Description Sequence

• identifier: Identifier of the sample.
• anatomical_sites: List of codes describing the primary anatomic structures of interest in the sample.
• sampled_from: The sampling (of another specimen) that was done to produce the sample (if known). If the sampled specimen also was produced through sampling, this property will give access to the full hierarchy of (known) specimens.
• uid: Unique identifier for the sample.
• localization: Description of the placement of the sample on the slide. Should be present if more than one sample is placed on the slide.
• steps: List of preparation steps performed on the sample.
• short_description: Short description of the sample (should not exceed 64 characters).
• detailed_description: Unlimited description of the sample.

Samplings

The optional sampled_from property can either be a Sampling or a UnknownSampling. Both of these specify a sampled specimen, with the difference that the UnknownSampling is used when the sampling conditions are not fully know. A Sampling is more detailed, and specifies the sampling method and optional properties such as sampling date_time, description and location.

Specimens

The specimen property of a Sampling or a UnknownSampling links to either a Specimen or a Sample. A Specimen has no known parents (e.g. could be the specimen extracted from a patient), while a Sample always is produced from one or more samplings of other Specimens or Samples. The samplings used to produce a Sample is given by its sampled_from-property. Both Specimen and Sample contain additional properties describing the specimen:

• identifier: Identifier of the specimen.
• type: Optional anatomic pathology specimen type code (e.g. "tissue specimen"). Should be a specimen type defined in CID 8103.
• steps: List of processing steps performed on the specimen.
• container: Optional container type code the specimen is placed in. Should be a container type defined in CID 8101.

Processing and staining steps

The processing steps that can be performed on a sample are:

• Sampling: Sampling of the specimen in order to produce new specimen(s). The sampling method should be a method defined in CID 8110.
• Collection: Collection of a specimen from a body. This can only be done on a Specimen, i.e. not on a specimen produced by sampling. The collection method should be a method defined in CID 8109.
• Processing: Processing performed on the specimen. The processing method should be a method defined in CID 8113.
• Embedding: Embedding done on the specimen. The embedding medium should be a medium defined in CID 8115.
• Fixation: Fixation of the specimen. The fixative should be a fixative defined in CID 8114.
• Receiving: Receiving of the specimen.
• Storage: Storage of the specimen.

The Staining(s) for a Slide contains a list of substances used for staining. The substances used should defined in CID 8112.

Every processing step (including staining) also have the optional properties date_time for when the processing was done and description for a textual description of the processing.

These steps are parsed from the SpecimenPreparationSequence following TID 8004 for each specimen identifier in the item sequence.

Exporting to json

The metadata can be exported to json:

from wsidicom.metadata.schema.json import WsiMetadataJsonSchema

with WsiDicom.open("path_to_folder") as wsi:
    metadata = wsi.metadata

schema = WsiMetadataJsonSchema()
metadata_json = schema.dump(metadata)

Settings

The strictness of parsing of DICOM WSI metadata can be configured using the following settings (see Settings):

• strict_specimen_identifier_check: Controls how to handle matching between specimen identifiers if one of the identifiers have a issuer of identifier set and the other does not. If True the identifiers are considered equal (provided that the identifier value is the same), if False the issuer of identifier must always also match. This setting is useful if for example a issuer of identifier is specified in the Specimen Description Sequence but steps in the Specimen Preparation Sequence lacks the issuer of identifier. The default value is True.
• ignore_specimen_preparation_step_on_validation_error: Controls how to handle if a step in the Specimen Preparation Sequence fails to validate. If True, only steps that fails will be ignored. If False all steps will be ignored. The default value is True.

Saving files

An opened WsiDicom instance can be saved to a new path using the save()-method. The produced files will be:

• Fully tiled. Any sparse tiles will be replaced with a blank tile with color depending on the photometric interpretation.
• Have a basic offset table (or optionally an extended offset table or no offset table).
• Not be concatenated.

By default frames are copied as-is, i.e. without re-compression.

with WsiDicom.open("path_to_folder") as slide:
    slide.save("path_to_output")

The output folder must already exists. Be careful to specify a unique folder folder to avoid mixing files from different images.

Optionally frames can be transcoded, either by a encoder setting or an encoder:

from wsidicom.codec import JpegSettings

with WsiDicom.open("path_to_folder") as slide:
    slide.save("path_to_output", transcoding=JpegSettings())

Settings

wsidicom can be configured with the settings variable. For example, set the parsing of files to strict:

from wsidicom import settings
settings.strict_uid_check = True
settings.strict_attribute_check = True

Annotation usage

Annotations are structured in a hierarchy:

• AnnotationInstance Represents a collection of AnnotationGroups. All the groups have the same frame of reference, i.e. annotations are from the same wsi stack.
• AnnotationGroup Represents a group of annotations. All annotations in the group are of the same type (e.g. PointAnnotation), have the same label, description and category and type. The category and type are codes that are used to define the annotated feature. A good resource for working with codes is available here.
• Annotation Represents a annotation. An Annotation has a geometry (currently Point, Polyline, Polygon) and an optional list of Measurements.
• Measurement Represents a measurement for an Annotation. A Measurement consists of a type-code (e.g. "Area"), a value and a unit-code ("mm")

Codes that are defined in the 222-draft can be created using the create(source, type) function of the ConceptCode-class.

Load a WSI dataset from files in folder.

from wsidicom import WsiDicom
slide = WsiDicom.open("path_to_folder")

Create a point annotation at x=10.0, y=20.0 mm.

from wsidicom import Annotation, Point
point_annotation = Annotation(Point(10.0, 20.0))

Create a point annotation with a measurement.

from wsidicom import ConceptCode, Measurement
# A measurement is defined by a type code ('Area'), a value (25.0) and a unit code ('Pixels).
area = ConceptCode.measurement('Area')
pixels = ConceptCode.unit('Pixels')
measurement = Measurement(area, 25.0, pixels)
point_annotation_with_measurment = Annotation(Point(10.0, 20.0), [measurement])

Create a group of the annotations.

from wsidicom import PointAnnotationGroup
# The 222 supplement requires groups to have a label, a category and a type
group = PointAnnotationGroup(
    annotations=[point_annotation, point_annotation_with_measurment],
    label='group label',
    categorycode=ConceptCode.category('Tissue'),
    typecode=ConceptCode.type('Nucleus'),
    description='description'
)

Create a collection of annotation groups.

from wsidicom import AnnotationInstance
annotations = AnnotationInstance([group], 'volume', slide.uids)

Save the collection to file.

annotations.save('path_to_dicom_dir/annotation.dcm')

Reopen the slide and access the annotation instance.

slide = WsiDicom.open("path_to_folder")
annotations = slide.annotations

Setup environment for development

Requires poetry installed in the virtual environment.

git clone https://github.com/imi-bigpicture/wsidicom.git
poetry install

To watch unit tests use:

poetry run pytest-watch -- -m unittest

The integration tests uses test images from nema.org that's needs to be downloaded. The location of the test images can be changed from the default tests\testdata\slides using the environment variable WSIDICOM_TESTDIR. Download the images using the supplied script:

python .\tests\download_test_images.py

If the files are already downloaded the script will validate the checksums.

To run integration tests:

poetry run pytest -m integration

Data structure

A WSI DICOM pyramid is in wsidicom represented by a hierarchy of objects of different classes, starting from bottom:

• WsiDicomReader, represents a WSI DICOM file reader, used for accessing WsiDicomFileImageData and WsiDataset.
• WsiDicomFileImageData, represents the image data in one or several (in case of concatenation) WSI DICOM files.
• WsiDataset, represents the image metadata in one or several (in case of concatenation) WSI DICOM files.
• WsiInstance, represents image data and image metadata.
• Level, represents a group of instances with the same image size, i.e. of the same level.
• Pyramid, represents a group of levels, i.e. the pyrimidal structure.
• Pyramids, represents a collection of pyramids, each with different image coordate system or extended depth of field.
• WsiDicom, represents a collection of pyramids, labels and overviews.

Labels and overviews are structured similarly to levels, but with somewhat different properties and restrictions. For DICOMWeb the WsiDicomFile* classes are replaced with WsiDicomWeb* classes.

A Source is used to create WsiInstances, either from files (WsiDicomFileSource) or DICOMWeb (WsiDicomWebSource), and can be used to to Initiate a WsiDicom object. A source is easiest created with the open() and open_web() helper functions, e.g.:

slide = WsiDicom.open("path_to_folder")

Code structure

• codec - Encoders and decoders for image pixel data.
• file - Implementation for reading and writing DICOM WSI files.
• group - Group implementations, e.g. Level.
• instance - Instance implementations WsiIsntance and WsiDataset, the metaclass ImageData and ImageData implementations WsiDicomImageData and PillowImageData.
• metadata - Metadata models and schema for serializing and deserializing to DICOM and json.
• series - Series implementations Levels, Labels, and Overview.
• web - Implementation for reading DICOM WSI from DICOMWeb.
• conceptcode.py - Handling of DICOM concept codes.
• config.py - Handles configuration settings.
• errors.py - Custom errors.
• geometry.py - Classes for geometry handling.
• graphical_annotations - Handling graphical annotations.
• source.py - Metaclass Source for serving WsiInstances to WsiDicom.
• stringprinting.py - For nicer string printing of objects.
• tags.py - Definition of commonly used DICOM tags.
• threads.py - Implementation of ThreadPoolExecutor that does not use a pool when only single worker.
• uid.py - Handles DICOM uids.
• wsidicom.py - Main class with methods to open DICOM WSI objects.

Adding support for other file formats

Support for other formats (or methods to access DICOM data) can be implemented by creating a new Source implementation, that should create WsiInstances for the implemented formats. A format specific implementations of the ImageData is likely needed to access the WSI image data. Additionally a WsiDataset needs to be created that returns matching metadata for the WSI.

The implemented Source can then create a instance from the implemented ImageData (and a method returning a WsiDataset):

image_data = MyImageData('path_to_image_file')
dataset = create_dataset_from_image_data(image_data)
instance = WsiInstance(dataset, image_data)

The source should arrange the created instances and return them at the level_instances, label_instances, and overview_instances properties. WsiDicom can then open the source object and arrange the instances into levels etc as described in 'Data structure'.

Other DICOM python tools

• pydicom
• highdicom
• wsidicomizer
• dicomslide
• openslide-python

Contributing

We welcome any contributions to help improve this tool for the WSI DICOM community!

We recommend first creating an issue before creating potential contributions to check that the contribution is in line with the goals of the project. To submit your contribution, please issue a pull request on the imi-bigpicture/wsidicom repository with your changes for review.

Our aim is to provide constructive and positive code reviews for all submissions. The project relies on gradual typing and roughly follows PEP8. However, we are not dogmatic. Most important is that the code is easy to read and understand.

Acknowledgement

wsidicom: Copyright 2021 Sectra AB, licensed under Apache 2.0.

This project is part of a project that has received funding from the Innovative Medicines Initiative 2 Joint Undertaking under grant agreement No 945358. This Joint Undertaking receives support from the European Union’s Horizon 2020 research and innovation programme and EFPIA. IMI website: <www.imi.europa.eu>