Abstract
Quantitative analysis of clinical image data is an active area of research that holds promise for precision medicine, early assessment of treatment response, and objective characterization of the disease. Interoperability, data sharing, and the ability to mine the resulting data are of increasing importance, given the explosive growth in the number of quantitative analysis methods being proposed. The Digital Imaging and Communications in Medicine (DICOM) standard is widely adopted for image and metadata in radiology. dcmqi (DICOM for Quantitative Imaging) is a free, open source library that implements conversion of the data stored in commonly used research formats into the standard DICOM representation. dcmqi source code is distributed under BSD-style license. It is freely available as a precompiled binary package for every major operating system, as a Docker image, and as an extension to 3D Slicer. Installation and usage instructions are provided in the GitHub repository at https://github.com/qiicr/dcmqi. Cancer Res; 77(21); e87–90. ©2017 AACR.
Introduction
Quantitative analysis of imaging data in clinical setting holds the promise for objective, noninvasive, and reproducible characterization of disease. Development of the algorithms to enable such analyses has been an active area of clinical research for decades, leading to a proliferation of the analysis tools and public datasets. In parallel, there is an increasing push for improving reproducibility of scientific research and sharing of both the original image data and the results of analysis in a machine-readable and reusable format (1). Meaningful data sharing and interoperability are hampered by the use of ad hoc data formats in the research community.
We introduce a free open source library, dcmqi, that implements conversion of the results commonly produced in quantitative imaging research into the standard Digital Imaging and Communications in Medicine (DICOM; ref. 2) format.
DICOM (2) is the most widely used standard for clinical image data and metadata. DICOM has been adopted by most equipment manufacturers in the field of radiology. DICOM enables interoperability of the various processing, archival, and visualization components, both clinical and research. DICOM provides a framework for communicating not only images, but also imaging-related information, accompanied by the structured and often coded “composite context” attributes, such as those identifying the patient, and describing acquisition and provenance of the data. Composite context enables indexing and cross-referencing of the data. DICOM applications are not restricted to clinical use only. The standard defines objects for storing image and image-related data, which are also suitable to support imaging research, including that done using preclinical and animal imaging.
Adoption of DICOM for research applications has been limited for a number of reasons relating to the overall complexity of the standard, lack of understanding of its capabilities, and the significant effort required for the DICOM software maintenance. Existing DICOM toolkits reduce implementation effort, but often lack specific higher level semantic abstractions, structures, and concepts relevant in quantitative imaging. dcmqi aims to lower some of the barriers toward adoption of the DICOM standard for the purposes of quantitative imaging research. This article provides a high-level overview of dcmqi. The details are available in the dcmqi user guide.
Materials and Methods
dcmqi is a collection of software libraries and command line tools for converting data between the commonly used research formats and their DICOM representation counterparts. dcmqi is the result of generalization of the conversion tools we developed in an earlier project that demonstrated application of DICOM for imaging-based treatment response assessment in nuclear medicine oncological application (3).
We chose C++ as the primary language for implementing dcmqi to integrate with the existing open source libraries and tools. dcmqi builds upon DICOM Toolkit (DCMTK; ref. 4) and Insight Toolkit (ITK; ref. 5). DCMTK provides a general purpose application programming interface for reading, creating, and writing DICOM objects. ITK is commonly used for developing image segmentation and registration tools and provides support for popular research data formats. The decision to create a new library instead of implementing the new functionality within DCMTK was motivated by the need of using ITK and the wish to keep this dependence compartmentalized and well defined.
Conversion from research formats into DICOM relies on metadata necessary to create a valid and meaningful DICOM object. dcmqi aims to reduce the user burden in providing such metadata, for example, by automatically propagating attributes of the composite context that are available in the reference DICOM data specified by the user (e.g., patient and study information), and initializing references to the source objects used in the derivation. Inevitably, some of the metadata are not present in the source data and must be populated by the user. These attributes are specified using JavaScript Object Notation (JSON), formalized by a JSON Schema (6) defined for the individual data types we support. This approach results in documents that are both human and machine readable and can be automatically validated. dcmqi uses ajv (7) for validating the conformance of JSON documents with the schema. This validation is done as part of the continuous integration (CI) testing and can be conveniently accessed via the web-based validation tools that accompany dcmqi. Reverse conversion from DICOM into research formats is also supported to enable integration with the tools that are not DICOM aware.
Results
This article accompanies the initial release of dcmqi. The source code, available at https://github.com/qiicr/dcmqi, has been in development since January 2016 and is distributed under a nonrestrictive BSD-style license. Installation and usage instructions are available within the repository. CI is implemented for Windows, Linux, and macOS to exercise the build process, functionality of the individual converters, and to check the validity and compliance with the schema for the JSON components of the library. Compliance of the resulting objects with the standard is verified using dciodvfy from dicom3tools (8). dcmqi converters can be built from source, are available as a ready-to-use Docker image or via a visual user interface within 3D Slicer (9). OS-specific installers of dcmqi are provided for Windows, Linux, and macOS. An introductory tutorial is available at http://qiicr.org/dcmqi-guide/tutorials/intro.html. The process of setting up the environment and installing dcmqi Docker image is explained in Supplementary Video S1.
Supplementary Video S1
This video explains how to set up the environment, install related tools, and install dcmqi from Docker Hub. Supplementary Video S1
As of Spring 2017, dcmqi supports the following types of quantitative image analysis results:
(i) Rasterized image segmentations defining image regions of interest as DICOM Segmentation image objects. Examples of segmentation tasks include generation of atlases of normal anatomy, or definition of the neoplastic areas for subsequent volumetric evaluation and other quantitative characterization tasks, including radiomics analysis (10). Metadata include coded semantics of the segmented region, anatomic location of the finding, type of the algorithm used, and recommended color for the visualization of the segmentation. Supplementary Video S2 demonstrates the process of converting segmentation result generated using ITK-Snap (11), a popular image segmentation tool, into a DICOM Segmentation image object.
Supplementary Video S2
This video guides the viewer over the process of how a segmentation created using ITK-Snap can be converted into a standard DICOM representation using dcmqi. Supplementary Video S2
(ii) Parametric maps as DICOM Parametric Map objects: result of postprocessing of image data. As an example, apparent diffusion coefficient maps derived by fitting various models to diffusion-weighted MRI data have shown promise in characterizing aggressiveness of prostate cancer (12). Resulting objects contain coded quantity and units for the pixel-wise values, and the method used to derive those values.
(iii) Measurements calculated over the volumetric regions defined by rasterized image segmentations, as DICOM Structured Reporting (SR) Template ID 1500 (SR TID 1500) objects. Measurements can range from simple calculation of region volume to more complex quantitative measures, such as image textures (13). The resulting SR objects contain coded quantity, units, and reference to the segmentation object used for measurements calculation. The process of loading and visualizing segmentation objects and the associated measurements from a publicly available TCIA dataset (3) using 3D Slicer platform (9) is illustrated in Supplementary Video S3.
Supplementary Video S3
This video guides the user over the steps how to visualize the DICOM segmentation objects and structured measurements using 3D Slicer platform. Supplementary Video S3
Functionality of dcmqi is exposed via graphical user interface in the QuantitativeReporting extension of 3D Slicer. QuantitativeReporting supports the interactive image annotation and quantitative analysis workflow that allows an operator to load a DICOM image series, use automated segmentation tools of 3D Slicer to define the volume of interest, automatically calculate the volume of interest-based measurements, and save the result of this workflow as a cross-referenced collection of DICOM objects. QuantitativeReporting can also be used to reload the volumetric measurements stored using SR TID 1500 together with the referenced image and segmentation data (see Fig. 1 and Supplementary Video S3).
Result of loading postprocessing results stored in DICOM for subject 139 from the TCIA QIN-HEADNECK collection (3) using QuantitativeReporting extension of 3D Slicer, which uses dcmqi for DICOM data conversion. The interactive visualization includes multiplanar reformats of the original PET image data with overlay of segmented areas, 3-dimensional rendering (top right), coded terminology of the annotated regions (yellow tooltip rectangle), and a table of the image-based measurements over the segmented areas. All of these data components were restored from DICOM objects retrieved from TCIA.
We are participating in community projects to raise awareness of the importance of interoperable communication of derived imaging information and the relevant capabilities of the DICOM standard and supporting tools. DICOM for Quantitative Imaging (DICOM4QI) is a “connectathon”-style initiative that we organized within Radiological Society of North America (RSNA) annual convention (14). DICOM4QI gathers developers of both academic and commercial platforms to test DICOM-based interoperability. In 2016, a number of both open source and commercial platforms chose dcmqi to implement DICOM research data storage support (14). Within the NCI Quantitative Imaging Network (QIN), we apply dcmqi for converting the analysis results produced by the QIN algorithm challenges and community projects into DICOM representation.
Discussion
The goal of dcmqi is to simplify the use of the DICOM standard for the purposes of quantitative imaging research to support standardized collection, archival, and sharing of research results. Previous related efforts aimed at providing more general-purpose DICOM conversion and editing tools (e.g., dccp from dicom3tools, dcmodify from DCMTK, and RSNA Clinical Trials Processor). We limit the scope of dcmqi specifically to the needs and data types relevant in quantitative imaging research and focus on the task of use case-driven conversion of the data.
A key challenge in maintaining dcmqi converters is that the standard itself does not provide machine-readable, formalized descriptions of the DICOM objects. The metadata schemas we introduced formalize dcmqi conversion process, but do not aim to redefine the standard. Although those schemas will evolve in the future, the output objects will comply with the same DICOM standard. Our future plans include improved flexibility and harmonization of dcmqi parameter schema with the standard and the standard-defined JSON model of DICOM (15). The latter was not used directly due to our prioritization for readability of the metadata JSON.
The DICOM standard is evolving alongside dcmqi. We work with the DICOM community to resolve gaps, limitations, or inconsistencies that we identify in the process of implementation, as discussed in ref. 3. We expect further refinements of both the library and the standard as we expand the collection of use cases and applications of dcmqi in quantitative imaging research.
Disclosure of Potential Conflicts of Interest
S. Pieper reports receiving other commercial research support from Siemens. D. Clunie has ownership interest (including patents) in PixelMed and is a consultant/advisory board member for Carestream, Curemetrix, CVUS, GE Healthcare, ImagingAppMarketplace, Laitek, MDDX, NCI, NEMA, Protom, and Sorna. No potential conflicts of interest were disclosed by the other authors.
Authors' Contributions
Conception and design: C. Herz, J.-C. Fillion-Robin, C. Pinter, S. Pieper, D. Clunie, R. Kikinis, A. Fedorov
Development of methodology: J.-C. Fillion-Robin, M. Onken, A. Lasso, C. Pinter, S. Pieper, D. Clunie, R. Kikinis, A. Fedorov
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D. Clunie, R. Kikinis, A. Fedorov
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D. Clunie, R. Kikinis, A. Fedorov
Writing, review, and/or revision of the manuscript: C. Herz, J.-C. Fillion-Robin, M. Onken, J. Riesmeier, C. Pinter, G. Fichtinger, S. Pieper, D. Clunie, R. Kikinis, A. Fedorov
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Herz, R. Kikinis, A. Fedorov
Study supervision: R. Kikinis, A. Fedorov
Other (development of the software described in the article): C. Herz, J.-C. Fillion-Robin, M. Onken, J. Riesmeier, D. Clunie, A. Fedorov
Grant Support
This work was supported in part by NIH awards U24CA180918 (principal investigators: A. Fedorov and R. Kikinis; this is the main award funding the development of dcmqi through 2018), P41EB015902 (to principal investigator R. Kikinis), P41EB015898 (to principal investigator C. Tempany), and R01EB014955 (to principal investigator R. Kikinis).
- Received February 3, 2017.
- Revision received April 26, 2017.
- Accepted June 23, 2017.
- ©2017 American Association for Cancer Research.
Volume 77, Issue 21
