Page 161 - ITU KALEIDOSCOPE, ATLANTA 2019

ICT for Health: Networks, standards and innovation

financing. It includes major R&D project budget agencies need to answer. Recently, the International
support, profit distribution and national health Telecommunication Union (ITU) established an
insurance management, etc. Especially with the ITU/WHO Focus Group on artificial intelligence for
progress of aging, how to effectively reduce the cost of health (FG-AI4H), which works in partnership with
medical insurance has become a common problem that WHO to especially establish a standardized assessment
needs to be solved in various countries. framework for the evaluation of AI-based methods for
health, diagnosis, triage or treatment decisions [42]
5.2 Standardization
➢ Security：It includes the concerns of data security,
Under the overview control of the collaboration mechanism, network security and product security. Data security
standardization could technically act as an accelerator for refers to data ownership, data handling policy and
integration innovation. Consideration on standardization privacy protection during usage. Network security
could be prepared as the following part. refers to cybersecurity, avoiding products being
attacked by illegal cyberattacks. Product security refers
➢ Data format and interface: Much previous efforts have to the safe use of the product. Currently with the rapid
been made to meet the demand of medical system development of AI for health application, most of the
interconnection and compatibility. Personal health applications investigated could not provide relevant
device standards (ISO / IEEE 11073) and Digital evidence or peer-reviewed research to support their
Imaging and Communications in Medicine (DICOM) products. According to a study published in Nature
are known as addressing the interoperability. Though Digital Medicine, only two of the 73 applications in
AI for health focuses more on the application layer, their survey provided evidence of research [43].
updates on data formats and interfaces should also be
considered to meet the development requirement. ➢ Ethics：Ethics are important to consider especially for
the health field. Major countries and international
➢ Data quality: It refers to the standardization of content organizations have established AI ethics institutions
requirement input to AI algorithms, and it is a new focusing on the discussion of ethical guidelines and
demand due to new technology of AI. Medical images standards. In June 2019 the US Food and Drug
used by AI may contain undesirable artefacts (e.g. Administration released a discussion paper of proposed
background noise), lack focus, exhibit uneven ‘Regulatory Framework for Modifications to Artificial
illumination or under/overexposure, etc. [35]. Intelligence/Machine Learning (AI/ML)-Based
Moreover, the quality of the annotation for AI training Software as a Medical Device (SaMD)’ and also
is also critical. To form a unified understanding and requested feedback including on ethical aspects [44].
workflow on annotation among different groups of
clinicians is a difficult but necessary task. Several 6. CONCLUSION
public datasets are released for research, including
Kaggle, ImageNet, Messidor database [36-40], but for In this work, we choose the perspective of interaction of ICT
long-term development and scaled application, on the health industry. An industrial framework of the digital
standardization on data content and annotation are very health industry was proposed to better understand the
necessary requirements for sustainable development. interaction between component factors from the health and
ICT sides. We extracted and reconstructed different
5.3 Regulation component factors to expand the framework from the
traditional health industry to digital health. The traditional
With collaboration mechanism acting as a macro-control, health industrial framework is divided into service and
standardization as a technical accelerator, regulation is to management parts, and ICT factors are listed as sensors,
define the bottom line of the industry and maintain its networks, data resources, platforms, applications and
legality. The International Medical Device Regulators solutions. This paper also tracks the interaction through the
Forum (IMDRF) was established to discuss the common development history of the digital health industry, from
problems in international medical device regulation, with institutional informationization to regional
representatives from regulatory authorities in Australia, informationization, and finally to service intelligentization.
Brazil, Canada, China, European Union, Japan and the Following such a developmental roadmap, AI was chosen as
United States, as well as WHO [41]. From their working one of the most powerful technologies to discuss the key
group setting, main concerns on regulation can be divided trends from data, computing power and algorithms. Service
into reliability, security and ethics parts. and management processes in the health industry were
observed on the effects of ICT penetration. In the end,
➢ Reliability: The performance of AI algorithms can be exogenous factors such as a collaboration mechanism,
evaluated in the metrics of accuracy, precision, ROC, standardization and regulation were proposed and discussed
F-measures, interpretability, robustness, generalization, to better prepare for supporting the sustainable development
etc. In the face of such an emerging technology, how to of digital health in the AI era.
evaluate AI/ML-based software as a medical device
(SaMD) is a problem that all national regulatory

– 141 –