If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Image fusion, clustering and resampling operations.
Abstract
Introduction
One of the main issues in the field of clinical research is to enhance clinical databases with information from imaging (CT, MR, PET-scan), contouring (RTstruct), or produced by TPS such as dose distribution (RTdose) or treatment plans (RTplan). To perform these analyses automatically, we propose the new open-source package “espadon”, developed in R environment. This package also opens up numerous perspectives for TPS-independant calculation, automation and processing of DICOM data.
Results
The espadon package converts DICOM objects into espadon objects. Several tools have been developed to manipulate these objects and extract the desired information. In addition to decode DICOM files and pseudonomize them, the great advantage of espadon is that it presents the links between patient data (images, structures, treatment plans) in a didactic way, respecting the dates of the examinations. It can visualize volumes or structures in 2D or 3D, resample volumes, segment them, and change geometric frames of reference. It integrates dose-volume histogram functions on a selection, with Monte Carlo calculations of random shifts of contours. It offers the automatic calculation of several usual radiotherapy indices, as well as the calculation of Gamma and Chi indices.
Conclusions
Espadon is a toolkit designed to be easily used by radiotherapists, medical physicists or students. Espadon's functions are implemented in an R script, and allow the automatic extraction or calculation of data from DICOM files, which can be used for statistical modelling or machine-learning in the R environment. This package is available on the Comprehensive R Archive Network (CRAN) repository.
] Indeed, finding the influencing factors of the responses to a treatment is one of the key issues to optimize the treatment proposed for a specific patient and its post-treatment management. Espadon (acronym of Easy Study of PAtient Dicom data in ONcology) has been developed in the framework of this research. Its objective is to enhance clinical databases, generally containing the outcomes and pathways of patients, with objective data from imaging, dose distribution and treatment geometry or organ location in relation to the tumor.
] environment, which is also a very suitable language for statistical studies, data mining, machine learning and database creation. It is available for free on the Comprehensive R Archive Network (CRAN) repository [
Fontbonne Cathy et al (Online; accessed 2023-03-21) “Easy Study of Patient DICOM Data in Oncology [R Package Espadon Version 1.3.1].” The Comprehensive R Archive Network , Comprehensive R Archive Network (CRAN), 3 Mar. 2023, https://CRAN.R-project.org/package=espadon.
]. Espadon is a coherent and logical set of functions that can be easily implemented in a study automation script. It allows to manipulate and visualize DICOM objects, and to systematically calculate most of the usual radiotherapy indices. Espadon can also manipulate the raw data of DICOM files: for example, it has been used in ArDCore [
Combes S, Bacry E, Fontbonne C. Health Data Hub ; plateforme des données de santé en France, application à l’oncologie radiothérapie Cancer Radiother 2020;24(6-7):762-767, ISSN 1278-3218. doi: 10.1016/j.canrad.2020.07.003.
] software that automatically pseudonymizes and sorts patient DICOM files.
In this article, we will use simple examples to illustrate how the espadon package can be used to access or compute information of interest to research actors in radiotherapy.
2. Methods
Espadon was designed and authored by Cathy Fontbonne and Jean-Marc Fontbonne, with Nathan Azemar as contributor. It is a package written in R language, with some functions using C++ for better performances. The package uses the capabilities offered by other packages: for example qs [
Schlager S (2017). “Morpho and Rvcg - Shape Analysis in R.” In Zheng G, Li S, Szekely G (eds.),_Statistical Shape and Deformation Analysis, 217-256. Academic Press. ISBN 9780128104934.
As the objective of espadon is to study DICOM exports related to radiotherapy, the package is first able to decode a DICOM file into an R data.frame or R list. It also includes a function to modify the ASCII fields of the DICOM file.
Espadon incorporates a tag dictionary, which can be called using the dicom.tag.dictionary() function. The tag dictionary is defined by the nema standard [
]. It can be completed, as needed, by constructor dictionaries. In addition to the NEMA standard dictionary, the espadon package provides the Raysearch Laboratories dictionary [
Espadon then interprets all the DICOM files by creating espadon objects that belong to specific classes. The class of an object ensures how the object is used by the espadon functions, and thus simplifies their use. Each object of a specific class contains information (fields) that are necessary for the use of that object. For example, RTdose dose distribution, CT and MRI scans create objects of the volume class. This class necessarily includes the number of voxels in the three dimensions of space, the size of the voxel in millimeters, the value (if requested) and the unit of each voxel, as well as the information to locate this volume in space. The DICOM RTstruct files form struct class objects that necessarily include a table of information about the regions of interest (name, color of the RoI in the DICOM file, but also calculated data such as volume, barycenter, coordinates of the end points) and a list with the coordinates of the contours in space.
In the current version there are other object classes, which are listed by calling the espadon.class() function and explained in the help page accessible by the command help(espadon.class). One of the important classes is the “patient” class. A patient object contains not only a description of all the patient's examinations, the examinations themselves sorted by modality, but also a t.mat class object managing the spatial registrations. To simplify the management of geometric reference frames, the espadon package automatically assigns to each reference frame defined by the DICOM tag (0020,0052) a pseudonym “ref1″ then ”ref2″… Each DICOM object loaded is then assigned a new alias by default, consisting of, among other things, its study date, the pseudonym of its reference frame, and its modality. This new alias makes it possible to quickly visualize the chronology of a patient's examinations, as well as the links that exist between them.
As the interpretation of DICOM data to create an espadon object can be a long process, the package proposes to convert DICOM files into *.Rdcm files with the function dicom.to.Rdcm.converter(). The backup of the object being performed with the qs package, the speed of access to the data as well as the amount of memory necessary for the backup are considerably improved. As a comparison, a database including the imaging and dose distribution data of 441 patients uses 239 GB of memory for 324,164 DICOM files, and uses only 79.5 GB for 4499 Rdcm files. Considering now the data of a single patient including 2 CT-scan, 5 RTdose, 2 RTplan, 2 RTstruct, the data are loaded entirely in 130 s from the 907 DICOM files (717 MB) with the function load.patient.from.dicom(), compared to 21 s from the 11 Rdcm files (248 MB) with the function load.patient.from.Rdcm(). In this second case, it is nevertheless necessary to perform a one-time 114 s conversion from DICOM files to Rdcm files beforehand with the function dicom.to.Rdcm.converter(). The measurements were preformed on computer with Intel Core i9 CPU (2.40 GHz) and 64 GB of system memory, running 64-bit Microsoft Windows 10 Professional. The complete content of a *.Rdcm file can be accessed with the function load.Rdcm.raw.data(). It is a list composed of, among other things, a “header” list containing interpreted data (e.g. name of the patient, number of voxels or number of contours, etc.), an “address” list giving the addresses of the content of the DICOM tags in the raw data of the DICOM file, and a “data” list giving the list of exports in the R-list of the various DICOM files making up the object.
Espadon's primary objective is to facilitate quality control of patient data. Several display functions with palette management and view overlays are available.
Espadon will also allow the creation of new objects that have not been created by the TPS: RTstruct, differential histogram, dose-volume histogram (DVH), and meshes. It has several tools for resampling imaging volumes or reducing volumes to focus on the area of interest and speed up calculations. Binary volumes can define regions of interest, creating a new “binary” modality. Operations on binary volumes such as logical, morpho-mathematical or segmentation operations are proposed, allowing the user to make new selections undefined in the original data.
Finally, espadon will allow the calculation of most of the radiotherapy indices such as Prescription Isodose Target Volume (PITV), Prescription Dose Spillage (PDS), Conformity Index (CI), Conformity Number (CN), New conformity index (NCI), Dice Similarity Coefficient (DSC), Conformity Distance Index (CDI), Triple Point Conformity Scale (CS3), Underdosed lesion factor (ULF), Overdosed healthy tissues factor (OHTF), Geometric Conformity Index (gCI), COIN Conformity Index, generalized COSI, several Homogeneity Index (HI), Gradient Index (GI) and modified Gradient Index (mGI). All these indices are compiled by Kaplan et al. [
A systematically compiled set of quantitative metrics to describe spatial characteristics of radiotherapy dose distributions and aid in treatment planning.
] and referenced in the reference manual of espadon package.
With the functions rt.gamma.index() and rt.chi.index(), the espadon package can also compute, on a 3D volume or 2D image, the γ-index and χ-index, as defined by Low et al. [
In this section, we will focus on a few example applications that provide a quick understanding of how to exploit espadon features to understand and control a patient's imaging and dose distribution data, and to calculate the usual indicators in radiation therapy. We have the DICOM data of patient A, consisting of a CT scan, an MRI, three RTdose, an RTplan, two RTstruct, and a registration matrix.
The DICOM files of patient A are pseudonymised data, generated by the dicom.raw.data.anonymizer() function of the espadon package. This function replaced the patient's name and identifier with a user-selected string, removed all other patient data except age, weight, height, gender, and birthday, shifted all dates by a selected number of days, removed all address, phone, physician, operator, author, reviewer, and department data. More generally, the dicom.set.tag.value() function can be used to modify the content of any tag, provided that this content is a character string.
Appendix A provides the instructions that were used to generate the CT, MR, RTdose displays, DVH numbers and index calculations in this paper.
3.1 Loading and displaying patient imaging and dose distribution data
The data from patient A is loaded into memory by creating a “patient” class object, either by the simple function load.patient.from.dicom(), or, if it is preferable to convert the data into pre-formatted Rdcm data for faster later use, by using the functions dicom.to.Rdcm.converter() followed by load.patient.from.Rdcm(). To manage memory space, objects can only be partially loaded, depending on whether the data argument is TRUE or FALSE. If FALSE, the objects of the volume class, for example, will be missing the values of the voxels ($vol3D.data field), which can then be loaded using the load.obj.data() function. In all cases, the links between the objects contained in the espadon class object “patient” can be represented using the display.obj.links() function, taking into account the spatial registrations (Fig. 1a) or not (Fig. 1b).
Fig. 1Links between the different objects, defined by the display.obj.link() function. The arrows indicate the objects that have been used to build another object. The lines indicate that the objects are sub-objects belonging to the same object. The objects with a green outline are approved for treatment. (a) The argument group.by.connected.For is set to TRUE: all the objects that are connected by a spatial registration are represented with the same color. (b) The argument group.by.connected.For is set to FALSE: the objects having a different frame of reference have a different color. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
The loaded objects can easily be represented by espadon in any reference frame. For example, patient A has a CT scan and an MRI scan with different geometric reference frames. Fig. 2 shows different views obtained with the display.plane() function of the CT or MRI, with superposition of the dose and PTV contour.
Fig. 2(a) Display of the CT transverse view with palette pal.RVV
. Display of transverse (b) and frontal (c) overlap of CT, RTdose, and the PTV contour. (d) Overlap of transverse view of the MR, and RTdose in the MR frame of reference.
Fig. 3 shows the use of espadon 3D functions: the display of the PTV contours (display.3D.contour() function), the stacking of the selected MRI planes in space (display.3D.stack() function), and the three cross-sections, frontal and sagittal at a given point (display.3D.sections() function).
Fig. 33D Display in CT frame of reference using three espadon functions: - display.3D.contour for the PTV contour display in blue, - display.3D.sections for transverse, frontal and sagittal views, with palette pal.RVV
and transparency management, - display.3D.stack for display of some MR slices in the default grayscale palette. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
The user has the choice to create his/her own color palette. However espadon also proposes the use of the pal.RVV palette in which Hounsfield units in the range −1000 HU to 1000 HU are converted to realistic colors, as defined by J.C. Silverstein et al [
3.2 Histograms computation and radiotherapy indices
Espadon allows to compute very quickly differential histograms of volume class objects in predefined areas either by a region of interest (RoI) coming from an RTstruct file with the function histo.from.roi(), or by a selection by binary volume with the histo.from.bin() function. In addition, the histo.from.roi() function can simulate random displacements of the contours according to a normal distribution, by Monte Carlo method.
Fig. 4 illustrates the result of computing the dose-volume histogram of a small RoI in the RTstruct file of patient A: the chiasma. In this example, the histo.from.roi() function first calculates the differential histogram with a Monte Carlo simulation on 100 samples (argument MC). The cumulative histogram, called in this case the dose-volume histogram or DVH, is then calculated using the histo.DVH() function. The display.DVH() and display.DVH.pc() functions are used to display clear graphs of the DVH, including the 100%, 95% and 50% quantile areas of the Monte Carlo data.
Fig. 4Use of the display.DVH() and display.DVH.pc() functions to display the optic chiasm DVH in cm3 (a) or % (b), with a simulation of a shift of 1 mm in all directions. The grey areas represent from lightest to darkest, respectively, the 100%, 95% and 50% quantiles of the MC data.
It may be convenient to compute only a few dose points in the DVH. This is done by means of the rt.indices.from.roi() function. Table 1 gives for example the mean dose and D.5%, D.95%, D0.1cc and D0.2cc dose point values:
The rt.indices.from.roi() and rt.indices.from.bin() functions actually provide the ability to compute many more radiotherapy indices. By providing the RTdose volume, the prescription dose, the target volumes to be treated (e.g., the different PTVs, depending on which function is used either as RoI in an RTstruct or as a binary volume), and the healthy volumes to be protected, it is possible, as shown in Table 2, to compute the Prescription Isodose Target Volume (PITV) as defined by E.Shaw [
Table 2Use of the function rt.indices.from.roi(), with the PTV as target region of interest (RoI), chiasma as healthy RoI, and 48.77 Gy as prescription dose. The function provides the requested dose data, volume, conformity and gradient table.
The available conformity, homogeneity and gradient indices are all defined and referenced in the reference manual of espadon package.
3.3 Geometric tools
The available geometrical tools are used to simplify the manipulation of volume objects. First, it is possible to increase or reduce a volume. The add.margin() function adds or removes a margin to the volume, the nesting.cube() function restricts the volume to a volume defined by the two vertices, the nesting.bin() function reduces the volume to a binary selection and nesting.roi() reduces the volume to the selection defined by a region of interest defined in an RTstruct, as illustrated in Fig. 5:
Fig. 5(a) 3D view of MRI and brain RoI. (b) 3D view of the reduction of the MRI volume around the contours of the “brain”.
In order to add volumes together (vol.sum() function), it is also possible to regrid a volume, thanks to the vol.regrid() function. One of the applications is, for example, to cumulate two dose distributions from different treatments. Fig. 6 shows, for example, the total dose of a patient who received a proton treatment and a boost with a CyberKnife machine. Each dose distribution was built from the contours of two different RTstruct files. The barycenters of the common Regions of Interest (RoI) were used to connect the two dose distributions, as explained in the manual Dose-cumulation.pdf available on the espadon website [
Fontbonne C., Fontbonne J.M. (Online; accessed 2023-03-17) Espadon, easy study of patient dicom data in oncology: toolbox for automation, exploitation and processing of DICOM files and images. https://espadon.cnrs.fr.
Fig. 6(a) Transverse view of the RTdose from proton treatment machine. (b) Transverse view of the boost RTdose from CyberKnife machine. (c) Sum of the two dose distributions.
Binary volumes are important tools for selecting only the desired part of a volume. They work like masks. The values of the voxels are either TRUE, FALSE or NA (i.e. indeterminate). Fig. 7 illustrates how to select the MRI region of patient A bounded by the brain. Only three functions were used: nesting.bin() to limit the working area and thus to reduce the computation time, bin.from.roi() to select the brain from the RoI provided by the RTstruct file, and vol.from.bin(), to select the relevant area in the MRI.
Fig. 7Selection of the brain from the patient A MRI. (a) MRI reduced to the brain (with a 2 cm margin). (b) Binary selection of the brain. (c) Selection of the brain in the MRI.
In the espadon toolkit, several mathematical morphology functions allow to increase (bin.dilation()), decrease (bin.erosion()), merge and fill holes (bin.closing()), and smooth (bin.opening()) binary volumes.
One of the applications of binary volumes is the definition of isodoses. A simple binary selection on the voxel value of a dose distribution (bin.from.vol() function) followed by an export to a espadon struct object (struct.from.bin() function) builds a network of isodoses (struct.merge() function), as illustrated in Fig. 8.
Fig. 8Isodoses from dose distribution, using espadon functions bin.from.vol(), struct.from.bin() and struct.merge().
Fontbonne Cathy et al (Online; accessed 2023-03-21) “Easy Study of Patient DICOM Data in Oncology [R Package Espadon Version 1.3.1].” The Comprehensive R Archive Network , Comprehensive R Archive Network (CRAN), 3 Mar. 2023, https://CRAN.R-project.org/package=espadon.
]. It allows radiotherapists, medical physicists and students to access an R tool, independent of the treatment planning system (TPS), to extract data from DICOM files in order to enhance clinical data a posteriori.
By forming a coherent set of functions, it allows users to write their own script in R, which may eventually enrich the package. The applications are numerous: from the creation of new contours to the automatic calculation of indices, it opens the way to multi-TPS analysis, the analysis of inter-operator practices, intra- and inter-centric comparisons and dose calculation or contouring challenges.
The espadon package will improve and develop over time: we plan to include TCP/NTCP and geometric index calculations.
Like all R packages, help is available for each espadon function in the reference manual, or with the simple “help (function)” instruction: the input parameters are explained and an example of use is proposed. Other explanations and use cases are available on the espadon website [
Fontbonne C., Fontbonne J.M. (Online; accessed 2023-03-17) Espadon, easy study of patient dicom data in oncology: toolbox for automation, exploitation and processing of DICOM files and images. https://espadon.cnrs.fr.
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
Espadon was developed as part of the PMRT project, co-funded by the European Union and the Normandy Region (ERDF-FSE fund 2014-2020, under the grant agreement n°18P03532/ 18 E01765). We would like to thank all our collaborators who shared their needs with us and allowed us to test the espadon package: A. Chaikh (Institut de Radioprotection et de Sûreté Nucléaire France), Dr. S. Guihard (Institut de Cancérologie Strasbourg Europe, France) , C. Legrand (Institut de Cancérologie de l'Ouest [Angers], France), C. Mazzara (Union Mutualiste pour la Gestion du Groupe Hospitalier Mutualiste de Grenoble, France), A. Moignier (Institut de Cancérologie de l'Ouest [Nantes], France), Pr. J. Thariat (Centre François Baclesse, Centre de Lutte contre le Cancer, Caen France).
Appendix A. Supplementary data
The following are the Supplementary data to this article:
Fontbonne Cathy et al (Online; accessed 2023-03-21) “Easy Study of Patient DICOM Data in Oncology [R Package Espadon Version 1.3.1].” The Comprehensive R Archive Network , Comprehensive R Archive Network (CRAN), 3 Mar. 2023, https://CRAN.R-project.org/package=espadon.
Combes S, Bacry E, Fontbonne C. Health Data Hub ; plateforme des données de santé en France, application à l’oncologie radiothérapie Cancer Radiother 2020;24(6-7):762-767, ISSN 1278-3218. doi: 10.1016/j.canrad.2020.07.003.
Schlager S (2017). “Morpho and Rvcg - Shape Analysis in R.” In Zheng G, Li S, Szekely G (eds.),_Statistical Shape and Deformation Analysis, 217-256. Academic Press. ISBN 9780128104934.
A systematically compiled set of quantitative metrics to describe spatial characteristics of radiotherapy dose distributions and aid in treatment planning.
Fontbonne C., Fontbonne J.M. (Online; accessed 2023-03-17) Espadon, easy study of patient dicom data in oncology: toolbox for automation, exploitation and processing of DICOM files and images. https://espadon.cnrs.fr.
00057-1&id=gr1.jpg){kind=link}
00057-1&id=gr2.jpg){kind=link}
00057-1&id=gr3.jpg){kind=link}
00057-1&id=gr4.jpg){kind=link}
00057-1&id=gr5.jpg){kind=link}
00057-1&id=gr6.jpg){kind=link}
00057-1&id=gr7.jpg){kind=link}
00057-1&id=gr8.jpg){kind=link}