IEEE Systems, Man and Cybernetics Magazine - July 2020 - 45

Table 1. Sensor embodiments and their technical characteristics.
Embodiment

Reference

Targets

Power

Communication

Sensing
-Technology

Footprint
(mm)

Surgical-Care
Pathway

Wearable

[15]

StO 2

Battery

Bluetooth

Optical

11 # 18

Pre/post

[21]

GRF

Battery

2.4-GHz radio
frequency

Physical

55 # 40 # 10

Pre/post

[22]

Temperature, sweat
pH, Na +, lactate

Battery

Bluetooth

Electrochemical

12 # 30 # 30

Pre/post

[14]

pH, Na +, K +,
-bioimpedance

Tethered

Electrochemical/
electrical

5 # 5+
100 # 12

Peri/post

[34]

pH, temperature,
lactate, glucose

Ultrasound

LSK

Electrochemical

20-mm
-diameter

Peri/post

[36]

Skin temperature,
hydration

Radio
frequency

NFC

Optical/
electrical

10 # 5

Peri/post

Tattoo-like

[46]

Skin hydration, pH,
glucose, temperature

Tethered

Electrochemical/
electrical

15 # 15

Peri

Catheter-like

[27], [54]

SSI biomarkers,
bacteria

Tethered

Optical

0.3-mm
-diameter

Peri/post

Implantable

information locally, increases power consumption. Onnode processing in real time can discriminate between sensor recordings and store/transmit only clinically useful
data and processed information, which may also be compressed, reducing volumes and power consumption while
enabling real-time anomaly detection.
Apart from probabilistic, dimensionality reduction, and
ensemble techniques, deep learning has been proposed for
sensing [50], where features can be gleaned directly from
data, and a generic classifier can be formed to provide
detection, even in unforeseen scenarios. Although the
training process is computationally expensive, feedforward classification requires relatively little computation,
which low-power processors can perform in real time.
Cloud computing, decision-support systems, and the
fusion of fundamentally different data also play an important role in these developments [21], [51], [52].
Neuromorphic computing aims in the envelopment
brain-inspired paradigms. This contrasts with classical
von Neumann computing architectures characterized by
the physical separation of memory and computation, relying on serial and iterative processing with a high power
consumption. In the brain, computation is performed in
the time domain by neurons propagating spikes, while
memories are stored locally in the synapses. Processing is
event-driven, leading to sparse, low-power computation.
Such bioinspired neuromorphic systems can facilitate
more power- and chip-area efficient, parallel, and real-time
computation and learning, enabling cognitive sensor networks to manifest. The recent realization of the memristor
sparked interest in neuromorphic computing, since it
enables memory and computation to coexist by emulating
biological synapses [53].

The Road Ahead
Sensing technologies are likely to play an ever-increasing
role in global health care and will continue to evolve as
biomarkers for disease detection, treatment, and posttreatment monitoring are developed. As our ability to process information advances, we are likely to discover novel
applications for physiological data streams. Combining
longitudinal data, from organism to cellular levels, will
add further insight to currently available models. The standardization of sensing data across health-care institutions,
sectors, and systems will provide a powerful means to
evaluate disease treatment efficacy. Furthermore, these
data will enable the modeling of pathology, from individual
to populational levels, which is crucial for identifying new
treatment modalities. During the short term, the incorporation of sensing technologies into routine clinical pathways can make great socio-economic impacts in pathologies
such as SSIs, where a limited early intervention can greatly
affect the treatment duration and cost.
Nevertheless, many challenges need to be dealt with
before routine deployment. Consideration must be given to
ethical and regulatory pathways to enable commercialization
and facilitate bench-to-bed device translation. Patient and
public engagement are important to ensure that devices are
designed with, rather than for, users so that they are sympathetic to patients' needs. Consumers are different from
patients. The former buy devices of their own free will; their
purchases are not dictated or enforced. That is not the case
for patients, and societal and psychological aspects will play
a role in the adoption of new sensor technologies. Finally,
economic evaluations must be conducted to ensure that
health-care providers can afford new interventions with
proven benefits.
Ju ly 2020

IEEE SYSTEMS, MAN, & CYBERNETICS MAGAZINE

IEEE Systems, Man and Cybernetics Magazine - July 2020

Table of Contents for the Digital Edition of IEEE Systems, Man and Cybernetics Magazine - July 2020