Keywords economics - teleradiology - cost-effectiveness - acceptance testing - diagnostic radiology
- pediatric radiology
Introduction
Pediatric and adolescent radiology has made a major contribution to medical successes
in the fight against diseases of childhood and adolescence. Its continued development
depends largely on insights from clinical specialties and on technical innovations
from other imaging fields [1 ]
[2 ]. There is no broad pool of doctors specializing in pediatric and adolescent radiology
in German-speaking countries. The strategy paper of the Society for German-speaking
Pediatric Radiologists (GPR) calls for providing adequately staffed regional referral
centers (with both medical and non-medical personnel) to focus on pediatric and adolescent
radiology; these centers will cooperate closely with the referring physicians from
conservative and surgical pediatric and adolescent medicine, as well as outpatient
or inpatient radiology [1 ]. According to the GPR strategy paper, technologies such as artificial intelligence
(AI) and teleradiology could open new opportunities for more efficient and improved
radiology care for children in Germany, Austria, and Switzerland (i.e. the DACH region)
[1 ].
The role of AI in medical imaging is growing steadily [3 ]. The pediatric radiology workflow holds many points of reference at which AI applications
could potentially be implemented. These can range from patient allocation and appointment
scheduling to dose and artifact reduction, as well as image interpretation, e.g. using
computer-assisted diagnosis (CAD) systems [4 ]
[5 ]
[6 ]. Virtual grids can be used for X-rays in intensive care units to reduce radiation
exposure and improve image quality [7 ]. With AI, image acquisition in MRI or CT examinations can be improved through faster
MRI sequences [8 ] or dose reduction can be optimized in CT examinations [9 ]
[10 ]. In post-processing, AI can handle, in addition to image optimization, more time-consuming
activities such as segmentation [6 ]. Of the more than 200 approved AI products for radiology, individual applications
have been developed specifically for pediatrics [3 ]
[11 ]. One prime example of the application of computer-assisted diagnostics is the determination
of bone age based on hand X-rays. As a result, in 2019 the Radiological Society of
North America (RSNA) announced a competition to develop a corresponding AI solution
[12 ], which contributed to the development of products such as BoneXpert, VUNO Med-BoneAge,
or IB Lab PANDA [11 ]
[13 ]
[14 ].
Due to the shortage of personnel in pediatric radiology, commercial teleradiology
concepts were developed early on in the United States [15 ]. By pooling expertise in larger pediatric radiology centers, the workload in departments
without pediatric and adolescent radiologists can be reduced, because medical staff
with a specialty in pediatric radiology no longer has to be available continuously
[16 ]. In order to cope with the increased workload in pediatric radiology centers, it
is essential to have adequate staffing. The GPR strategy paper recommends a similar
cooperation between centers and referring physicians [1 ]. In addition, a second diagnosis by specialized pediatric radiologists can lead
to substantial and therapy-relevant changes in the findings or assessment [17 ].
In the following, we present the results of a survey regarding the use of AI and teleradiology
in pediatric radiology care in Germany, Austria, and Switzerland. The survey of pediatric
radiologists aimed to evaluate the attitudes and use of technologies recommended in
the GPR strategy paper in the field of pediatric radiology – also with regard to a
potential solution to the shortage of pediatric and adolescent radiologists in Germany
and Austria [1 ]. Additionally, in contrast to previous studies, we intend to pay special attention
to the needs of pediatric radiology.
Materials & Methods
The survey was conducted in German on the SurveyMonkey platform (SurveyMonkey Inc.,
San Mateo, CA, USA). The survey was sent by the GPR and the SGPR to pediatric and
adolescent radiologists in Germany, Austria, and Switzerland, and it was open from
March to May 2023. The survey was advertised on the homepages of the GPR and the SGPR,
as well as in the newsletters of both professional associations. In addition to the
5-point Likert scales (“fully agree”, “partly agree”, “neither agree nor disagree”,
“partly disagree”, “fully disagree”), we included multiple choice questions with the
option for multiple answers as well as free-text responses. The answers were voluntary
and anonymous. Information about the place of work was not requested. Each individual
question could be skipped. This manuscript was shortened editorially using ChatGPT
(OpenAI, San Francisco, USA).
Results
Demographics
We received 36 responses from pediatric radiology departments in the DACH region,
which corresponds to a response rate of 8.6% for 350 GPR and 69 SGPR members (excluding
emeritus/retiree members in both organizations). Additional demographic information
is included in [Table 1 ]. The distribution in the table roughly corresponds to the distribution of characteristics
in the GPR.
Table 1 Demographics.
Parameters
Quantity (%)
Level of education
Doctor in training
1 (2.8%)
Specialist doctor
2 (5.6%)
Specialist doctor + specialty designation
33 (91.7%)
Age
< 30
0 (0%)
31–40
8 (22.2%)
41–50
8 (22.2%)
51–60
11 (30.6%)
>60
9 (25.0%)
Type of department
University hospital
27 (75.0%)
Non-university hospital
7 (19.4%)
Doctor’s office/practice
2 (5.6%)
Country
Germany
26 (72.2%)
Austria
5 (13.9%)
Switzerland
5 (13.9%)
Teleradiology
77.8% of respondents stated that teleradiology is an important topic for pediatric
radiology (50% “fully agree”; 27.8% “partly agree”), but it only plays a major role
in the daily work of half of the respondents (25% “fully agree”; 22.2% “partly agree”)
([Fig. 1 ]
a ). For further differentiation, we also asked about the use of teleradiology consultations/second
opinions as well as teleradiology in line with radiation protection regulations (Section
14 of the Radiation Protection Act, StrSchG, Germany, or Section 32 of the Medical
Radiation Protection Ordinance, MedStrSchV, Austria). Teleradiology second opinions
are frequently carried out in half of the departments (50% “fully agree”; 8.3% “partly
agree”) and teleradiology consultations are regularly used by one-third of the departments
(19.4% “fully agree”; 16.7% “partly agree”). For a quarter of the respondents, teleradiology
examinations are carried out in line with radiation protection regulations (16.7%
“fully agree”; 8.3% “partly agree”) ([Fig. 1 ]
b ). The main barriers to further expansion are legal issues (n = 14, multiple answers),
billing difficulties (n = 12), and technical limitations (n = 11). Other difficulties
(n=15) include staff capacity, billing options, lack of quality assurance, and lack
of guidelines. The greatest benefits of teleradiology are considered the all-around
improved pediatric radiology care (47.2%) and the increased expertise (41.7%). Only
one respondent sees no benefit (2.8%) ([Fig. 2 ]).
Fig. 1
a Relevance and utilization of teleradiology in pediatric radiology; b Utilization and reporting on teleradiology consultations and examinations in line
with the StrSchG and MedStrSchV medical radiation regulations.
Fig. 2
a Limiting factors in the further implementation of teleradiology (multiple answers
possible); b Greatest benefit of teleradiology.
Artificial Intelligence
Over 60% of pediatric radiologists consider AI to be important for pediatric radiology
(38.9% “fully agree”; 22.2% “partly agree”). 30% are undecided (“neither agree nor
disagree”). Only 8% see AI as a topic of no importance (5.6% “partly disagree”; 2.8%
“fully disagree”). However, AI plays a major role in the daily work of less than 15%
of the pediatric radiologists surveyed (2.8% “fully agree”; 11.1% “partly agree”)
([Fig. 3 ]). The most common areas of application are segmentation or volumetry (n = 15, multiple
answers), dose reduction (n = 14), image acquisition/processing (n = 12), and detection/CAD
(n = 10). We asked respondents whether three areas of application in pediatric radiology
were suitable for an AI solution and were already regularly in use: over 80% view
bone age determination as a suitable area of application (61.1% “fully agree”; 22.2%
“partly agree”), but only 30% use such systems regularly (30.6% “fully agree”). The
use of AI in pediatric chest X-rays is considered appropriate by one-third of respondents.
In contrast, actual usage is 14% for image acquisition (11.1% “fully agree”; 2.8%
“partly agree”) and just under 3% for the detection of pathologies (2.8% “fully agree”).
75% of respondents stated that AI can help improve CT protocols (38.9% “agree”; 36.1%
“partly agree”). In more than half of the departments, innovative algorithms such
as iterative reconstruction techniques and AI applications such as deep learning are
used to improve image quality and reduce dose in CT examinations of children (44.4%
“fully agree”; 11.1% “partly agree”) ([Fig. 4 ]
a , b , c ). The main obstacles to implementation of AI applications are costs (n = 21, multiple
answers) and technical limitations (n = 19). Other obstacles include skepticism (n
= 11), lack of expertise (n = 9), and legal issues (n = 7). Other obstacles (n=5)
include low availability of pediatric AI applications, low return on investment, and
administrative hurdles. The greatest benefit is the reduction in workload (37.1%),
followed by time savings (17.1%). Other benefits (17.1%) include improved computing
power, image quality, and error prevention, especially as support for doctors inexperienced
in pediatric radiology ([Fig. 5 ]
a , b ).
Fig. 3 Relevance and use of AI in pediatric radiology.
Fig. 4 Examples of AI applications in pediatric radiology; a Bone age determination; b Chest X-rays; c Computed tomography.
Fig. 5
a Limiting factors in the further implementation of AI (multiple answers possible);
b Greatest benefit of AI.
Discussion
This article presents a first survey on attitudes towards teleradiology and AI in
pediatric and adolescent radiology in German-speaking countries.
In an international comparison, Shelmerdine et al., in a survey among members of the
European Society of Paediatric Radiology (ESPR), the Society of Pediatric Radiology
(SPR), and other professional associations, reported that pediatric radiologists had
a positive perception of AI [18 ]. Other surveys on the use of teleradiology or AI are limited to adult radiology
[19 ]
[20 ] or specific applications, for example, surveys by an individual commercial provider
(bone age determination using BoneXpert ) [13 ]. The pediatric radiologists surveyed in the present study view both teleradiology
and AI and its applications as forward-looking and relevant, although the current
level of use lags behind the positive attitudes towards these topics. This may indicate
possible administrative hurdles in implementation or inadequacies of these techniques.
One limiting factor of our study is the number of only 36 responding pediatric radiologists.
The aforementioned global survey by ESPR, SPR, etc. on AI in pediatric radiology received
240 responses. Of these, only 159 responses were from pediatric radiologists, the
rest were from non-medical professions. Overall, the response rate of all medical
and non-medical respondents was 9.9% [18 ]. With a response rate of 8.6% from 419 members of the GPR and SGPR, our study has
a comparable response rate. Other surveys in pediatric radiology on various topics,
such as neonatal neurosonography or home office during the COVID-19 pandemic, have
shown higher response rates between 13% and 17% [21 ]
[22 ]. Our survey received responses from Germany and, in equal proportions, from Austria
and Switzerland. We received more responses from the three DACH countries compared
to the percentages from other multinational surveys previously [13 ]
[18 ]. The responses came predominantly from university departments, which is consistent
with previous studies [18 ].
Our survey shows that 75% of respondents consider teleradiology to be important for
pediatric radiology, but actual use is moderately low. Legal uncertainties and personal
concerns are cited as the main limiting factors resulting in lower usage. Teleradiology
is subject to various laws and standards, for example, the Radiation Protection Act
(StrSchG) in Germany and the Medical Radiation Protection Ordinance (MedStrSchV) in
Austria. In addition to radiation protection regulations, other legal areas such as
data protection or employment law have to be followed. The Swiss Society of Radiology
(SSR), for example, summarizes the multitude of legal areas and standards to be observed
in Switzerland in its white paper on teleradiology [23 ]. When preparing the survey, it was important for us to deliberately differentiate
between a teleradiology examination in line with the StrSchG regulation and a teleradiology
consultation in line with German case law. In the former case, the specialist doctor
remotely supervises the examination performed by a medical technologist for radiology
(MTR), which requires close and extensive cooperation in advance with the relevant
referring physician [24 ]. In Switzerland, this corresponds to performing the examination in the teleradiology
satellite [23 ]. The latter are less extensive and more readily available. Our survey confirms the
perception of many pediatric radiologists that teleradiology consultations for second
opinions are regularly carried out in more than half of the departments. In cases
of substantial discrepancies between the assessments by general or pediatric radiologists,
a pediatric radiology second opinion can bring about therapy-relevant changes to the
findings and thus help to ensure that children are receiving the best possible medical
care [17 ]. However, this leads to a significant increase in the workload for pediatric radiologists.
Compared to consultations, fewer teleradiology examinations are carried out in line
with the StrSchG, MedStrSchV, or the Swiss satellite model. The comprehensive network
of referring physicians and pediatric radiology centers recommended in the GPR strategy
paper could further promote teleradiology cooperation, especially for second opinions
and consultations [1 ]. The pediatric radiologists surveyed see the benefits of teleradiology in the improved
availability of pediatric radiology expertise. In addition to legal uncertainties,
billing issues and technical limitations are cited as limiting factors for developing
the offering of teleradiology services. The GPR strategy paper therefore recommends
technical and personnel needs assessments in order to plan the necessary expansion
of services [1 ]. The lack of guidelines on the use of teleradiology for children and adolescents
is considered a limiting factor. The heterogeneity of teleradiology services and related
regulations was further identified as a limiting factor. Based on the GPR strategy
paper, we focused the survey on the cooperation of pediatric radiology centers with
peripheral referring physicians and not on possible diagnosis from home. Even though
home office and reduced workload could basically make the discipline more attractive
[16 ]
[22 ], the respective centers are responsible for the relevant regulations. An Italian
survey from general radiology also supports our observations that teleradiology is
seen as an approach oriented on the future [19 ]. The concern described in that study – that teleradiology could impair cooperation
with referring clinicians – was not borne out by our study. On the one hand, this
could be due to the focus on cooperation with other radiology departments; on the
other hand, pediatric radiology generally works closely already with pediatrics and
pediatric surgery [1 ]. Overall, despite existing concerns, teleradiology is seen as an opportunity to
improve radiology care for children and adolescents.
The pediatric radiologists surveyed also consider AI to be an important technology
for pediatric and adolescent radiology. The international survey by Shelmerdine et
al. mentioned above underscores this positive perception [18 ]. Due to our focus on key technologies to improve pediatric radiology care on an
all-around basis, we have deliberately omitted questions related to the fear of being
replaced. Especially since previous studies provided no evidence of these concerns
[18 ]. However, the use of AI in everyday clinical practice, similar to the use of teleradiology,
lags behind the positive attitudes towards these topics.
Although AI is applicable in almost every step of the radiology workflow [4 ]
[5 ]
[6 ], the pediatric radiologists surveyed use AI applications mainly for segmentation/volumetry,
dose reduction, image acquisition and processing, and less for detection. Due to the
special radiation risk of children [25 ], it is not surprising that a major focus of AI use is on dose reduction and optimal
image acquisition [9 ]
[10 ]. We further differentiated the use of AI using the following three examples: bone
age determination, chest X-ray, and CT.
The biggest discrepancy was found regarding determination of bone age, which over
80% of respondents considered a suitable activity for AI, but only a third of respondents’
departments actually use a relevant AI solution. In general, bone age determination
is considered one of the prime examples of AI solutions for pediatric radiology, which
is why the “RSNA Pediatric Bone Age Machine Learning Challenge” was launched in 2019
[12 ]. Three approved commercial solutions are now available on the market [11 ], and they provide solid age determinations according to Greulich and Pyle [14 ]. A Europe-wide survey by Thodberg et al. among users of the commercial application
BoneXpert may offer possible explanations for this discrepancy: although AI-based
analysis saves time, it is usually only used as a supplement to radiology reporting.
Human expertise is still required when assessing the X-ray of the left hand, as AI
applications do not yet detect syndromal or metabolic abnormalities [13 ]. This applies primarily to the initial presentation for bone age determination.
The assessment of bone age during the follow-up examinations, some of which are annual,
can easily be carried out by the AI application. Implementation and operating costs
for AI solutions are expensive [5 ]. In addition, the quality of continuing professional development of doctors could
be compromised.
Few respondents consider the use of AI to be suitable for chest X-rays. Relevant AI
solutions are hardly used by the pediatric radiologists surveyed. This applies particularly
to diagnostic AI applications, while 13.8% of respondents use AI applications to improve
image quality in chest X-rays. One explanation for the low level of use is the focus
of diagnostic AI applications on individual diseases such as tuberculosis. The more
intensive dissemination of AI applications is also hampered by the lack of pediatric
cohort studies and data sets [5 ]
[26 ]. In contrast, AI is used more frequently in image acquisition for quality improvement
and dose savings [9 ]. Using virtual scatter grids in X-ray bed scans can improve image quality and reduce
radiation exposure [5 ]
[7 ]
[27 ]. The smallest discrepancy between potential and actual use of AI applications can
be observed in the application for dose reduction in CT. More than half of respondents
use AI for image enhancement and dose optimization in CT. AI methods are making it
possible to develop ultra-low-dose protocols [6 ]
[28 ]. Iterative reconstruction algorithms have long played an important role in dose
reduction in CT. Even though iterative reconstruction techniques do not correspond
to AI applications such as deep learning, we have included them in this question because
they are currently combined with AI methods [29 ].
The pediatric radiologists surveyed in our study consider the biggest future benefits
of AI to be reduced workloads, time savings, and error prevention. This reflects the
generally positive attitude towards AI that was reported by surveys in general radiology
[20 ]
[30 ]. Our results confirm the positive attitude of pediatric radiologists towards AI
as observed in the international pediatric radiologist survey by Shelmerdine et al.
[18 ]. AI applications to optimize radiation exposure are being used more frequently,
particularly in German-speaking countries [9 ]. The main limiting factors for the use of AI are costs and technical limitations.
These can be exacerbated by inadequate funding of pediatric medicine, as is the case
in Germany, for example. Interestingly, in contrast to teleradiology, significantly
fewer legal concerns were expressed with regard to AI.
Conclusions
Our survey shows that teleradiology and artificial intelligence are seen as future
technologies for pediatric and adolescent radiology. However, there are still several
obstacles to their more widespread use. Increased integration of these technologies
could be promoted through clear legal frameworks, adapted billing options, and technical
improvements. Additional studies and targeted pilot projects could help to establish
these technologies further in clinical practice and subsequently evaluate whether
the anticipated benefits, such reduced workloads, have actually materialized.
