IEEE Circuits and Systems Magazine - Q3 2022 - 36

pulse was turned on for 30 seconds and subsequently
turned off for 3 minutes. If there was no seizure onset detected
after a period of stimulation, the IPG would be turned
off and it would send a message back to the WER.
B. Test Results
The coastline length and the slope of the TESD were calculated
both in the inter-ictal period and in the ictal period.
We took
UT 156Wms, UT 780Sms (or M ),5=
=
=
and a sampling frequency of 256 Hz. The mean ACL, AS, SCL,
SS (both in the ictal period and in the inter-ictal period),
THCL and THS of each patient are presented in Table II. For
each patient, parameter k can be determined by the ratio
of the ACL in the ictal period and the ACL in the inter-ictal
period. Similarly, parameter d can be determined by the
ratio of the AS in the ictal period and the AS in the interictal
period. As a result, the mean ACL and AS of the six
patients in the inter-ictal period are 205 μV and 3.9 μV/ms,
respectively, while the mean ACL and AS in the ictal period
are 602 μV and 8.7 μV/ms, respectively. These result in a
mean k of 3 and a mean d of 2. Therefore, the THCL (with
a mean value of 294 μV) and THS (with a mean value of
7.3 μV/ms) were obtained based on equations (7) and (8).
Further, we took one UTS as unit number, and checked the
number of true positives (TP), true negatives (TN), false
positives (FP) and false negatives (FN) of each patient
within 68 UTS (each including 40 UTS in the inter-ictal
period and 28 UTS in the ictal period). The results are listed
in Table III, where the accuracy, sensitivity and specificity
of the TESD are defined as [38], [39]
Accuracy =
Sensitivity =
Specificity =
TP TN FP FN
TP TN
++ +
+
TP FN
TP
+
TN FP
TN
+
#100 %
#10 %0
# 10 %0
(7)
(8)
(9)
respectively. An average accuracy of 88.2%, sensitivity
of 80.7% and specificity of 95.3% were achieved in the TESD.
Table III.
Performance of the TESD.
Patients
No.1
No.2
No.3
No.4
No.5
No.6
Average
36
Total
(UTS)
68
68
68
68
68
68
68
TP
(UTS)
26
27
25
26
27
27
26.3
IEEE CIRCUITS AND SYSTEMS MAGAZINE
FP
(UTS)
2
1
3
2
1
1
1.7
TN
(UTS)
34
35
34
32
33
34
33.7
FN
(UTS)
6
5
6
8
7
6
6.3
Accuracy
(%)
88
91
87
85
88
90
88.2
Sensitivity
(%)
81.3
84.4
80.7
76.5
79.4
81.8
80.7
Specificity
(%)
94.4
97.2
91.9
94.1
97.1
97.1
95.3
THIRD QUARTER 2022
Fig. 18 shows the recorded scalp EEG signals of the
rat in the inter-ictal period, during the induced epileptic
seizure period and 15 minutes after stimulation. It
can be found that the spikes were very strong when an
epileptic seizure was induced by the PTX. However, the
number of the spikes was significantly reduced after the
15-minute-stimulation and thereby the seizure has been
effectively suppressed.
Table IV exhibits the performance comparison of the
state-of-the-art epileptic seizure detection algorithms in
the CLNS. In [40], a feature extraction algorithm called
MinMaxHist was used to measure the topological patterns
of the EEG signals and achieve a high classification accuracy.
[41] employed a spike-train detector (which detects
high-energy segments of the EEG and analyses the correlation
between them) and a oscillatory seizure detector running
in parallel. [42] exploited discrete wavelet transform
(DWT) to obtain features from the logarithm of variance
of detail and approximate coefficients of single-channel
scalp EEG signals for classification. [43] used the DWT for
feature extraction and artificial neural network (ANN) for
classification. [44] applied the empirical wavelet transform
(EWT) for feature extraction, which can determine the
joint instantaneous amplitudes and frequencies in signal
adaptive frequency scales. It can be found that the proposed
TESD has not only low computation complexity but
also acceptable accuracy for the real-time CLNS.
The quiescent power of the WER, IVNS and DFAICL
transmitter (or each in standby mode) are 220 μW, 305 μW
and 1.6 mW, respectively. The power consumption of
the WER in recording mode is 25.3 mW. The power consumption
of the IVNS in stimulating mode depends on
the amplitude of the stimulating pulse and within the region
of 3~97.7 mW. The minimum power consumption of
the DFAICL transmitter in the battery charging mode is
149~724 mW. Therefore, the total power consumption of
the HybridVNS is 174.3~847 mW. Table V presents the performance
summary of the HybridVNS and a comparison
with commercially available closed-loop VNSs.
IEEE Circuits and Systems Magazine - Q3 2022

Table of Contents for the Digital Edition of IEEE Circuits and Systems Magazine - Q3 2022
