Page 27 - ITU KALEIDOSCOPE, ATLANTA 2019

ICT for Health: Networks, standards and innovation

while the TCP-based network throughput is growing some strategies to balance the volume of engaged label
exponentially until reaching the system’s limitation. parameters and the processing accuracy. To this end,
This is because in an edge computing-assisted ICN a joint force from the off-the-shelf cooperative edge
scenario, we may always obtain the required content computing chips, dedicated image processing chips and
from the nearest caches without routing back to the state-of-the-art algorithms, are comprehensively needed.
remote server through the BS.
Edge computing-assisted ICN indeed has some 5. CONCLUSION
drawbacks, one of which is sub-network congestion.
In this work, the 5G-enabled health systems are
It happens especially when a bunch of sub-network
introduced. By leveraging the 5G NR and AI-based
subscribers simultaneously request the same content.
technologies, we can greatly improve the medical service
For future studies, effective routing and cache
quality for the remote areas, and upgrade in-hospital
distribution strategies can be good topics. On the
medical services. The solutions and demonstrations
other hand, optimal content division and labeling
of the 5G-enabled health systems are introduced.
strategy are some other topics for future studies.
For future studies, some new 5G NR technologies,
1000
network architecture redesigned from being data-centric
900 to information and user-centric, the image processing
algorithms and specialized devices are needed for better
800 implementation of the 5G-enabled health systems.
System throughput (MBit/s) 600 This work is supported by the Zhengzhou University
700
ACKNOWLEDGMENT
500
400
Research Startup Foundation under grant:124-32210907
300
Technology Major Projects of China under Grants:
200 and 124-32211247; the Natural Science and
2017ZX03001001-004; the JSPS KAKENHI of Japan
100
under Grant JP18K18044, the National Funding from
0 the FCT-Fundação para a Ciência e a Tecnologia
10 0 10 1 10 2 10 3
through the UID/EEA/500008/2019 Project; and
Client number
by the Brazilian National Council for Research and
Figure 7 – System throughput comparison between Development (CNPq) via Grant No. 309335/2017-5.
ICN-assisted edge computing and conventional scheme.
REFERENCES

4.3 VR/AR, medical image processing and the [1] D. Zhang, Z. Zhou, S. Mumtaz, J. Rodriguez, and
AI-based technologies T. Sato, “One integrated energy efficiency proposal
for 5G IoT communications,” IEEE Internet of
Currently, image label and detection are needed for Things Journal, vol. 3, no. 6, pp. 1346–1354, Dec.
the implementation of remote surgery and rescue. For 2016.
example, in machine learning based image detection,
we need to label the targets beforehand. We can [2] Y. Fan, L. Yang, D. Zhang, G. Han, and D. Zhang,
unmistakenly detect the target after training with “An angle rotate-qam aided differential spatial
existing objectives. However, it is still hard for the AI modulation for 5G ubiquitous mobile networks
to detect the unexperienced targets. The omitted target (accept),” Mobile Networks and Applications, 2019.
poses risks to the automatic ambulance and remote
[3] J. J. P. C. Rodrigues, I. de la Torre, G. FernÃąndez,
surgery. Similarly, for the medical image applications,
and M. LÃşpez-Coronado, “Analysis of the security
we also need to label these disease symptoms, once
some labels are omitted or mistakenly studied by the and privacy requirements of cloud-based electronic
algorithm, it will result in misdiagnosis. health records systems,” Journal of Medical
Internet Research, vol. 15, no. 8, p. e186, Aug. 2013.
Apart from the labeling and detection method, ML
algorithm’s processing speed is slowed down by the large [4] Y. Xue and H. Liang, “Analysis of telemedicine
scale and even growing AR/VR-based three-dimensional diffusion: The case of china,” IEEE Transactions
data volume. This is mainly due to the limited ability on Information Technology in Biomedicine, vol. 11,
of existing processing devices. In order to accelerate no. 2, pp. 231–233, Mar. 2007.
the processing speed, large scale deep learning (DL)
high-rank matrix factorization (MF) algorithms and [5] S. Thelen, M. Czaplik, P. Meisen, D. Schilberg,
dedicated processors are needed in the future. Due to and S. Jeschke, “Using off-the-shelf medical devices
the large scale and even growing data, we also need for biomedical signal monitoring in a telemedicine

– 7 –