Modeling a Smart Teleradiology:
Decision Support System based on Ontology
EUSTACHE MUTEBA A.1, PATRICK ANELIA L.2
1Correspondent of International Medical Informatics Association,
P.O. Box 14769, Kinshasa,
DR CONGO
2Interdisciplinary Center for Research in Medical Imaging,
Department of Radiology, University Clinics of Kinshasa,
DR CONGO
Abstract: - Increasingly, hospitals are producing information related to additional examinations for reasons of
in-depth investigations or diagnoses. Medical imaging plays an essential role in medical action, mainly for
diagnosis, therapeutic planning, intraoperative navigation, postoperative monitoring, and biomedical research.
From the perspective of Universal Health Coverage, teleradiology is one of the solutions to the lack of
radiologist practitioners in certain territories. Given the situation of the health system in developing countries
and in particular in DR Congo, we therefore aim to contribute by providing a solution under a project related to
teleradiology. The system designed to make a link between clinical information, data extracted from images,
and the radiological ontology for decision-making based on semi-supervised machine learning. This article
presents the theoretical foundations of the study and highlights the implementation of our radiology ontology
called Smart Ontology of Radiology (SORad).
Key-Words: - Teleradiology, Image analysis, Ontology, Semi-supervised machine learning, Intelligent system,
Decision support.
Received: August 15, 2023. Revised: November 9, 2023. Accepted: December 5, 2023. Published: December 31, 2023.
1 Introduction
Increasingly, hospitals are producing information
related to additional examinations for reasons of in-
depth investigations or diagnoses.
Medical images are an integral part of healthcare
data that are acquired mainly for diagnosis, therapy
planning, intraoperative navigation, post-operative
monitoring, and biomedical research, [1].
It should be noted that the global teleradiology
market size was valued at USD 2.44 billion in 2022
and is anticipated to grow at a compound annual
growth rate (CAGR) of 12.9% from 2023 to 2030.
The growth is majorly driven by the increasing
prevalence of target diseases, and rising demand for
teleradiology for second opinions and emergencies,
[2].
As stated in [3], the ability to exploit data, in
particular imaging, is at the heart of the challenges
of tomorrow's medicine.
However, given the situation of the health system
in developing countries and in particular in DR
Congo, we noted that there are the following
recurring priority problems: (a) low coverage (b)
weak operational capacity of structures at all levels
to carry out interventions, (c) poor quality of care
and services offered, (d) low use of available care
and services, and (e) weak public accountability of
health services.
From the perspective of Universal Health
Coverage, teleradiology is one of the solutions to
the lack of radiologist practitioners in certain
territories.
We therefore aim to contribute by providing a
solution under a project related to teleradiology. The
project involves the radiology department of the
University Clinics of Kinshasa (CUK). The choice
of the CUK is justified by its position on the
national level as a major center for medical research
application given the number of experts.
Our system is called Smart Teleradiology. The
system designed to make a link between clinical
information, data extracted from images, and the
radiological ontology for decision-making based on
semi-supervised machine learning.
To enable optimum and specialized patient care
through timely interventions by expert radiologists
WSEAS TRANSACTIONS on COMPUTERS
DOI: 10.37394/23205.2023.22.37
Eustache Muteba A., Patrick Anelia L.
E-ISSN: 2224-2872
324
Volume 22, 2023
in any location at any time of day, our system is
built in a cloud.
This article presents the theoretical foundations
of the study and highlights the implementation of
our radiology ontology called Smart Ontology of
Radiology (SORad).
2 Methods
The methodology adopted to develop our system
involves classical analysis of the different concepts
involved. This is a classic way that gives the limits
of our study.
2.1 Radiology Information System (RIS)
A Hospital Information System (HIS) is a federation
of functionally distinct but not disjointed
subsystems, within and between which flows of
information circulate. The radiology information
system is one of the subsystems.
As clearly stated in [4], radiological center
within the hospital information system (HIS)
requires a special and its information system,
radiology information system (RIS).
Ultimately, a radiology information system is a
subsystem that manages medical imagery and
associated data.
2.1.1 Different Functionalities of RIS
The RIS is an element of the HIS that is dedicated to
interacting with the medical record.
Within RIS there is the PACS (picture archiving
and communication systems) as an important
component.
PACS is a computerized means of replacing the
roles of the conventional radiological film: images
are acquired, stored, transmitted, and displayed
digitally, [5].
In general, the RIS provides the following
functions: radiology examination order form,
modality interface and image acquisition, processing
and restitution of images, digital reporting and result
transmission, integration with the HIS medical
record, archiving in PACS, and Images tracking.
2.1.2 Nature and Type of Radiology Data
Routine clinical visits of a single patient might
produce digital data in multiple modalities,
including image data (i.e. pathology images,
radiology images, and camera images) and non-
image data (i.e. lab test results and clinical data).
The heterogeneous data would provide different
views of the same patient to better support various
clinical decisions (e.g. disease diagnosis and
prognosis, [6].
2.2 Imaging Modalities
It should be noted that in radiology, a medical image
is a materialization in the form of images of
anatomical or functional information in vivo of parts
(organs, tissues, cells) of the human body, as well as
the data extracted or derived from these images.
To obtain image data, there are both hardware
and software processing operations.
On the hardware side: it is the image acquisition
process by an image constructor device. An image is
the optical representation of an object illuminated
by a radiation source. The following elements are
present in an image formation process: an object, a
radiation source (visible light, X-rays, electrons,
etc.), and an image formation system, [7].
On the software side: it is the image processing.
Image processing may include the following steps:
Image import via image acquisition tools, image
manipulation, and analysis.
Table 1. Types and Mechanisms of Different
Imaging Modalities
Modalities
Mechanisms
Types
X-ray
X-ray
absorption
Morphological
Digital
angiography
X-ray
absorption
Morphological
Rotational
angiography
X-ray
absorption
Morphological
CT (scanner)
X-ray
absorption
Morphological
Ultrasound
Ultrasound
reflection
Morphological
Ultrasound 3D
Ultrasound
reflection
Morphological
Scintigraphy
Emission of
gamma
photons
Functional
Tomoscintigraphy
Emission of
gamma
photons
Functional
PET
Émission de
positons
Functional
MRI
Echoes of the
magnetization
of the nuclei
Mixed
Imaging modalities relate to image acquisition
techniques using imaging equipment. As stated in
[8], imaging modalities are often categorized by the
method in which images are generated normally by
physical phenomena. Usually, the physical
phenomena, on which all the stages of image
production can be: X-rays (Radiology), gamma rays
(Nuclear Medicine), magnetic waves (Magnetic
Resonance Imaging), ultrasonic waves (ultrasound),
and optics (endoscopy).
WSEAS TRANSACTIONS on COMPUTERS
DOI: 10.37394/23205.2023.22.37
Eustache Muteba A., Patrick Anelia L.
E-ISSN: 2224-2872
325
Volume 22, 2023
Modern radiological imaging uses a standard
called Digital Imaging and Communications in
Medicine (DICOM), [9]. The DICOM incorporates
standards for imaging modalities as described in
Table 1.
2.3 Ontology
Defining an ontology for knowledge representation
means defining, for a given domain and problem,
the functional and relational signature of a formal
language of representation and the associated
semantics, [10].
A formal system represents a specific domain of
knowledge using basic elements, the concepts, that
are defined and organized each one in relation to the
others, [11].
In practice, the modeling of knowledge by an
ontology will have to take the following elements:
semantics (knowing how to understand each other),
syntactics (knowing how to communicate), and
technique (being able to communicate).
Assuming that the physician has a cognitive
causal model for performing a medical act
(investigating, diagnosing, and prescribing) of a
patient. This causal model, incorporating the
expert’s knowledge of anatomy and physiology, can
be used to simulate the normal working of the body,
its pathological behavior in a diseased state, and the
idiosyncrasies that characterize a particular patient.
Therefore, we can imagine that this causal model
influences the creation of the classes
of a medical ontology.
In practical terms, developing an ontology includes:
- defining classes in the ontology,
- arranging the classes in a taxonomic (subclass–
superclass) hierarchy,
- defining slots and describing allowed values for
these slots,
- filling in the values for slots for instances.
In radiology, there are ontologies which of
course have different specificities, to date, the
radiology lexicon, Radiological Society of North
America's radiology lexicon (Radlex) ontology [12],
[13], seems the most widespread.
In any case, the integrated Radiology Gamuts
Ontology (RGO) system is interesting. The RGO is
an ontology that links diseases and imaging findings
to support differential diagnosis in radiology, to
terms in three key vocabularies for clinical
radiology: the International Classification of
Diseases, version 10, Clinical Modification (ICD-
10-CM), the Radiological Society of North
America's radiology lexicon (RadLex), and the
Systematized Nomenclature of Medicine Clinical
Terms (SNOMED CT), [14].
2.4 Radiology Decision Support
In medical practice, the clinical decision support
systems (CDSSs) rely on formalized knowledge
bases and are usually integrated into electronic
patient records to assist the clinician in her everyday
practice, [15].
The radiology decision support takes its source
in the modalities of the images according to the
specific problem of the patient as presented in the
medical record
Among the most concerned activities in
radiology is the interpretation of images for
diagnosis.
Technically, the interpretation of medical images
requires the modeling of spatial relationships and
the view of the appearances of objects. The
modeling of spatial relationships uses segmentation
and recognition.
The use of computer-assisted decision support is
not new in radiology. As mentioned in [16],
radiology requires diagnosis and decision-making
under uncertainty and AI may help automate some
of the labour-intensive tasks such as radiograph
interpretation and reporting.
It is stated that machine learning, the cornerstone
of today’s artificial intelligence (AI) revolution,
brings new promises to clinical practice with
medical images, [17]. In accordance with [18],
initially, machine learning typically begins with the
machine learning algorithm system computing the
image features that are believed to be of importance
in making the prediction or diagnosis of interest.
Moreover, it is interesting to know that recent
advances in machine learning have the potential to
recognize and classify complex patterns from
different radiological imaging modalities such as x-
rays, computed tomography, magnetic resonance
imaging, and positron emission tomography
imaging, [19].
In the previous section, it was discussed and
demonstrated that ontology can play an important
role in decision support regarding the protocol of
imaging findings.
2.5 Teleradiology
Teleradiology is the remote practice of radiological
medicine. It is one of the telemedicine applications
that benefit from the longest clinical experience and
the greatest technological maturity. Teleradiology
mainly includes telediagnosis and teleexpertise.
Telediagnosis consists, for the radiologist, of
organizing remotely and under his control the
performance, by a manipulator, of an imaging
examination then interpreting it and reporting its
result, in the most similar way possible to what
WSEAS TRANSACTIONS on COMPUTERS
DOI: 10.37394/23205.2023.22.37
Eustache Muteba A., Patrick Anelia L.
E-ISSN: 2224-2872
326
Volume 22, 2023
would have been done in an office or an
establishment.
Teleexpertise consists of allowing a healthcare
professional to remotely seek the opinion of one or
more healthcare professionals because of their
training or their particular skills for the care of a
patient.
Teleradiology can help increase imaging
efficiency and mitigate both geographic and
temporal discrepancies in imaging care, [20].
3 Results
Let us take advantage of the different techniques
previously described to develop a smart
teleradiology system. It is a very complex system as
you can see in its architecture. However, we will
present to you the essentials, namely the models of
our: machine learning, ontology, and decision
support. For the sake of pragmatism, we carry out
simulations.
3.1 Smart Teleradiology System Architecture
The system architecture, as presented in Figure 1, is
structured around the following components: User
Interface, PACS, Semi-supervised Machine
Learning, Ontology, and Decision Support.
Fig. 1: Diagram of the Smart Teleradiology System
3.2 Semi-Supervised Machine Learning
The idea behind Semi-Supervised Machine
Learning, [21], stands somewhere between
supervised learning (in which all training examples
are labeled) and unsupervised learning (in which no
label data are given).
The SSML-based problem-solving approach
caught our attention for its realistic aspect. In our
previous work [22], we implemented and
experimented with this method to perform the
automatic diagnosis of malaria. However, this was
limited to microscopic images. And so for this
current study which relates to macroscopic images,
we have made modifications especially in terms of
the image analysis algorithms.
A succinct description of the framework of
SSML is shown in Figure 2 and described below:
- HumanExpert.class: allows the user interface.
- Manager.class: provides activities of managing
and controlling all components of SSML.
- Contractor.class: supports and allows grants to
the manager.class to perform activities. This ensures
and meets SSML requirements in terms of
flexibility and efficiency under a contract
specification, [23].
- PACS.class: provides activities to manage file
acquisition.
- Uploader.class: provides activities to manage
uploads or acquisition of files.
- Viewer.class: provides activities to display
files.
- ImageAnalyzer.class: allows the extraction of
significant information and manages image analysis.
- Extractor.class: quantifies the significant
characteristics of objects in the image and selects
regions of interest containing relevant information.
- PatternBuilder.class: provides activities to build
pattern from a selected image region. It is the pre-
classification process.
- Classifier.class: provides label assignment and
classification activities based on a prediction model.
- Profiler.class: provides clustering activities based
on Classifier results.
Fig. 2: Framework of Semi-Supervised Machine
Learning (SSML)
3.3 Smart Ontology of Radiology
The developed ontology is called “Smart Ontology
of Radiology (SORad)”. SORad is based on a
WSEAS TRANSACTIONS on COMPUTERS
DOI: 10.37394/23205.2023.22.37
Eustache Muteba A., Patrick Anelia L.
E-ISSN: 2224-2872
327
Volume 22, 2023
multiaxial classification and it is built based on
aspects related to topography, morphology, etiology,
functional, treatment, and procedure.
We present in the following six figures, namely
Figure 3, Figure 4, Figure 5, Figure 6, Figure 7 and
Figure 8, that show different views of our first
version of SORad, It is structured around 5 main
concepts: Diagnosis, Equipment, Human, Imaging,
and Procedure classes.
Fig. 3: SORad abstract classes
Fig. 4: Diagnosis abstract class
Fig. 5: Human class
Fig. 6: Procedure class
Fig. 7: Equipment class
WSEAS TRANSACTIONS on COMPUTERS
DOI: 10.37394/23205.2023.22.37
Eustache Muteba A., Patrick Anelia L.
E-ISSN: 2224-2872
328
Volume 22, 2023
Fig. 8: Imaging class
3.4 Smart Decision Support
When we interpret the SSML, there is a straight line
in the center that goes from the ImageAnalyzer class
to the Classifier class. The Profiler class is the
bridge between the Single Block and the Decision
Support class. Indeed, once the machine provides a
classification, the DecisionSupport class will have
to select the results obtained based on the calculated
and/or predefined rule-criteria.
Ultimately, radiology relies on exchanges
between medical personnel and uses language as a
vector of communication through requests,
protocols, and reports. Considering that medical
imaging is by definition linked to the image and that
this is in its production as well as in its
interpretation. The radiologist interprets and makes
the image speak using words. There is a clear need
for knowledge management to facilitate common
understanding among radiologists.
The SORad ontology can act in carrying out the
protocol to give meaning and support the diagnosis.
At this level, a special component of Natural
Language Processing (NLP) intervenes. This is the
Automatic Radiology Protocol Generator (ARPG).
3.5 Simulation
The simulation focuses on the initial results of the
SORad ontology component. The screenshots in
Figure 9, Figure 10 and Figure 11, show the user
interface and query/search terms in SORad.
4 Discussion and Conclusion
Presenting a model and methodologies to develop a
smart teleradiology system is the aim of this article.
Fig. 9: Screenshot of the term related to human
class
Fig. 10: Screenshot of the term related to
equipment’s class
Fig. 11: Screenshot of the term related to image’s
class
The architecture of the system shows that there
are four integrated components, namely: Radiology
Information System, Semi-supervised Machine
Learning, Ontology and Decision Support. Each of
them plays a role.
In itself, the smart teleradiology system is an
essential solution for access to health care and to
enable optimum and specialized patient care through
timely interventions by expert radiologists in any
location at any time of day, particularly, in
developing countries.
WSEAS TRANSACTIONS on COMPUTERS
DOI: 10.37394/23205.2023.22.37
Eustache Muteba A., Patrick Anelia L.
E-ISSN: 2224-2872
329
Volume 22, 2023
Upstream, the radiology information system
intervenes as an interface with the cloud. It manages
the patient's clinical information, acquisition,
transmission, and traceability of images.
The core of the system is semi-supervised
machine learning. This choice is explained by the
fact that an SSML is controllable, especially given
the delicacy of radiological imaging.
It turns out that a decision support system
supported by an SSML interacting with a dedicated
ontology allows causal reasoning which gives the
ability to explain complex logical interconnections
between investigation, diagnosis, and prescription
about a given case.
Furthermore, for pragmatic reasons, a simulation
is performed and presents the SORad ontology.
This initial implementation was carried out on
our Virtual Community of Healthcare Facilities
development platform, [24]. Note that the project is
underway for a prototype. And as we go along, we
bring innovations. It’s like leveraging Radiology
Gamuts Ontology (RGO) to enrich our system.
Certain aspects of our study will be the subject of
other future articles.
References:
[1] Preim B., Botha C., Acquisition of Medical
Image Data, in Visual Computing for
Medicine (Second Edition), 2014.
[2] Teleradiology Market Size, Share & Trends:
Analysis Report 2023-2030, Grand View
Research, 2023, [Online].
https://www.grandviewresearch.com/industry-
analysis/teleradiology-market (Accessed Date:
February 1, 2024).
[3] Seymour K., Benyahia N., Hérent P.,
Malhaire C., Exploitation of data for research
and artificial intelligence: medical, ethical,
legal, technical issues, Women’s Imaging
(Exploitation des données pour la recherche
et l’intelligence artificielle : enjeux médicaux,
éthiques, juridiques, techniques, Imagerie de
la Femme), 2019,
https://doi.org/10.1016/j.femme.2019.04.004.
[4] Rade B., Zoran M., Đinđić B., Gordana B.,
Radiology Information System, ACTA Medica
Medianae, Vol. 51, 2012, pp.39-46.
[5] Strickland N.H., PACS (picture archiving and
communication systems): filmless radiology.
Arch Dis Child, Vol. 83, No.1, 2000, pp. 82-6.
[6] Cui C., Yang H., Wang Y., Zhao S., Asad Z.,
Coburn LA., Wilson KT., Landman BA., Huo
Y., Deep multimodal fusion of image and
non-image data in disease diagnosis and
prognosis: a review, Prog. Biomed. Eng.
(Bristol), Vol. 5, No 2, 2023.
[7] Kindratenko V., Development and
Application of Image Analysis Techniques for
Identification and Classification of
Microscopic Particles, Ph.D. Thesis,
University of Antwerpen, 1997, [Online].
https://users.ncsa.illinois.edu/kindr/phd/index.
pdf (Accessed Date: February 1, 2024).
[8] Jacques S., Christe b., Chapter 2 - Healthcare
technology basics, Introduction to Clinical
Engineering, Academic Press, 2020, pp. 21-
50.
[9] Gibaud, B., The DICOM standard: a brief
overview, NATO Security through Science
Series B: Physics and Biophysics, 2008.
[10] Baneyx A., Charlet J., Jaulent M.C.,
Methodology to Build Medical Ontology from
Textual Resources, AMIA Annu Symp Proc.
2006, 2006, pp.21-25.
[11] Lasbleiz J., Use of ontologies in the context of
Magnetic Resonance Imaging. Thesis,
Biomedical Engineering, Rennes 1 University
(Utilisation des ontologies dans le contexte de
l’Imagerie par Résonance Magnétique. Thesis,
Ingénierie biomédicale, Université Rennes 1),
2013, [Online]. https://theses.hal.science/tel-
00950357 (Accessed Date: February 1, 2024).
[12] RadLex, RadLex radiology lexicon. RSNA,
2020, [Online]. https://radlex.org/ (Accessed
Date: February 1, 2024).
[13] Chepelev L.L, Kwan D, Kahn C.E., Filice
R.W., Wang K.C., Ontologies in the New
Computational Age of Radiology: RadLex for
Semantics and Interoperability in Imaging
Workflows, RadioGraphics, Vol.43-3, 2023.
[14] Filice R.W., Kahn C.E., Integrating an
Ontology of Radiology Differential Diagnosis
with ICD-10-CM, RadLex, and SNOMED
CT, J Digit Imaging, Vol. 32, No 2, 2019,
pp.206-210.
[15] Galopin A., Bouaud J., Pereira S., Seroussi B.,
An Ontology-Based Clinical Decision Support
System for the Management of Patients with
Multiple Chronic Disorders, Stud Health
Technol Inform, Vol. 216, 2015, pp. 275-9.
[16] Farič N., Hinder S., Williams R., Ramaesh R.,
Bernabeu M., Van Beek E., Cresswell K.,
Early experiences of integrating an artificial
intelligence-based diagnostic decision support
system into radiology settings: a qualitative
study, Journal of the American Medical
Informatics Association, Vol. 31, Issue 1,
2023, pp. 24–34.
WSEAS TRANSACTIONS on COMPUTERS
DOI: 10.37394/23205.2023.22.37
Eustache Muteba A., Patrick Anelia L.
E-ISSN: 2224-2872
330
Volume 22, 2023
[17] Varoquaux G., Cheplygina V., Machine
learning for medical imaging: methodological
failures and recommendations for the future,
npj Digit. Med., Vol. 5, No 48, 2022.
[18] Erickson B. J., Korfiatis P., Akkus Z., Kline
T. L., Machine Learning for Medical Imaging,
RadioGraphics, Vol. 37, No 2, 2017.
[19] Zhang Z., Sejdić E., Radiological images and
machine learning: Trends, perspectives, and
prospects, Computers in Biology and
Medicine, Vol. 108, 2019, pp. 354-370.
[20] Hanna T.N., Steenburg S.D., Rosenkrantz
A.B., Pyatt R.S., Duszak R., Friedberg E.B.,
Emerging Challenges and Opportunities in the
Evolution of Teleradiology, American Journal
of Roentgenology, Vol. 215, Issue 6, 2020, pp.
1411-1416.
[21] Chapelle O., Schölkopf B., Zien A., Semi-
Supervised Learning, Cambridge,
Massachusetts: Massachusetts Institute of
Technology Press, 2006, [Online].
https://www.molgen.mpg.de/3659531/MITPre
ss--SemiSupervised-Learning.pdf (Accessed
Date: February 1, 2024).
[22] Muteba A.E., Modeling an Automatic
Diagnosis of Malaria based on Digital
Holographic Microscopy, Congress, poster,
3rd ICEHTMC Proceedings, Vol. 2, 2019.
[23] Muteba A.E., Modeling Software Agents:
Web-Based Decision Support System for
Malaria Diagnosis and Therapy, Journal of
Health Informatics in Africa, Vol. 3, No 1,
2015.
[24] Muteba A.E., Radiology Ontology (Smart
Ontologie de Radiologie), (SORad), Virtual
Community of Healthcare Facilities (VCHF),
2023, [Online].
https://www.vchf.net/clorad/sorad/index.php
(Accessed Date: February 16, 2024).
Contribution of Individual Authors to the
Creation of a Scientific Article
- Patrick Anelia posed the problem and provided
useful specifications relating to the field of study:
"radiological imaging".
- Eustache Muteba carried out all theoretical and
technical aspects, and also wrote the paper.
Sources of Funding for Research Presented in a
Scientific Article or Scientific Article Itself
No funding was received for conducting this study.
Conflict of Interest
The authors have no conflicts of interest to declare.
Creative Commons Attribution License 4.0
(Attribution 4.0 International, CC BY 4.0)
This article is published under the terms of the
Creative Commons Attribution License 4.0
https://creativecommons.org/licenses/by/4.0/deed.en
_US
WSEAS TRANSACTIONS on COMPUTERS
DOI: 10.37394/23205.2023.22.37
Eustache Muteba A., Patrick Anelia L.
E-ISSN: 2224-2872
331
Volume 22, 2023