Page 178 - Read Online

Page 26 of 33 Arab Hassani. Soft Sci 2023;3:31 https://dx.doi.org/10.20517/ss.2023.23

may be able to find an optimum trade-off between spatial resolution and interference. The data collected
from multimodal sensors through a signal acquisition system could be processed using mathematical
algorithms or, more often, ML algorithms to obtain better predictions [177-180] . However, multimodal signals
from multimodal sensors are often redundant and complementary. Therefore, human experts should
extract key features from the original data and provide them as inputs to ML. To avoid this difficult and
time-consuming manual extraction, multimodal fusion algorithms could be applied. The structure of these
algorithms is hierarchical and consists of at least three main levels: (1) data-level fusion to combine data
from multimodal sensors; (2) feature-level fusion to extract features from the pre-processed data of each
sensory modality and combine them to create a unified feature set; and (3) decision-level fusion to obtain
feature-based classifications and their fusion to generate the final decision . The signals obtained using
[181]
multimodal sensors need to be labelled so that they can be processed using supervised ML algorithms,
which is a time-consuming process. Moving towards semi- and unsupervised ML algorithms might be
beneficial from the perspective of reducing data labelling, but doing so reduces accuracy. The use of hybrid
supervised and unsupervised methods could be a solution to leverage the strengths of each method, but this
approach requires better data integration, model selection, and performance evaluation. Apart from these
data processing challenges, data collection over the long term will lead to security and privacy concerns
related to the processes of data collection, transmission, and storage . The high-level architecture of
[182]
Internet of Things (IoT) components in healthcare can be divided into three layers: (1) the perception layer
(i.e., sensing devices); (2) the network layer that connects the perception layer devices through wired and
[183]
wireless connections; and (3) the application layer (i.e., end-user and data storage) . By performing a
security risk management analysis of each layer individually, the process of implementing appropriate
security measures can be simplified. Wireless wearable sensors have attracted great interest in healthcare.
Usually, sensor data are wirelessly communicated to a smartphone and then sent to a centralised cloud-
based repository, where the relevant medical team can work on them. Wearable sensors usually have low
power consumption because their power resources are limited. Therefore, the type of wireless technology
used in each sensor is determined by bandwidth, power, and transmission range limitations. The data
security of wireless sensors is extremely important, but the applied security measures should be consistent
[184]
with the low-power requirements of the sensors, which translates into less efficient security . The
incorporation of various energy harvesting techniques can help mitigate the power source limitations of
sensors and, therefore, enhance their security efficiency. The development of self-powered sensors,
[185]
integration of energy harvesting sources with multifunctional sensors , and hybridisation of several
energy harvesting mechanisms could help elongate the lifetimes of sensor arrays, apart from the efforts to
[186]
reduce sensor power consumption [187,188] .
The development of new approaches that consider the personal needs of each patient opens new horizons in
the diagnosis, treatment, and prevention of diseases. Despite all the mentioned challenges and considering
the complex nature of diseases, multimodal sensor arrays that can simultaneously monitor multiple analytes
or stimuli are the way forward for obtaining more information on the health status of patients and
providing accurate diagnoses.

DECLARATIONS
Author contributions
Initiated the review and wrote the manuscript: Arab Hassani F

Availability of data and materials
Not applicable.

Financial support and sponsorship
Not applicable.