Open Implementation of DICOM for Whole-Slide Microscopic Imaging

ebastien Jodogne

Eric Lenaerts

, Lara Marquet

, Charlotte Erpicum

Roland Greimers

, Pierre Gillet

, Roland Hustinx

and Philippe Delvenne

Department of Medical Physics, University Hospital of Li

ege, Avenue de l’H

opital, 4000 Li

ege, Belgium

Department of Pathology, University Hospital of Li

ege, Avenue de l’H

opital, 4000 Li

ege, Belgium

University Hospital of Li

ege, Avenue de l’H

opital, 4000 Li

ege, Belgium

http://www.chuliege.be/

Keywords:

Whole-slide Imaging, Telepathology, DICOM.

Abstract:

This paper introduces an open implementation of DICOM for whole-slide microscopic imaging, following

Supplement 145 of the DICOM standard. The software is divided into two parts: (a) a command-line tool to

convert an whole-slide image to the DICOM format, and (b) a zero-footprint Web interface to display such

DICOM images. The software architecture leverages the DICOM server Orthanc. The entire framework is

available as free and open-source software. The existence of this software supports the development of digital

pathology and telepathology in clinical environments, featuring a smooth integration with existing EHR and

PACS solutions.

1 INTRODUCTION

Anatomopathology plays important medical roles by

detecting preneoplastic lesions and by giving a di-

agnosis, prognosis and evaluation of therapeutic re-

sponse based on visual observations of cellular or

tissue samples. Telepathology is an application of

telemedicine that allows the practice of the anato-

mopathology over a long distance with the use of im-

ages in an electronic format rather than viewing glass

slides (Ling and Krishnappa, 2012). Telepathology is

potentially useful for several applications:

• Intraoperative Consultation: Intraoperative con-

sultations are emergencies in anatomopathology

since the purpose of this examination is to guide

immediate surgical management. For such con-

sultations, the results should be given within 30

minutes (Ribback et al., 2014). Telepathology can

be used in this context for obtaining consultation

by a pathologist present in a remote physical lo-

cation.

• Secondary Consultation from Experts: Secondary

consultation refers to situations where a primary

diagnosis has been performed on the primary ma-

terial and further opinion is needed. Telepathol-

ogy is beneﬁcial by accelerating the process and

by reducing the risks of material loss or break

compared to postal services (Farahani and Pan-

tanowitz, 2016).

• Education and Research: Telepathology is highly

promising in medical education and clinical re-

search. Currently, digital imaging starts to replace

traditional glass slides and high quality micro-

scopes to discuss interesting cases during lessons,

symposia and conferences. Telepathology facili-

tates interactions between multiple users and al-

lows the consistency and longevity of imaged ma-

terials (Mar

ee et al., 2016).

• Pathology Archiving: The storage of histolog-

ical slides is mandatory for a long period of

time. Since the virtual slide perfectly reproduces

the glass slide without any loss of information,

telepathology could enable the storage of images

in an entirely electronic way. As a consequence,

the currently-used, costly physical archives could

become legacy systems in the future (Webster and

Dunstan, 2014).

Telepathology requires microscopic images to be

put quickly and easily online, from a slide scanner

onto a secured Web server. To this end, at least three

technical difﬁculties must be overcome.

Firstly, any pathology laboratory requires high

electronic storage capacities to store their whole-slide

images, as the size of the latter may range from hun-

dreds of megabytes to hundreds of gigabytes, de-

pending on the scanning objective and the tissue sec-

Jodogne S., Lenaerts ÃL’., Marquet L., Erpicum C., Greimers R., Gillet P., Hustinx R. and Delvenne P.

Open Implementation of DICOM for Whole-Slide Microscopic Imaging.

DOI: 10.5220/0006155100810087

In Proceedings of the 12th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2017), pages 81-87

ISBN: 978-989-758-227-1

tion. A typical, uncompressed whole-slide image of

80,000×60,000 pixels acquired in 24-bit color (RGB)

weights over 10GB (DICOM Standards Committee,

Working Group 26, Pathology, 2010). Image com-

pression (JPEG or JPEG 2000) can divide this size by

a factor 10, but this still roughly corresponds to the

size of 3D medical images (e.g. computed tomogra-

phy or magnetic resonance imaging). Storing and in-

dexing such a large amount of data on the long run is

obviously not compatible with the manual administra-

tion of a ﬁlesystem: It implies the use of automated,

scalable, enterprise-ready database systems similar to

the PACS of the hospital (Picture Archiving and Com-

munication System).

The second technical difﬁculty is the lack of inter-

operability between the proprietary ecosystems that

are implemented by the manufacturers of slide scan-

ners. A single hospital might host several scanners

from different manufacturers, making it hard to con-

solidate all the whole-slide images inside a central-

ized tool because of a large variety of ﬁle formats. For

the same reason, it is hard to link the EHR (Electronic

Health Record) of a patient to her anatomopathology

images. This lack of interoperability in the clinical

workﬂow is also an obstacle to the exchange of im-

ages between hospitals, as well as to the use of stan-

dalone post-processing tools that are necessary for

big-data analysis, medical research or education.

A possible solution to this problem consists in

taking advantage of the DICOM standard for medi-

cal imaging (NEMA — National Electrical Manufac-

turers Association, 2016). Indeed, in 2010, the DI-

COM Committee deﬁned a standard way of exchang-

ing whole-slide microscopic imaging that is vendor-

independent and that is fully compatible with exist-

ing PACS systems (DICOM Standards Committee,

Working Group 26, Pathology, 2010). Unfortunately,

there is currently no open implementation of this re-

cent standard, most probably because of the existence

of many patents (Cucoranu et al., 2014): As a con-

sequence, even if any modern PACS can ingest such

DICOM images, few will render them. There is also a

lack of sample ﬁles, which strongly calls for an open

infrastructure to share knowledge about this complex

standard, to the beneﬁt of hospitals, manufacturers

and researchers.

The third technical challenge for telepathology

consists in serving the images over Internet. Be-

cause of the size of whole-slide images, a raw transfer

of the ﬁles would not be compatible with real-time

constraints or mobile applications. It is also highly

desirable to avoid the installation of some heavy-

weight client software on the computers of an hos-

pital, as this operation is often prevented for secu-

rity reasons: A telepathology viewer should there-

fore be able to run entirely inside a standard Web

browser, as such a browser is almost always installed

by default on any computer. A Web application also

has the advantage of being compatible with mobile

devices. The OpenSlide software can already serve

whole-slide images generated by commonplace scan-

ners onto the Web (Goode et al., 2013). Similarly,

the Cytomine software builds upon the OpenSlide ex-

perience to provide an advanced collaborative anal-

ysis framework for research and education in digital

pathology (Mar

ee et al., 2016). Unfortunately, nei-

ther of those platforms currently supports the DICOM

standard that is mandatory for clinical workﬂow.

The previous discussion calls for the engineer-

ing of a framework built upon the DICOM standard

that would support the development of workﬂow for

digital pathology and telepathology in clinical envi-

ronments, featuring a smooth integration with exist-

ing hospital information systems (EHR and PACS).

The needed software would be able to convert whole-

slide images encoded using proprietary formats, to

a standard DICOM ﬁle in accordance with the “VL

Whole Slide Microscopy Image Information Object

Deﬁnition (IOD)”, as speciﬁed in the “PS3.3: Infor-

mation Object Deﬁnitions” part of the DICOM stan-

dard (NEMA — National Electrical Manufacturers

Association, 2016). It would then be able to for-

ward the converted images to a generic PACS sys-

tem, while serving them directly over Internet, using

a zero-footprint Web client.

This paper introduces such an innovative frame-

work and makes it available as free and open-source

software to support the development of digital pathol-

ogy and telepathology in hospitals.

2 SYSTEM AND METHODS

According to the Introduction section, our software

framework for digital pathology and telepathology is

divided into two separate components:

• A standalone command-line tool that takes as in-

put a non-DICOM whole-slide microscopic im-

age, and that generates a compliant DICOM ﬁle.

• A DICOM server that can easily publish such im-

ages on the Web.

This section will fully describe our technical solu-

tion, after an introduction to the DICOM ﬁle format

for whole-slide microscopic imaging.

VISAPP 2017 - International Conference on Computer Vision Theory and Applications

2.1 DICOM for Visible Light

Whole-Slide Microscopic Images

Because images for digital pathology are very large

(possibly dozens of gigabytes), they most often can-

not entirely ﬁt inside the memory of a standard com-

puter. For this reason, whole-slide images are in prac-

tice divided as a regularly-spaced set of tiles, each tile

being a small patch taken from the full image. A tile is

most often a square whose sides contain between 256

and 1024 pixels. Each individual tile can be accessed

separately, allowing a digital pathology application to

bring a tile into memory only when needed.

Compressing some whole-slide image amounts to

compressing each individual tile either using a loss-

less algorithm (JPEG 2000), or a destructive algo-

rithm (JPEG). Furthermore, to enable the navigation

over the entire whole-slide image, the full-sized im-

age is downscaled several times, leading to a pyramid

of images with decreasing spatial resolutions. Each

level of the pyramid is encoded as a separate tiled im-

age. This process is illustrated in Figure 1. Note that

similar encoding schemes are also commonly used to

serve cartography maps on the Web.

The DICOM standard models the real world as

follows: A given patient beneﬁts during her life from

a set of medical imaging studies. Each study is made

of a set of series. Each series is in turn a set of in-

stances, the latter being a synonym for a DICOM ﬁle.

A single DICOM instance can be multi-frame, mean-

ing it can store several independent images (provided

all of its individual frames share the same size). As

a consequence, the whole-slide pyramid represented

in Figure 1 corresponds to one DICOM series, whose

parent study might contain other series acquired dur-

ing the same clinical episode, possibly coming from

other medical imaging modalities. This series is made

of several instances (the DICOM ﬁles), each instance

storing the individual tiles of one given pyramid level

as separate frames. A single instance is not allowed to

store tiles from multiple pyramid levels, yet the same

level can be spread over multiple instances, so as to

prevent the appearance of huge DICOM ﬁles.

Besides its individual frames, each DICOM in-

stance is associated with clinical data under the form

of a recursive key-value associative array. Such keys

are called the DICOM tags and are indexed with two

16-bit hexadecimal numbers. The DICOM standard

lists which DICOM tags are mandatory, conditional

or allowed for whole-slide images in the so-called

“VL Whole Slide Microscopy Image IOD” (NEMA —

National Electrical Manufacturers Association, 2016,

PS3.3, Section A.32.8).

A full enumeration of this set of DICOM tags is

obviously out of the scope of this paper, and the inter-

ested reader is kindly invited to refer to the DICOM

standard. In the context of this paper, it is sufﬁcient

to know that each single instance must specify the

size of the pyramid level it is related to: The DICOM

tag “Total Pixel Matrix Columns” (0x0048,0x0006)

stores the width of the pyramid level, whereas “To-

tal Pixel Matrix Rows” (0x0048,0x0007) stores its

height. These two tags allow to know to which level

of the pyramid a given DICOM instance belongs.

Similarly, because the tiles of one pyramid level

can be shufﬂed over several multi-frame DICOM in-

stances, each frame is associated with the (x, y) po-

sition of the corresponding tile in the corresponding

pyramid level: The tag “Column Position In Total Im-

age Pixel Matrix” (0x0048,0x021e) contains the x-

position of one frame, and “Row Position In Total Im-

age Pixel Matrix” (0x0048,0x021f) its y-position.

This information is collected for each frame of the DI-

COM instance inside the tag “Per Frame Functional

Groups Sequence” (0x5200,0x9230). Our frame-

work almost exclusively relies on this set of tags.

2.2 DICOM-izer

The ﬁrst component of our software framework for

telepathology is the tool that converts a whole-slide

image from a non-DICOM format to DICOM. This

tool will be referred to as the DICOM-izer. It takes

the form of a standalone, cross-platform command-

line tool, so that it can easily be integrated into any

pathology department.

The DICOM-izer features built-in support to read

the most widespread open ﬁle format for whole-slide

imaging (i.e. hierarchical TIFF), as well as common-

place image formats (PNG, JPEG and JPEG 2000). It

is also able to decode proprietary ﬁle formats for slide

scanners (SVS, BIF, VMS. . . ) through the OpenSlide

toolkit (Goode et al., 2013)

. It extracts as much clin-

ical information as possible from the meta-data of the

input image (such as the scanner manufacturer), but

additional information that is not verbatim available

inside the source ﬁle must be provided alongside (e.g.

the identiﬁer of the patient, or the optical parameters

of the acquisition). The output of the DICOM-izer is

a set of compliant DICOM ﬁles that can be sent to any

DICOM modality, including the PACS of the hospital,

thanks to the standard DICOM C-Store command.

If the input image does not contain the full pyra-

mid but only its ﬁnest level, the DICOM-izer can au-

tomatically generate all the upper levels of the pyra-

mid. Similarly, the DICOM-izer can change the com-

Note that OpenSlide does not support the generation of

DICOM ﬁles by itself.

Open Implementation of DICOM for Whole-Slide Microscopic Imaging

Figure 1: Mapping a multi-resolution pyramid according to the DICOM standard (DICOM Standards Committee, Working

Group 26, Pathology, 2010). Each level of the pyramid is a downscaled version of the whole-slide image, and is decomposed

as a set of tiles. The tiles are encoded as separate frames of multi-frame DICOM instances (ﬁles).

pression scheme of the input image to one of the al-

gorithms supported by the DICOM standard (i.e. no

compression, JPEG or JPEG 2000). Unsurprisingly,

decoding a proprietary format, changing the compres-

sion, and/or rebuilding the pyramid are CPU-intensive

re-encoding operations that can last dozens of sec-

onds, even though our DICOM-izer efﬁciently takes

advantage of multi-threading. However, the baseline

process of simply transcoding a hierarchical TIFF to

DICOM only takes a few seconds, with almost no

CPU usage.

Note that microscopy images are often stained or

multi-spectral. This means that multiple channels

might be needed to store the full microscopy image.

The DICOM standard requests to store the various

spatial and spectral channels in separate DICOM se-

ries. In such situations, the DICOM-izer can be sepa-

rately invoked for each channel, which will generate a

set of DICOM series, all belonging to the same parent

DICOM study.

2.3 Orthanc Plugin for Whole-Slide

Imaging

The second component of our software framework

for telepathology allows laboratories to immediately

publish the DICOM images that are produced by the

DICOM-izer over Internet. As argued in the Introduc-

tion, this component is a necessary companion to the

DICOM-izer, as most PACS do not currently support

the rendering of whole-slide images. Because our

framework is designed to be as open as possible to the

beneﬁt of the worldwide community of pathology lab-

oratories and researchers, this component leverages a

free and open-source DICOM server.

The two most well-known free and open-source

DICOM servers are DCM4CHE (Warnock et al.,

2007, written with Java and JBoss) and Orthanc

(Jodogne et al., 2013, written in C++). Besides its

small footprint, Orthanc has the advantage of propos-

ing a RESTful API that makes it ready for Web ap-

plications (Fielding, 2000), and to propose a plugin

mechanism that can be used by third-party developers

to extend the core REST API without using an addi-

tional Web server. As a consequence, our Web pub-

lishing component is built upon the Orthanc vendor

neutral archive, and takes the form of a C++ plugin.

Our Orthanc plugin is a shared library that is dy-

namically loaded by Orthanc during its startup, and

that is responsible for:

1. Transparently indexing all the tiles of a given DI-

COM series as a whole-slide pyramid.

VISAPP 2017 - International Conference on Computer Vision Theory and Applications

Figure 2: Some screenshots of a Web browser displaying real-world pathology images stored inside Orthanc, at various zoom

levels. The Web application is zero-footprint: It is entirely written in JavaScript, and no heavyweight client must be installed

(all is done by the Web browser). The Web interface is built upon OpenLayers version 3, a free and open-source JavaScript

library for displaying raster tile maps (Open Source Geospatial Foundation, 2010). Note that the Web application can be

served through the HTTPS protocol, meaning that the medical communication can be secured through proper authentication

and encryption.

2. Serving the individual tiles according to their

(x, y, z) location (z corresponds to the level of the

tiles in the pyramid), after dynamically transcod-

ing them to an image format that is compatible

with Web browsers (either PNG for uncompressed

or losslessly-compressed whole-slide images, or

JPEG for destructively-compressed images).

3. Publishing all the HTML and JavaScript static re-

sources that are necessary for the Web viewer.

The indexing process of Step (1) ﬁrst groups the

instances according to the size of their associated

pyramid level z, then extracts the (x, y) position of

each frame in these instances, only by considering

the DICOM tags that were introduced at the end of

Section 2.1. Figure 2 shows screenshots of a Web

browser accessing our viewer of DICOM whole-slide

images

The DICOM-izer can be conﬁgured to automat-

ically push its output images to any Orthanc server

through its RESTful API. Note also that because

Orthanc is a fully-featured vendor neutral archive, it

can be used to re-transmit (resp. query/retrieve) the

whole-slide images to (resp. from) other DICOM

modalities, including the PACS of the hospital. Fi-

nally, an ofﬁcial plugin is available to make Orthanc

use a PostgreSQL database, making it fully scalable

and enterprise-ready if need be.

The demonstration server from which these

screenshots were taken is publicly available online at:

http://wsi.orthanc-server.com/demo/. This demonstration

server illustrates the fact that the Web interface is entirely

zero-footprint: Any modern Web browser will display

it out-of-box, without having to install any additional

software.

Open Implementation of DICOM for Whole-Slide Microscopic Imaging

Figure 3: Possible clinical workﬂow for digital pathology. The innovative contributions introduced by this paper are high-

lighted in red.

3 APPLICATIONS

Figure 3 illustrates a real-world clinical workﬂow that

becomes possible thanks to our open framework for

digital pathology and telepathology. In this work-

ﬂow, the pathology department sends its images to the

PACS of the hospital, which enables the continuous

integration with the electronic health record (EHR) of

the patient: This would solve the “pathology archiv-

ing” objective explained in the Introduction.

In parallel, the DICOM-izer also sends its out-

put images to an Orthanc server equipped with our

plugin. As soon as the DICOM images are received

by Orthanc, they are immediately made available on-

line for real-time viewing: Opening an image in the

browser takes less than one second, and the user ex-

perience is similar to well-known Web mapping sys-

tems. This allows telepathologists to quickly and eas-

ily review whole-slide images remotely, using any

standard Web browser. This remote, read-only access

is also possible using a smartphone or a tablet.

The CPU power that is required by the Orthanc

server is very low if the images are encoded using

JPEG, which is the most common case: As Web

browsers natively support JPEG, the plugin can serve

the compressed tiles without decoding them. Fur-

thermore, as the telepathology plugin is entirely Web-

based, the network administrators of the hospital can

setup a reverse proxying system and HTTPS encryp-

tion to ensure the proper authentication and conﬁ-

dentiality for the access to the medical information:

This would solve the “intraoperative consultation”

and “secondary consultation” objectives. Also note

that Orthanc can query/retrieve images archived in the

PACS so as to put them back online.

Finally, whole-slide images that are found to be

of interest for research or education, can be exported

as hierarchical TIFF directly from our Web viewer to

a richer Internet application for collaborative analysis

such as Cytomine (Mar

ee et al., 2016). This would

solve the “education and research” objective.

4 CONCLUSION

This paper introduces an implementation of the “VL

Whole Slide Microscopy Image IOD”, as speciﬁed in

the “PS3.3: Information Object Deﬁnitions” part of

the DICOM standard (NEMA — National Electrical

Manufacturers Association, 2016), in order to support

the development of digital pathology and telepathol-

ogy in clinical laboratories and hospitals. The source

code of our framework is provided as free and open-

source software, under the terms of the AGPL li-

cense

. It is notably compatible with Microsoft Win-

dows, Apple OS X and GNU/Linux environments.

To the best of our knowledge, this is ﬁrst pub-

lic, open, reference implementation of DICOM for

whole-slide microscopic imaging. Our framework

consists of a standalone DICOM-izer, that converts

The source code is accessible from the ofﬁcial home-

page of Orthanc: http://wsi.orthanc-server.com/. Full tech-

nical documentation of the framework and of the un-

derlying open Web API is part of the Orthanc Book:

http://book.orthanc-server.com/.

VISAPP 2017 - International Conference on Computer Vision Theory and Applications

whole-slide images to compliant DICOM ﬁles, to-

gether with a dedicated plugin extending the vendor

neutral archive Orthanc. The latter plugin extends Or-

thanc with a lightweight, zero-footprint Web viewer

of whole-slide images. This viewer can be accessed

remotely from any Web browser or mobile device.

It has also been discussed how our framework for

telepathology can be integrated inside a typical hos-

pital workﬂow to meet real-world challenging objec-

tives such as remote consultation, education, clini-

cal research, or long-term archiving. Future work

will consist in taking advantage of our framework

to support a multi-centric clinical study that will de-

velop and assess new algorithms to quantify relevant

biomarkers on digitized immunostained slides of neo-

plastic lesions.

REFERENCES

Cucoranu, I. C., Parwani, A. V., Vepa, S., Weinstein, R. S.,

and Pantanowitz, L. (2014). Digital pathology: A sys-

tematic evaluation of the patent landscape. Journal of

Pathology Informatics, 5(1):16.

DICOM Standards Committee, Working Group 26, Pathol-

ogy (2010). Supplement 145: Whole slide micro-

scopic image IOD and SOP classes.

Farahani, N. and Pantanowitz, L. (2016). Overview of tele-

pathology. Clinics in Laboratory Medicine, 36(1):101

– 112.

Fielding, R. (2000). Architectural Styles and the Design

of Network-based Software Architectures. PhD thesis,

University of California, Irvine.

Goode, A., Gilbert, B., Harkes, J., Jukic, D., and Satya-

narayanan, M. (2013). OpenSlide: A vendor-neutral

software foundation for digital pathology. Journal of

Pathology Informatics, 4(1):27. Freely available at:

http://openslide.org/.

Jodogne, S., Bernard, C., Devillers, M., Lenaerts, E., and

Coucke, P. (2013). Orthanc – Lightweight, RESTful

DICOM server for healthcare and medical research.

In IEEE 10th International Symposium on Biomedical

Imaging (ISBI), pages 190–193, San Francisco, USA.

Freely available at: http://www.orthanc-server.com/.

Ling, C. and Krishnappa, P. (2012). Telepathology – An up-

date. International Journal of Collaborative Research

on Internal Medicine and Public Health, 4(12).

Mar

ee, R., Rollus, L., St

evens, B., Hoyoux, R., Louppe,

G., Vandaele, R., Begon, J.-M., Kainz, P., Geurts, P.,

and Wehenkel, L. (2016). Collaborative analysis of

multi-gigapixel imaging data using cytomine. Bioin-

formatics. Freely available at: http://cytomine.be/.

NEMA — National Electrical Manufacturers Association

(2016). NEMA PS3 / ISO 12052, Digital Imaging

and Communications in Medicine (DICOM) standard.

Freely available at: http://medical.nema.org/.

Open Source Geospatial Foundation (2010). Openlay-

ers: Free maps for the web. Freely available at:

http://www.openlayers.org/.

Ribback, S., Flessa, S., Gromoll-Bergmann, K., Evert, M.,

and Dombrowski, F. (2014). Virtual slide telepathol-

ogy with scanner systems for intraoperative frozen-

section consultation. Pathology - Research and Prac-

tice, 210(6):377 – 382.

Warnock, M. J., Toland, C., Evans, D., Wallace, B., and

Nagy, P. (2007). Beneﬁts of using the DCM4CHE DI-

COM archive. Journal of Digital Imaging, 20(Supp.

1):125–129.

Webster, J. D. and Dunstan, R. W. (2014). Whole-slide

imaging and automated image analysis: Considera-

tions and opportunities in the practice of pathology.

Veterinary Pathology, 51(1):211–223.

Open Implementation of DICOM for Whole-Slide Microscopic Imaging