IEEE Solid-States Circuits Magazine - Fall 2023 - 44

such as the beta-band power of LFP
within the 13- to 30-Hz range, were
utilized in an adaptive or closed-loop
manner to control DBS [Figure 2(a)].
The adaptive DBS resulted in significant
improvements in patients'
motor scores compared to continuous
stimulation, while delivering
Amplified, Digitally Filtered Over Beta
Band (13-30 Hz), and Thresholded LFPs
Stimulation Trigger
DBS
Beta Filtered LFP
DBS Trigger
(a)
∗
∗
-80
-60
-40
-20
cDBS
∗
∗
aDBS Random
(b)
FIGURE 2: (a) The concept of adaptive DBS (aDBS) for PD [16]. Bipolar LFP is filtered within
the beta band and smoothed. When the beta power crosses a predefined threshold, stimulation
is triggered in a monopolar configuration between the bipolar recording electrodes.
Stimulation ceases when the beta power drops below the threshold. (b) Improvements in
clinical motor score (Unified PD Rating Scale), and the speech intelligibility for various stimulation
conditions on eight Parkinsonian patients. Modified from [16] and [17]. *Significant
differences after applying false discovery rate for multiple comparisons correction. cDBS:
continuous deep-brain stimulation.
50
60
70
80
Off DBS
aDBS
Speech
Baseline
less than half the energy and reducing
speech side effects [Figure 2(b)]
[16], [17]. Moreover, it demonstrated
superior performance compared to
random intermittent stimulation.
Biomarker-driven closed-loop control
has also found widespread application
in epilepsy [5], [18], [19],
where seizure biomarkers from
ECoG or intracranial EEG can trigger
stimulation for seizure prevention,
as in the RNS device. Similarly, in
the context of Tourette's syndrome
and treatment-resistant depression,
stimulation can be activated based
on features linked to tic onset [20] or
mood state [21]. In general, closedloop
stimulation offers the potential
for more effective stimulation compared
to its open-loop alternatives,
resulting in reduced side effects and
improved energy efficiency, thereby
extending the device's battery life.
cDBS
Closed-Loop Implant
Off-Chip
Neural Input
Amplifiers, ADC
Biomarker
Extraction
Machine
Learning
Therapeutic
Feedback
Stimulation
Model
Training
AI-Enabled Closed-Loop
Neuromodulation
Despite its advantages over openloop
stimulation, the simple method
of thresholding individual biomarkers
may not be optimal for closed-loop
control. Accurate symptom prediction
often necessitates multiple biomarkers
from different input channels,
increasing system complexity. Furthermore,
recent studies [22], [23],
[24], [25], [26], [27] highlight the potential
of personalized decoders, particularly
those utilizing modern machine
learning (ML), for reliable decoding of
brain states in movement disorders
[25], [28], epilepsy [29], [30], [31], [32],
depression [24], migraine [26], and
memory [22], as well as enhanced
motor decoding for BCI [23], [33],
[34], [35]. However, AI-based closedloop
control that utilizes external
or cloud-based computing resources
may lead to high telemetry power
and prolonged loop latency, hindering
real-time feedback.
Alternatively, Figure 3 illustrates
FIGURE 3: The framework for AI-driven closed-loop stimulation in neural interfaces: On-chip
biomarker extraction and ML processor facilitate rapid symptom detection and therapeutic
feedback, while model training is performed externally.
44
FALL 2023
IEEE SOLID-STATE CIRCUITS MAGAZINE
a conceptual framework for a closedloop
neural interface empowered by
AI. This framework integrates on-chip
biomarker extraction and advanced
ML algorithms, facilitating real-time
% Motor Improvement
Speech Intelligibility
Symptom Detection
Deterioration
IEEE Solid-States Circuits Magazine - Fall 2023

Table of Contents for the Digital Edition of IEEE Solid-States Circuits Magazine - Fall 2023
