pydicom-seg

Reading and writing of DICOM-SEG medical image segmentation storage files using pydicom as DICOM serialization/deserialization library.

Motivation

Converting DICOM-SEG files into ITK compatible data formats, commonly used for research, is made possible by the dcmqi project for some time. However, the project is written in C++ and offers only access to the conversion via the binaries itkimage2segimage and segimage2itkimage. After a conversion of a DICOM-SEG file to ITK NRRD file format, the user has to scan the output directory for generated files, load them individually and potentially combine multiple files to the desired format.

This library aims to make this process much easier, by providing a Python native implementation of reading and writing functionality with support for numpy and SimpleITK. Additionally, common use cases like loading multi-class segmentations are supported out-of-the-box.

Installation

Install from PyPI

pip install pydicom-seg

Install from source

This package uses Poetry as build system.

git clone https://github.com/razorx89/pydicom-seg.git
cd pydicom-seg
poetry build
pip install dist/pydicom_seg-<version>-py3-none-any.whl

Getting Started

Loading binary segments

import pydicom
import pydicom_seg
import SimpleITK as sitk

dcm = pydicom.dcmread('segmentation.dcm')

reader = pydicom_seg.SegmentReader()
result = reader.read(dcm)

for segment_number in result.available_segments:
    image_data = result.segment_data(segment_number)  # directly available
    image = result.segment_image(segment_number)  # lazy construction
    sitk.WriteImage(image, f'/tmp/segmentation-{segment_number}.nrrd', True)

Loading a multi-class segmentation

dcm = pydicom.dcmread('segmentation.dcm')

reader = pydicom_seg.MultiClassReader()
result = reader.read(dcm)

image_data = result.data  # directly available
image = result.image  # lazy construction
sitk.WriteImage(image, '/tmp/segmentation.nrrd', True)

Saving a multi-class segmentation

segmentation: SimpleITK.Image = ...  # A segmentation image with integer data type
                                     # and a single component per voxel
dicom_series_paths = [...]  # Paths to an imaging series related to the segmentation
source_images = [
    pydicom.dcmread(x, stop_before_pixels=True)
    for x in dicom_series_paths
]
template = pydicom_seg.template.from_dcmqi_metainfo('metainfo.json')
writer = pydicom_seg.MultiClassWriter(
    template=template,
    inplane_cropping=True,  # Crop image slices to the minimum bounding box on 
                            # x and y axes
    skip_empty_slices=True,  # Don't encode slices with only zeros
    skip_missing_segments=False,  # If a segment definition is missing in the
                                  # template, then raise an error instead of
                                  # skipping it.
)
dcm = writer.write(segmentation, source_images)
dcm.save_as('segmentation.dcm')

License

pydicom-seg is distributed under the MIT license.

jmgo/pydicom-seg