Commun Nonlinear Sci Numer Simulat 36 (2016) 219237

Contents lists available at ScienceDirect

Commun Nonlinear Sci Numer Simulat

journal homepage: www.elsevier.com/locate/cnsns

Optimal control of directional deep brain stimulation in the

parkinsonian neuronal network
Denggui Fan, Zhihui Wang, Qingyun Wang
Department of Dynamics and Control, Beihang University, Beijing 100191, China

a r t i c l e

i n f o

Article history:
Received 1 July 2015
Revised 6 November 2015
Accepted 8 December 2015
Available online 14 December 2015
Keywords:
Parkinsons disease
Subthalamic nucleus (STN)
Charge-balanced bi-phasic (CBBP) pulse
Deep brain stimulation (DBS)
Directional steering control

a b s t r a c t
The effect of conventional deep brain stimulation (DBS) on debilitating symptoms of Parkinsons disease can be limited because it can only yield the spherical eld. And, some side effects
are clearly induced with inuencing their adjacent ganglia. Recent experimental evidence for
patients with Parkinsons disease has shown that a novel DBS electrode with 32 independent
stimulation source contacts can effectively optimize the clinical therapy by enlarging the therapeutic windows, when it is applied on the subthalamic nucleus (STN). This is due to the selective activation in clusters of various stimulation contacts which can be steered directionally
and accurately on the targeted regions of interest. In addition, because of the serious damage
to the neural tissues, the charge-unbalanced stimulation is not typically indicated and the
real DBS utilizes charge-balanced bi-phasic (CBBP) pulses. Inspired by this, we computationally investigate the optimal control of directional CBBP-DBS from the proposed parkinsonian
neuronal network of basal gangliathalamocortical circuit. By appropriately tuning stimulation for different neuronal populations, it can be found that directional steering CBBP-DBS
paradigms are superior to the spherical case in improving parkinsonian dynamical properties including the synchronization of neuronal populations and the reliability of thalamus
relaying the information from cortex, which is in a good agreement with the physiological
experiments. Furthermore, it can be found that directional steering stimulations can increase
the optimal stimulation intensity of desynchronization by more than 1 mA compared to the
spherical case. This is consistent with the experimental result with showing that there exists
at least one steering direction that can allow increasing the threshold of side effects by 1 mA.
In addition, we also simulate the local eld potential (LFP) and dominant frequency (DF) of
the STN neuronal population induced by the activation of 32 different contacts with optimal
stimulation intensity and immediately after the stimulation, respectively. These can reveal regional differences in pathological activity within STN nucleus. It is shown that in line with
the experimental results directional steering stimulation can induce the low-amplitude LFP
which implies the occurrence of desynchronizing regime, as well as the distribution of DF can
locate at the 1340 Hz of beta frequency range. Hopefully, the obtained results can provide
theoretical evidences in exploring pathophysiologic activity of brain.
2015 Elsevier B.V. All rights reserved.

Corresponding author. Tel.: +86 10 82332003; fax: +86 10 -82332003.

E-mail address: nmqingyun@163.com (Q. Wang).

http://dx.doi.org/10.1016/j.cnsns.2015.12.005
1007-5704/ 2015 Elsevier B.V. All rights reserved.

220

D. Fan et al. / Commun Nonlinear Sci Numer Simulat 36 (2016) 219237

1. Introduction
Deep brain stimulation (DBS) involves an external pulse generator that can produce high frequency stimulations (HFS) for
the specic subcortical targets through the deeply implanted electrodes. Due to the restorability after stimulation and the ability
to effectively alleviate the pathological symptoms of patients with neurological disorders, DBS has been gradually becoming the
dominant surgical treatment for various neurodegenerative diseases including Parkinsons disease (PD) [1]. However, DBS cannot
be equally effective for all the parkinsonian symptoms [2]. Also, the stimulation parameters for the DBS can still only be adopted
empirically [3]. In order to achieve optimal stimulation effect, the stimulation parameters should be optimized and different
stimulation therapies also should be employed for various patients with different clinical characteristics. Presently, these stimulation paradigms involve the open-loop stimulation [4], closed-loop stimulation [5,6], adaptive feedback stimulation [7], coordinated reset stimulation [814] and coordinated delayed feedback stimulation [1517], and so on. Therein, feedback stimulation
can be applied in the necessary amount to suppress pathological activities. Batista et al. [16,17] compared the eciency of both
the conventional high-frequency stimulation and the delayed feedback stimulation procedures, and pointed out that the delayed
feedback stimulation procedure is thought to save energy when implemented by the surgical such as the implanted pacemaker
[16,17]. The multi-site stimulation modality is particularly relevant for the coordinated reset of neural sub-populations which is
introduced as an effectively desynchronizing stimulation technique by Tass [9]. During the coordinated reset stimulation, highfrequency but short sequences pulse trains are administered on the different sites in a coordinated manner. Many works [1014]
have highlighted the potential effects of the coordinated reset stimulation and encouraged the further development of this approach. In addition, due to the serious damage to the neural tissues, the charge-unbalanced stimulation is not typically indicated.
The real surgical DBS utilizes the charge-balanced bi-phasic (CBBP) pulses (see Fig. 2). However, the mechanism underlying the
effects of DBS is still enigmatic and under debate.
Parkinsons disease is the typical neurodegenerative disorder which originates from the degeneration of dopaminergic neurons in the substantia nigra pars compacta (SNc) [18]. The SNc is the key part of the basal gangliathalamocortical motor circuit
that is closely correlated to the pathophysiological movement disorder [19]. The schematic of the basal ganglia thalamocortical
motor circuit is shown in Fig. 1, where the basal ganglia is mainly composed of subthalamic nucleus (STN), globus pallidus pars
externa (GPe) and interna (GPi), striatum, substantia nigra pars compacta (SNc) and reticulata (SNr) [19]. We can nd that striatum and STN are the primary input nuclei of basal ganglia receiving information from the cortex, and as coped in [20,21] the main
output nuclei GPi and SNr are considered as the single structure due to their similar connections and cytological functions. The
damage of SNc (see Fig. 1(b)) rst leads to the depletion of dopamine, which can further indirectly modulate the information ow
received by the subcortical structures from cortex through the direct/indirect pathways consisting of the downstream synaptic
connections within the basal ganglia and modulated by the dopamine D2 /D1 -receptor-bearing striatal neurons in the striatum
[8,19], respectively. In addition, the loss of dopamine can eventually inuence the synaptic function of GPi to thalamus (Thal),
and then affect the ability of thalamus to relay the information of cerebral cortex. Continuous high-frequency deep brain stimulation (HF-DBS) has been widely used for the therapy of patients with advanced parkinsonian symptoms [5,22]. Specically, the
structures or nuclei under the cerebral cortex such as STN [23], GP (GPi and GPe) [2426] and the ventral intermediate (VIM)
[27,28] have been chosen as the stimulation targets of HF-DBS for the parkinsonian symptoms, where the STN is applied for the
PD, and the GPi, GPe and VIM are applied for the essential tremor (ET) [29,30]. Particularly, relevant works have demonstrated

Fig. 1. Schematic of the classic basal gangliathalamocortical model, adapted from Obeso in [19] and composed of cerebral cortex, basal ganglia and thalamus
(Thal), where basal ganglia mainly consists of substantia nigra (SN), striatum, globus pallidus pars externa (GPe), globus pallidus pars interna (GPi), substantia
nigra pars reticularis (SNr) and subthalamic nucleus (STN). The healthy (a) and parkinsonian (b) states are depicted. Excitatory projections are indicated by red
solid arrows. Inhibitory projections are shown in blue lines with closed circles. The dashed lines with open arrows are the external excitatory current inputs. The
thickness of lines represents the degree of activation of projections. The striatum can communicate with GPi or SNr not only through the direct pathway but also
the indirect pathway consisting of the GPe and STN. (For interpretation of the references to color in this gure legend, the reader is referred to the web version
of this article.)

D. Fan et al. / Commun Nonlinear Sci Numer Simulat 36 (2016) 219237

221

Fig. 2. Overview of the charge-balanced bi-phasic (CBBP) stimulus pulses: Biphasic asymmetric pulse, with short-duration, high-amplitude anodic leading phase
indicated by the light red bar and long-duration, low-amplitude cathodic lagging phase indicated by the light blue bar. (For interpretation of the references to
color in this gure legend, the reader is referred to the web version of this article.)

Fig. 3. (a) Conventional deep brain stimulation (DBS) electrode with four lengthwise cylindrical contacts and (b and c) Novel 32-contacts DBS electrode with
32 oval disc-shaped contacts latitudinally arranged on 8 rows of 4 contacts, adapted from Contarino et al. [1]. Activation of one contact (red) from traditional
electrode (a) or 12 contacts in 3 rows (red) from the experimental 32-contact electrode (b) can produce a spherical stimulation eld (red), which can inuence
both the target STN and its adjacent internal capsule (IC) (e). Activation of a cluster of 4 contacts (red) (c) from the 32-contact electrode in the rhomboidal
manner yields a stimulation eld steering stimulation approximately toward one quadrant, which can effectively keep away from the adjacent area (f). (e, f):
Atlas images drawn by hand adapted from Martens et al. [34]; ZI, zona incerta. (d) The structure of STN with the shape of spheroid. The specic contacts are
selected to activate directional steering stimulation from the anterior, posterior, lateral and medial paradigms, which correspond to the identications in (e) and
(f). (For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article.)

that the most effective stimulation region is within the STN, especially the dorsal/dorsomedial or the caudal zona incerta (ZI)
[31].
However, the DBS effect can be signicantly limited because the conventional stimulator only yields a spherical eld (see
Fig. 3(a)), which will inuence both the stimulation targets and their adjacent ganglia (see also Fig. 3(e)), hence inducing the
side effects [1,32]. This disadvantage of conventional DBS electrode, which harbors four cylindrical contacts (Fig. 3(a)), is based
on the technology of voltage-controlled stimulation with only a single source [33]. To optimize the ecacy of clinical therapeutic DBS and largely reduce the side effect, it is imperative to design novel electrodes. Fortunately, engineering advances have
promoted the emergence of new DBS devices with current-controlled multiple independent sources [3335]. Compared to the
traditional DBS electrodes with single source, these new stimulators can more specically and accurately steer the stimulated
current towards the targeted regions of interest [34,35] by the selective activation of targeted neuronal populations. Although
steering stimulated current has not be a new concept [36,37], its application for DBS is emerging and challenging due to the cost
for searching much larger stimulation parameter space. In order to improve the potential application of steering DBS, Contarino,
et al. recently developed a new type of DBS electrode with 32 small independent source contacts (see Fig. 3(b) or (c)) to further modulate the thresholds for benecial and side effects of stimulation surgery in the STN for the patients with Parkinsons
disease [1]. The 32 contacts in this new electrode can be independently and selectively activated in clusters to realize the directional steering of stimulation eld (see Fig. 3(f), compared to the Fig. 3(e)). They can also be used for directionally recording
the local eld potentials (LFP) to explore the local brain pathological activity, and provide the feedback of adaptive DBS [1,7].

222

D. Fan et al. / Commun Nonlinear Sci Numer Simulat 36 (2016) 219237

Fig. 4. Arrangement of the basal gangliathalamocortical model network, adapted from So, et al. in [53] originated from RT model in [50]. The network is
composed of STN, GPe, GPi and thalamocortical relay (TC) nuclei with 320 cells in each neuronal population. Sparse connections adopted within the basal ganglia
thalamocortical model network: Excitatory projections (inset) indicated by red arrows; Inhibitory projections indicated by black lines with closed circles; Each
STN cell sends excitations to the neighboring 2 GPe cells and 2 GPi cells; Each GPe cell sends inhibitions to the 2 neighboring STN cells and 2 GPe neurons with an
immediate neighboring one to the left and one after next to the right, as well as 2 GPi cells which skip 2 GPi cells nearest to the GPe; Each GPi cell sends inhibition
to one TC cell. Applied constant currents on GPe, STN and GPi are shown in black arrows, respectively. CBBP-DBS pulse and input of sensorimotor cortex into STN
and TC are indicated by open and solid arrows, respectively. As in the previous works [8,5456], our network architecture has a periodic structure, so that cells 1
and 320 in each array are neighbors. (For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article.)

Experimental results [1] have shown that directional steering STN-DBS (see Fig. 3(d)) by a novel 32-contact electrode can more
effectively control the DBS directions and improve the therapeutic effect compared to the conventional stimulation (Fig. 3(a)) or
current 32-contact stimulation in the standard spherical mode (Fig. 3(b)). Therefore, the steering current stimulation may be a
promising technique for the directional exploration of brain areas both common to DBS surgical therapy and hitherto unreachable. However, to fully reveal the mechanisms of DBS, great improvements should be made for the stimulation devices, strategies
and techniques, for example, the recent attempts to the treatment of Parkinsons disease by the focused ultrasound stimulation
or lesion [38,39].
The computational models for investigating the DBS effect of Parkinsons disease can give theoretical evidences for clinical applications with the understanding of origins and suppressions of pathological oscillations [37,4042]. There exists the consistent
view that parkinsonian symptoms are mainly characterized by the low reliability of relaying cortical information by thalamus
and high-frequency synchronous oscillations for the nuclei within basal-ganglia. Hence, various mathematical models have been
proposed to reproduce the physiological and pathological activities of these nuclei [6,4347]. Also, relevant desynchronizing DBS
techniques [48] have been adopted to destroy this synchronous behavior and restore the ability of thalamus to faithfully relay
the cortical information [8,49]. The most computational models for the parkinsonian state originated from the work of Rubin
and Terman [50], which is based on the basal gangliathalamocortical circuit and composed of a few different types of neuronal
populations with sparse and structured connectivity. However, these models can still not help us to well understand the physiological mechanisms underlying new DBS experiments in a convincing manner [51,52]. Recently, based on the model of Rubin
and Terman (RT) [50], So et al., modied the basal ganglia model network (see Fig. 4) [53] which can both reproduce the pathophysiological activity of basal ganglia network and realistically simulate the feature of HF-DBS [53]. Presently, we use this model
to explore the recent experimental results for the patients with Parkinsons disease [1], which shows that directional steering
STN-DBS by a novel 32-contact electrode is superior to the conventional stimulation (Fig. 3(a)) or current 32-contact stimulation
in the standard spherical mode (Fig. 3(b)). For ne-grained modulation of the proportion of stimulated individual neuron, in this
text, according to the architectural features of 32-contact electrode and 3-D anatomical structures of subcortical ganglia, 1280
conductance-based model neurons are equally divided into the four neuronal populations with assuming that stimulation eld
produced by each of the 32 contacts can inuence 10 cells around it. Meanwhile, we mainly implement ve various stimulation
paradigms of anterior, posterior, lateral, medial and spherical modes to investigate the directional control effects of STN-DBS on
the parkinsonian state through multiple electrode contacts.
2. Model and method
In this section, we will rst introduce the model network of basal gangliathalamocortical motor circuit used in this paper
and their concrete mathematical descriptions. Secondly, the setup for the computational process of directional steering DBS will
be provided.
2.1. The network model
The computational model used here is the modied basal ganglia thalamocortical model network (see Fig. 4) proposed by So
et al. [53], originated from RT model [50]. Based on the previous sparsely-connected framework of RT model, they reconstructed

D. Fan et al. / Commun Nonlinear Sci Numer Simulat 36 (2016) 219237

223

the sparse connectivity of modied model network which can reproduce more realistic electrophysiological observations [53].
In particular, this network is consisted of four equally-sized group of GPe, STN, GPi and thalamocortical relay (TC) neurons with
total 1280 conductance-based model cells (see Fig. 4). Each nucleus is represented by 320 neurons according to the structure of
32-contact electrode, where each contact can inuence 10 cells around it by the produced stimulation eld. Each GPe neuron
inhibits two immediate neighboring GPi neurons which skip the two nearest GPi neurons around GPe, as well as sends inhibitions
to the two immediate neighboring STN neurons and two GPe neurons with an immediate neighboring one to the left and one
after next to the right. Each STN neuron respectively sends excitations to the nearest two GPe neurons and two GPi neurons in
the array of neurons. Each GPi neuron only inhibits the immediate neighboring one TC neuron, and each TC neuron also receives
synaptic inhibition from only one nearest GPi neuron. As in the previous works [8,5456], our network architecture has a periodic
structure, so that cells 1 and 320 in each array are neighbors. In addition, as considered in Refs. [50,53,5759], all of the three
app app
app
nuclei located in the basal-ganglia receive constant positive bias currents, i.e., IGPe , IST N and IGPi , respectively, which can be seen
as the net synaptic current input from other brain regions to these cells to maintain the baseline ring rates of 70 Hz for GPe,
app
10 Hz for STN and 80 Hz for GPi [53,57,58]. Therein, IGPe can be viewed as the striatal synaptic inputs into the GPe cells. Moreover,
STN neurons receive excitatory deep brain high-frequency stimulation and TC neurons receive excitatory sensorimotor cortex
(SMC) signals [50,53]. Particularly, in order to match the physiological recordings, the applied currents used in this modied
basal gangliathalamocortical model [53] are much higher than that in the original RT model [50].
app
All the model parameter values can be found in the modied model [53], where decreasing these bias currents, IGPe/ST N/GPi ,
can drive the state transitions of basal gangliathalamocortical model from healthy condition to Parkinsons disease (PD), and
vice versa, which shows the increase of synchronization in the STN, GPe and GPi neuronal populations [6067]. This is consistent
with the modeling results [50,68,69], and can also be supported by the electrophysiological observations from clinical patient
with PD [70], dopamine-depleted rodents [71], as well as MPTP-induced primate model of PD [72]. Especially, the larger values
app
of IGPe/ST N/GPi correspond to the normal state of system which is characterized by the random or desynchronous dynamics, while
the parkinsonian state corresponds to the smaller values of bias currents indicated by the consistent or synchronous dynamics.
2.2. Model for each neuron type
As proposed in the work of So et al. [53], the conductance-based model HodgkinHuxley (HH) neurons are used to model
dynamical behaviors of the four type of STN, GPe, GPi and TC neurons. Generally, the canonical form of HH equations [8] can be
used to represent the change of membrane potentials in the STN, GPe and GPi model neuronal populations, which can be written
as follows,

CmV  =

Iion Isyn + Iapp + Idbs

(1)

where Cm = 1 F cm2 is the membrane capacitance for all cell models,

Iion = IL + IK + INa + IT + ICa + IAHP

(2)

is the total ionic current through the membrane, where IL = gL (V VL ) is the leak current, IK = gK n4 (V VK ) is the potassium current, INa = gNa m3 (V )h(V VNa ) is the sodium current, IT = gT a3 (V )r (V VCa ) is the low-threshold T-type Ca2+ cur[Ca]
) is the Ca2+ -activated, afterrent, ICa = gCa c2 (V )(V VCa ) is the high-threshold Ca2+ current and IAHP = gAHP (V VK )( [Ca]+k
AHP

hyperpolarization K+ current. Therein, intracellular Ca2+ concentration has the control action on the after-hyperpolarization
current, which follows the dynamics equation,

[Ca] = (ICa IT kCa [Ca](t )),

(3)

where is a constant related to the buffer effect of calcium, and kCa is the constant representing the calcium pump rate. Among
these ionic currents, gx represents the channel conductance and Vx represents the reversal potential for the ion x. The specic
parameter values for these ionic currents can be found in the previous works [8,5053] (also see Table 2).
Note that in the modied RT model STN neurons, IT = gT a3 (V )b2 (r )(V VCa ) is modied by replacing the r with b2 (r ),
where

b ( r ) =

1
rb

1 + exp( )
b

1
1 + exp( b )

(4)

is the steady-state of gating variable, which contributes to the rebound bursts [50]. And, ICa = gCa c2 (V )(V VCa ) is modied by
replacing the c with c. However, the other steady-states of the voltage-dependent gating variables in the ionic current equations
can be described as the following general form:

x (V ) =

1
1 + exp( V (t)xx )

(5)

where x = m, h, n, r, a, c, both x and x are constants. Hence,the following general equation can be used to describe the slow
gating activation and inactivation variables,

x = x (x x(t ))/x

(6)

224

D. Fan et al. / Commun Nonlinear Sci Numer Simulat 36 (2016) 219237

where x = h, n, r, c represents the sodium, potassium, T-current and Ca2+ current inactivation, respectively. x is a constant
parameter, and x is the time constant, which obeys the rst-order kinetics,

x = x0 +

1
 x V (t )x 

1 + exp

(7)

where x0 and x1 are constants. It has to be noted that Eq. (7) only holds for the cases of x = h, n, c. For the r , it follows the
Eq. (7) only for the STN cells, but in the ionic currents of GPe and GPi cells it is usually considered as a constant [55] .
In addition, Iapp is the applied constant current which is used to adjust the ring properties of various types of neurons. Isyn
is the synaptic current which will be elaborated in the next section. Idbs is the charge-balanced biphasic HF-DBS current injected
into the neuronal population by micro-electrodes.
Thalamic neurons or TC cells are also modeled by the HodgkinHuxley type equations, but simpler than those for the basal
ganglia [50]. As used in the previous works [50,73], the model used for the TC cells is a standardly reduced version of an earlier
formulation [74,75], which decreases the dimensionality of the model by one variable [76]. There are only three state variables,
VTh , hTh and rTh for the reduced TC neurons [76]:
 = I I I I I
CmVTh
L
Na
K
T
GPiTh + ISMC

(8)

hTh = (h (VTh ) hTh )/h (VTh )

(9)

 = (r (V ) r )/ (V )
rTh
Th
r
Th
Th

1 F cm2 ,

(10)

where Cm =
IL = gL (VTh VL ), INa =
VNa ), IK =
VK ) and IT =
gT a2 (VTh )rTh (VTh VT ). IGPi Th is the synaptic current sent by GPi to TC. Note that the gating variables hTh and rTh have the
similar state equations with the Eq. (6), but their time constants hTh and rTh differ slightly. Here,

hTh =

gNa m3 (VTh )hTh (VTh

gK (0.75(1 hTh ))4 (VTh

(11)

aTh
+ bTh
h
h

with

 V + 46 

aTh
h = 0.128exp

(12)

18

and

bTh
h =
while

1 + exp V +23
5

,

(13)


 V + 25 
rTh = 0.15 28 + exp
.
10.5

(14)

ISMC is the sensorimotor cortex (SMC) input, which is modeled as a series of monophasic current pulses with the form:

ISMC = iSMC H1 (sin(2 t/SMC )) [1 H1 (sin(2 (t + SMC )/SMC ))]

(15)

where H1 is the Heaviside bi-value step function (i.e., H1 (x) = 1, if x > 0 and H1 (x) = 0, if x 0). Usually, SMC and iSMC =
3.5 A cm2 are taken as the period and amplitude of ISMC , respectively. SMC = 5 ms is the positive input duration which determines the duty circle of the pulse train. In order to simulate the non-regular nature of SMC input from cortex to the thalamus, as
used in the modied network model [53] the instantaneous frequencies of incoming pulses, i.e., 1/ SMC , is drawn from a gamma
distribution with an average rate (AR) of 14 Hz and a coecient of variation (CV) of 0.2. In particular, SMC can be represented as
follow:

SMC = 1000/(A, B ),

where A =

1/CV2 ,

(16)

B = AR/A,  represents the gamma distribution and SMC has the unit of ms.

2.3. Synaptic currents

Isyn in Eq. (1) is the current from synaptic coupling, which can be written by the form of Ip q . This represents the direction
of the synaptic current from neuron p to neuron q, whose general form is,

Ipq = gpq (Vq Vpq )

skp

(17)

where p/q represents the pre-/post-synaptic neuron type, which can be the GPe, GPi, STN or TC neurons. gp q is the maxi
mal synaptic conductance, Vp q is the synaptic reversal potential, k skp sums over the pre-synaptic neurons, where synaptic
variable skp is the synaptic conductance of the kth pre-synaptic neuron.

D. Fan et al. / Commun Nonlinear Sci Numer Simulat 36 (2016) 219237

225

For the case of pre-synaptic neuron p = GPe, i.e., GPe STN, GPe GPi and GPe GPe, the sp still obeys the rst-order
dynamical differential equation of the following form as used in the original RT model [50],

sp = Ap (1 sp )H (Vp p ) Bp sp
where H (V ) =

is the smooth approximation of the bi-value step function H, and

)
1+exp( V H

3). Ap , Bp and p
H

(18)

and

are constant parame-

ters (see Table

are also constants contributing to the synaptic currents. However, for the case of p = STN or GPi,
i.e., STN GPe, STN GPi or GPi TH, the sp follows the alpha function as used in the modied model [53], that is governed by
a system of two coupled rst-order ODEs:

sp = z

(19)

z =

(20)

1 u(t ) 2 z 3 sp

where u(t) = 1, if Vp 10 mV at time t, representing the occurrence of action potential in the presynaptic cells. Otherwise,
u(t) = 0, 1 , 2 and 3 are constants (see Table 3).
2.4. Directional steering DBS setup
In what follows, we introduce the directional steering stimulation protocol based on the directional steering STN-DBS by a
novel 32-contact electrode and the newly proposed modied RT model [53]. On the basis of the physiological experiments [1], as
in the original RT model [50] or the modied RT model [53], high-frequency DBS pulse current, Idbs , are applied intracellularly to
STN neurons with varying stimulation amplitudes (SA) of 0.58 mA cm2 , invariant pulse width of 60 s and high frequency of
130 Hz. Whereas, due to the risk to damage the neural tissues, charge-unbalanced monophasic stimulation current is typically not
indicated and the charge-balanced bi-phasic (CBBP) pulses [7780], where, in most cases, a short cathodic pulse is followed by a
long anodic recharge phase, are frequently utilized for the real DBS surgical. However, in line with the experimental procedures,
we maintain the invariant short positive anodic pulse width of 60 s followed by a long cathodic recharge phase. Therefore, the
charge-balanced bi-phasic pulse current (see Fig. 2), Icbbpdbs , can be described by the following form of rectangular pulse:

Icbbpdbs (t ) = Istim H2 {H1 (sin(2 t/Tp )) [1 H1 (sin(2 (t + p )/Tp ))]}

(21)

where Istim is the stimulation intensity varying in [0.5, 8] and having the unit of mA, Tp = 1000/130 ms is the period of pulse and

p = 60/1000 ms is the duration of positive current. H1 is the Heaviside function, H2 is a bi-value discrete function as follow,

H2 (x ) =

Tp p

x = 1,
x=0

(22)

During the simulation, as performed in the experiments [1], we progressively increase the current intensity at a xed step of
0.5 mA with every second.
According to the physical structure of 32-contact electrode, we consider 320 model neurons in each nucleus and assume
that these neurons uniformly distribute around this electrode when applied on the targeted nucleus with the stimulation eld
activated by each contact inuencing 10 cells. In order to more clearly describe the directional steering stimulation, rstly, the
32 contacts of electrode are mapped into a plane with 32 numbered circles (see Fig. 5). Then, consistent with the experimental
process, four group of 4 contacts surrounded by the hexagon with different color of edges are selected in clusters to be activated, which represent the different directional steering stimulation modes of anterior (black), posterior (blue), lateral (pink)
and medial (light green) directions for the STN, respectively (see Figs. 3(d)(f) and 5). By contrast, all the three row of 12 contacts surrounded by a rectangle with rounded corners are activated to represent the stimulation of spherical mode of the STN.
Following the space scale of real stimulation eld activated by the four contacts of each directional stimulation mode [1], we
assume that each directional stimulation can eventually involve 16 contacts, i.e., 160 STN cells, while the spherical mode can
nally stimulate 280 STN neurons of the target nucleus STN with 28 contacts being involved in the stimulation eld activated by
the three row of 12 contacts [1] (see Fig. 5).
Because of the peer-to-peer equality between neurons, one basic problem, numerically, is how to determine or distinguish the
directivity of neurons stimulated within the STN neuronal population. However, the 3-D anatomical structure of basal ganglia
thalamocortical motor circuit provides a new way and notion for it [37,8183], which shows that cells in the posterior and
caudal of STN mainly project to the GPi, 1/3 medially located cells of STN mainly project to the GPe and 1/3 laterally located
cells of STN mainly project to the other areas of the globus pallidus. In addition, from Fig. 4 we can see that each STN cell
sends projections to two GPe cells and similarly each GPe cell sends projections to two STN cells, which leads to the equivalent
synaptic function between STN and GPe cells. By contrast, each STN cell sends projections to two GPi cells but does not receive
any feedback from GPi cells. Besides the local effects, DBS also generates systematic effects via the activation of the projection
pathways from the stimulated nuclei [84]. Therefore, the mechanisms of DBS cannot be understood by considering only the local
changes of activity in the stimulated area but the overall changes of activity throughout the connected network [84]. Presently,
we determine the steering directions in the relative method based on the anatomy. For example, on the basis of spatial scale
and structural framework of basal ganglia-thalamocortical motor circuit (see Fig. 1C in [81]), we can simultaneously stimulate

226

D. Fan et al. / Commun Nonlinear Sci Numer Simulat 36 (2016) 219237

Fig. 5. (Upper panel): Mapping of the 32-contact electrode into the 2-D plane. Numbered circles correspond to the 32 contacts in terms of top-to-bottom
and left-to-right order. Suppose that 320 model STN neurons uniformly distribute around this electrode when applied on the targeted nucleus STN with the
stimulation eld activated by each contact inuencing 10 STN cells. Each group of 4 contacts encircled by hexagons with different colored edges are selected to
steer anterior (black), posterior (blue), lateral (pink) and medial (green) current stimulations. 12 contacts in the 3 rows of 4 contacts, surrounded by the rounded
rectangle, are selected to perform spherical current stimulation eld. (Lower panel) According to the physical scale of stimulation eld yielded by the 32-contact
electrodes, directional steering stimulation eld can cover the area of 16 contacts, i.e., 160 STN neurons in the model network, while spherical stimulation eld
can cover the area of 28 contacts, i.e., 280 STN neurons in the model network. (For interpretation of the references to color in this gure legend, the reader is
referred to the web version of this article.)

a certain number of specic STN cells, a small proportion of random GPe cells and a lesser proportion of random GPi cells, to
realize the medially directional steering stimulation mode.
Moreover, stimulation of bers or projection pathways is sensitive to the intensity of stimulation [84]. As the stimulation
intensity increases, an increasingly broader range of bers can be involved due to its increasing excitability. Therefore, in accordance with the experiments, we must consider the strength effect of directional steering DBS on the stimulation targets with the
stimulation intensity increasing. To address this issue, we introduce two scale factors, 1 and 2 , to describe the effect of stimulation intensity applied in stimulation target STN on the adjacent GPe or GPi due to the activation of the projection bers. By this
method, we can exibly control the progressive stimulation scope for other nuclei as the stimulation intensity increases. These
scale factors are sigmoid function and dependent on the stimulation current, Istim , which can be represented by the following
equation (See Fig. 6):



Istim
1,2 (Istim ) = 1/ 1 + 1,2

(23)

where 1,2 determines the steepness referring to Ref. [85] (see Table 4). Throughout the whole paper, the directional steering
strategies are listed in the Table 1.
3. Numerical results
On basis of the directional steering DBS strategy proposed above, in this section, we aim to mathematically simulate the
therapeutic effect of novel 32-contact electrode applied on the STN of patients with Parkinsons disease by means of the model
proposed by So et al. [53], originated from the RT model [50]. In line with the experimental process [1], we perform ve different
stimulation modes of anterior, posterior, lateral, medial and spherical stimulation eld on the STN to computationally investigate the effect of directional steering DBS on the parkinsonian symptoms. For clarity and ne-grained modulation of directional
steering, we expand the 100 cells of modied model [53] (16 cells in the original RT model [50]) to 320 cells in each neuronal
population within the basal gangliathalamocortical motor circuit, with the assumption that 10 cells can be affected by stimulation eld activated by each contact of 32-contact electrode (see Figs. 24). In addition, parkinsonian symptoms are mainly

D. Fan et al. / Commun Nonlinear Sci Numer Simulat 36 (2016) 219237

227

0.95

0.9

0.85

1
2

(Stim Ratio)

0.8

0.75

0.7

0.65

0.6

0.55

0.5

4
Stim Amp (mA)

Fig. 6. Evolutions of the scale factors, 1 (red asterisk) and 2 (blue asterisk), as stimulation amplitude (SA), Istim , varies from 0.5 mA to 8 mA with a xed step
of 0.5 mA. The 1,2 in Eq. (14) are taken as 1.001 and 1.0001, respectively. (For interpretation of the references to color in this gure legend, the reader is referred
to the web version of this article.)
Table 1
Five various stimulation paradigms of anterior, posterior, lateral, medial and spherical modes, determined by simultaneously stimulating a xed number of specic STN cells and different portions of random GPe or GPi cells
according to the anatomical features of basal gangliathalamocortical motor circuit. 1, 2 (Istim ) represents the ratio
of stimulation intensity at specic Istim .
Nuclei
Directions

STN (xed)

GPe (random)

GPi (random)

Anterior
Posterior
Lateral
Medial
Spherical

160
160
160
160
280

1 1/15320
2 1/5320
1 1/10320
1 1/3320
1 (1/3 + 1/10 + 1/15 ) 320
+ 2 1/5 320

2 1/15320
1 2/3320
2 1/10320
2 1/5320
2 (1/5 + 1/10 + 1/15 ) 320
+ 1 2/3 320

characterized by the low reliability of thalamus relaying the information from the cortex and electrophysiologically observed
high-frequency synchronous oscillations. Hence, the improvement of these two pathological characteristics are used to be as the
measurement of stimulation effect under the ve various stimulation paradigms.
3.1. Electrophysiological characteristics for the basal gangliathalamocortical motor circuit
The modied model of basal ganglia-thalamocortical motor circuit proposed by So et al. [53], originated from RT model [50],
can reproduce the more realistic pathophysiological activities of Parkinsons disease [53]. From Fig. 7(a), we can see that in
the healthy state, the three nuclei of STN, GPi and GPe within this basal gangliathalamocortical loop can re irregularly with
the baseline ring rates of 10 Hz [57], 80 Hz and 70 Hz [58], respectively. Also, the thalamus or TC cells can faithfully relay
or respond to the information input from SMC. However, this irregular state can be destroyed by the pathological change of

228

D. Fan et al. / Commun Nonlinear Sci Numer Simulat 36 (2016) 219237

Fig. 7. Firing patterns transition from healthy condition to parkinsonian state: (a) in the healthy conditions, STN, GPe and GPi neurons exhibit random spiking
ring, and thalamus or TC cells can faithfully relay the information from the sensorimotor cortex (SMC); (b) in the parkinsonian state, STN, GPe and GPi neurons
re in burst-like manner, and the thalamus cannot always accurately respond to the information from SMC. (c) Under the application of CBBP-DBS current on the
target STN, the rings of STN, GPe and GPi neurons can all restore to random rings of normal state but with higher ring frequency, as well as the relay ability
of thalamus can also be restored. (For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article.)

D. Fan et al. / Commun Nonlinear Sci Numer Simulat 36 (2016) 219237

229

Table 2
All the parameters related to the ionic currents used for each cell type where all potentials have the unit of mV, currents have the unit of A/cm2 , conductances have the unit of
mS/cm2 , and time constants have the unit of ms. - indicates the NULL.
Ionic currents: Iion
Cells
Currents

TH

STN

GPe

GPi

IL

gL = 0.05
VL = 70

gL = 2.25
VL = 60

gL = 0.1
VL = 65

gL = 0.1
VL = 65

INa

gNa = 3
VNa = 50
m = 37
m = 7
h = 41
h = 4
-

gNa = 37
VNa = 55
m = 30
m = 15
h = 39
h = 3.1
h = 0.75
h0 = 1
h1 = 500
h = 57
h = 3

gNa = 120
VNa = 55
m = 37
m = 10
h = 58
h = 12
h = 0.05
h0 = 0.05
h1 = 0.27
htau = 40
h = 12

gNa = 120
VNa = 55
m = 37
m = 10
h = 58
h = 12
h = 0.05
h0 = 0.05
h1 = 0.27
h = 40
h = 12

IK

gK = 5
VK 75
h = 41
h = 4
-

gK = 45
VK = 80
n = 32
n = 8
n = 0.75
n0 = 1
n1 = 100
n = 80
n = 26

gK = 30
VK = 80
n = 50
n = 14
n = 0.1
n0 = 0.05
n1 = 0.27
n = 40
n = 12

gK = 30
VK = 80
n = 50
n = 14
n = 0.1
n0 = 0.05
n1 = 0.27
n = 40
n = 12

IT

gT = 5
VT = 0
a 60
a = 6.2
r = 84
r = 4
-

gT = 0.5
VT = 0
a 63
a = 7.8
b = 0.4
b = 0.2
r = 67
r = 2
r = 0.2
r0 = 7.1
r1 = 17.5
r = 68
r = 2.2

gT = 0.5
VT = 0
a = 57
a = 2
r = 70
r = 2
r = 1
r = 30
-

gT = 0.5
VT = 0
a 57
a = 2
r = 70
r = 2
r = 1
r = 30
-

ICa

gCa = 2
VCa = 140
c = 20
c = 8
c = 0.08
c0 = 1
c1 = 10
c = 80
c = 26

gCa = 0.15
VCa = 120
c = 35
c = 2
-

gCa = 0.15
VCa = 120
c = 35
c = 2
-

IAHP

gAHP = 20
VAHP = 80
 = 3.75e 5
kCa = 22.5
kAHP = 15

gAHP = 10
VAHP = 80
 = 1e 4
kCa = 15
kAHP = 10

gAHP = 10
VAHP = 80
 = 1e 4
kCa = 15
kAHP = 10

substantia nigra pars compacta (SNc) (see Fig. 1(b)) into the parkinsonian state characterized by the regular rings of STN cells
in a periodic tremor-like fashion and high-frequency bursting oscillations of the GPe and GPi cells [50,53]. These changes can be
seen in Fig. 7(b), where we can also see that the thalamus can not accurately respond to the input from SMC any more. Hence,
from Figs. 7(a) and (b), it can be seen that the electrophysiologically observed basic healthy and parkinsonian states related to
the basal gangliathalamocortical motor circuit can be represented by this model. What is more, the parkinsonian state can be
effectively improved, as seen in Fig. 7(c), when the CBBP-DBS electrode is applied on all the 320 STN cells of the modied model
with the stimulation parameters as used in the recent experiment [1]. Under the stimulation of CBBP-DBS electrode, the ring
patterns of GPe and GPi cells can restore towards their normal states, as well as the thalamus can faithfully respond to the input
from SMC. Therein, the ring rates of all the STN, GPe and GPi cells have been greatly increased.

230

D. Fan et al. / Commun Nonlinear Sci Numer Simulat 36 (2016) 219237

0.35
ANT
POST
LAT
MED
SPH
HEALTHY
PD

Error Index (EI)

0.3
0.25
0.2
0.15
0.1
0.05
0
0

Stim Amp (mA)

Fig. 8. Effect of stimulation (SA), corresponding to the Istim in Eq. (21), on the error index (EI) of thalamus cells under ve type of anterior (ANT: red solid circle
with blue edge), posterior (POST: blue asterisk), lateral (LAT: green square with blue edge), medial (MED: inverted black solid triangular with blue edge) and
spherical (SPH: yellow ve-pointed star with blue edge) stimulation paradigms. The blue and red squares on the SA = 0 indicate the EI corresponding to the
healthy and parkinsonian states, respectively. As used in [1], stimulation parameters are performed with 130 Hz and 0.06 ms, and SA is varied in [0.5, 8] at a xed
step of 0.5 mA. (For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article.)

3.2. Error index of information relay for the thalamus

One of the most important characteristics correlated to the parkinsonian state is the unfaithfulness of thalamic transmission
for the information from SMC. This unfaithfulness is originally and physiologically induced by the depletion of dopaminergic
neurons in the SNc, which can successively affect the downstream synaptic functions within the basal gangliathalamocortical
network (see Fig. 1) and eventually lead to the over-inhibitory action of output nucleus, GPi, on the thalamus, hence impairing
its responses to the input from SMC. However, this parkinsonian symptom can be improved by the external excitatory DBS input.
As in the previous RT model [50] and its modied model [53], in order to quantitatively understand the control effect of these
directional steering stimulation paradigms on the reliability of thalamus in relaying the information from SMC, we introduce the
error index (EI) [8,50,53] as the reliability-based measurement, which can be expressed as follow,

EI =

Nmiss + Nburst + Nspur

NSM

(24)

where we consider 3 type of errors in the thalamic transmission (see also Fig. 1C in [53]), which are misses, bursts and spurious
events. A miss occurs if the TC cell fails to spike, a burst occurs if the TC cell res more than one spike during the stimulating
input of 25 ms, and a spurious event occurs when the TC neuron res without the stimulation input. Nmiss , Nburst and Nspur are
the numbers of missed, bursting and spurious event of TC cells in response to SMC input, respectively. Hence, EI describes the
proportion of false responses to the input pulse from SMC accounting for the total number of NSM SMC inputs, which shows the
poor reliability of thalamic relay. Obviously, the larger values of EI reect the worse performance of thalamic relay.
Similar to the experimental process [1], we vary the stimulating amplitudes (SA) from 0.5 mA to 8 mA with an increasing
step of 0.5 mA to investigate the control effects of ve different stimulation modes of anterior, posterior, lateral, medial and
spherical stimulation eld on the reliability of TC cells. As seen in the Fig. 8, when SA = 0 mA, the EI corresponding to the healthy
state is approximately equal to zero indicated by the blue square, which biologically implies the optimal reliability of TC cells
relaying the information from the cortex. By contrast, under the parkinsonian state, the system shows a lower reliability of TC
cells corresponding to the larger value of EI more than EI = 3 indicated by the red square. However, after the application of
CBBP-DBS, as the SA is lower than SA = 1.5 mA sharp decreases entering into the dips of the EI values can be found for the
four directional steering stimulations, which means the ability of TC cell to relay the information from the cortex can be largely
restored. Whereas, for the spherical stimulation mode, there exists a sharp jump entering into the plateau phenomenon of EI
corresponding to the suddenly weakened reliability of thalamus. In addition, when the SA is lower than SA = 1 mA, we can see
from Fig. 8 that compared to the other four directional steering stimulation, under the spherical stimulation mode the system
achieves the smaller EI value of thalamus, which corresponds to the higher reliability of thalamus relaying the information from
SMC. This may be because spherical stimulation paradigm can produce the larger stimulation eld than that of the directional
steering stimulations and also scarcely affect the adjacent non-targeted regions due to the weak SA, hence achieving the better
relay reliability (i.e., lower EI) of thalamus than the cases of directional steering stimulations. However, as the stimulation current
increases more than SA = 1.5 mA, apart from the other three directional steering stimulation modes, the medially steering

D. Fan et al. / Commun Nonlinear Sci Numer Simulat 36 (2016) 219237

231

Table 3
All the parameters used for the synaptic connections among the four different neuronal populations.
Synaptic currents: Isyn
Parameters
Synapses

gsyn

Vsyn

Ap

Bp

w1

w2

w3

ISTN GPe
ISTN GPi
IGPi TH
IGPe STN
IGPe GPe
IGPe GPi

0.15
0.15
0.11
0.5
0.5
0.5

0
0
85
85
85
85

2
2
2

0.04
0.04
0.04

20
20
20

57
57
57

2
2
2

0.234
0.234
0.234

0.4
0.4
0.4

0.04
0.04
0.04

stimulation reaches to smaller values of EI than the spherical model which represents the better reliability of thalamus for the
information from SMC above 95%, i.e., EI is lower than 0.05. Particularly, as the SA increases more than SA = 2 mA, sharp decreases
for all the ve stimulation paradigms can be found, and with the further increasing of SA (more than SA = 3 mA) all the EI values
for the ve stimulation modes gradually reach to the steady state closely approximate to the zeros, which matches the reliability
of healthy state and represents the restoration of the ability to relay the information from SMC for the thalamus. As a whole,
the CBBP-DBS can effectively restore the ability of thalamus to relay the information from the cortex. More importantly, as the
stimulation intensity increasing the medially steering stimulation achieves the lowest value of EI corresponding to the optimal
control for the reliability of thalamus, which is qualitatively consistent with the experimental evidence.
3.3. The desynchronization of neuronal populations
It is known that under the condition of healthy state, the three main cells, STN, GPi and GPe within the basal ganglia thalamocortical model network electrophysiologically re irregularly and there are little correlations between the activities of different
neurons in the neuronal populations [53]. On the contrary, parkinsonian state is characterized by an increasing regularity of
the neuronal discharges, where each STN neuron res in a periodic tremor-like fashion and the GPe or GPi res in a similar
manner. The above properties are paralleled those in the original RT model [50] or the modied model [53]. Specially, experimental results have demonstrated that Parkinsons disease is also characterized by the excessive pathological beta-band synchronizations of STN, GPi and GPe neuronal populations [8,59,63,6568]). However, these parkinsonian dynamical properties,
especially the pathological synchronization, can be effectively eliminated by the application of DBS. In addition, because the
parkinsonian beta-band synchrony is essentially a synchrony of bursting, in order to quantitatively measure the desynchronizing
effect of the four directional steering DBS of anterior, posterior, lateral and medial paradigms, as well as the case of spherical
stimulation paradigm, on the STN, GPi and GPe neuronal populations [53], the standard deviation (SD) [8,86] of the slow variables rn from each of the STN, GPe and GPi neurons is used for the analysis [63], which can be formulated by the following
expression,

T
1 

=
( j ),

j=1


2
N
N

1 i
1 i

[rx ( j )]
rx ( j )
( j) = N
N
i=1

(25)

i=1

where i and j are the indexes of neurons and time steps, respectively. Totally N = 320 model neurons in each neuronal population
are numerically computed over the T time steps. rx is the slow variable of model neurons where the subscript x can be the STN,
GPe and GPi. As mentioned in the previous work [8], is the effective measurement for the spatiotemporal desynchronization of
neuronal populations, which can be used to reveal the desynchronizing effect of the ve various STN-DBS stimulation paradigms.
Specically, from Eq. (25) we can know that the larger values of correspond to the higher level of desynchronization of the
neuronal network.
It can be seen from Fig. 9 that without stimulations, i.e., SA = 0 mA, the system is in the parkinsonian state and the SD of
the STN (Fig. 9(a)), GPe (Fig. 9(b)) and GPi (Fig. 9(c)) neuronal populations are all smaller values indicated by the red squares,
which implies the high-level of bursting synchronizations. However, as the stimulation is introduced, the values of SD can be
immediately and evidently elevated. That means the bursting syschronizations of parkinsonian state can be effectively improved.
Particularly, with the increasing of stimulation amplitudes, the SD of the STN, GPe and GPi neuronal populations can be gradually
increased, representing the gradual improvement of the pathological bursting behaviors. Therein, as SA is taken small values, e.g.,
SA 2 mA in Fig. 9(a), SA 0.5 mA in Fig. 9(b) and SA 1 mA in Fig. 9(c), the value of SD in the spherical stimulation paradigms is
slightly larger than those of the other four directional steering stimulations cases. This may be due to the fact that the small SA can
not suciently affect other non-targeted regions, but can form the larger stimulation eld than any other four directional steering
cases, hence displaying better desynchronizing effect. However, as the SA is increased beyond these critical values the SD for

232

D. Fan et al. / Commun Nonlinear Sci Numer Simulat 36 (2016) 219237

(a) STN

Standard Deviation, (r )

0.35
ANT
POST
LAT
MED
SPH
PD

0.3
0.25
0.2
0.15
0.1
1 mA

0.05
0
0

Stim Amp (mA)

(b) GPe

Standard Deviation, (r )

0.35
ANT
POST
LAT
MED
SPH
PD

0.3
0.25
0.2
0.15
0.1
1.5 mA

0.05
0
0

Stim Amp (mA)

(c) GPi
ANT
POST
LAT
MED
SPH
PD

0.35

Standard Deviation, (r )

0.4

0.3
0.25
0.2
0.15
0.1
1.5 mA

0.05
0
0

Stim Amp (mA)

Fig. 9. Effect of stimulation amplitude (SA), corresponding to the Istim in Eq. (21), on the standard deviation (SD) of the (a) STN, (b) GPe and (c) GPi neuronal
populations under ve type of anterior (ANT: red solid circle with blue edge), posterior (POST: blue asterisk), lateral (LAT: green square with blue edge), medial
(MED: inverted black solid triangular with blue edge) and spherical (SPH: yellow ve-pointed with blue edge) stimulation paradigms. The red square on the
SA = 0 indicates the SD corresponding to the parkinsonian state. Stimulation parameters are as mentioned in Fig. 8 with 130 Hz and 0.06 ms, and SA is varied
in [0.5, 8] at a xed step of 0.5 mA. Light dashed green and blue vertical bars indicate the stimulation intensity, Istim , where the maximal SD, i.e., optimal
desynchronizing effect can be achieved for the spherical and directional steering stimulation paradigms, respectively. Accordingly, red double-arrows represent
the differences between the optimal stimulation intensity. (For interpretation of the references to color in this gure legend, the reader is referred to the web
version of this article.)

D. Fan et al. / Commun Nonlinear Sci Numer Simulat 36 (2016) 219237

233

the directional steering stimulation paradigms will largely (Fig. 9(a), corresponding to the STN neuronal population) or slightly
(Figs. 9(b) and (c) corresponding to GPe and GPi neuronal populations) exceed that of the spherical case soon. In particular,
as SA is increased into SA = 3 mA, as shown in Fig. 9(a), the SD of the STN neuronal population for the spherical stimulation
paradigm rstly reaches to maximum value indicated by the dashed green vertical bar, preceding those of the directional steering
ones. Thereafter, with the SA further increasing the desynchronizing effect of spherical stimulation can be gradually weakened
accompanied by gradually decreasing in SD. This may indicate the occurrence of side effect for the spherical stimulation when the
stimulation amplitude is larger than SD = 3 mA. By contrast, after SA = 3 mA and before SA = 4 mA, the steering stimulations can
further enhance the desynchronizing effect of STN neuronal populations along with the further increasing values of SD. And when
SA = 4 mA indicated by the dashed light blue bar, the SD for the steering stimulation paradigms reach to the maximum values.
Similarly, after SA = 4 mA side effect may be induced by the steering stimulation paradigms and as the SA further increasing
the SD of the STN neuronal population can be sharply decreased. Overall, the stimulation parameter region for the directional
steering stimulation is larger than that of the spherical case. And the difference between the stimulation amplitudes indicated
by the dashed light green and blue bars is 1 mA (see the red double arrow), which represents the maximum values of SD, i.e., the
optimal desynchronizing effect of the spherical and directional steering stimulation paradigms, respectively. This is consistent
with the experimental evidence that there is at least one steering direction that allowed increasing the threshold for side effects
of 1 mA [1]. The reason may be that increasing SA can gradually enhance the spherical stimulation eld larger than those of
steering cases, which can more strongly inuence the adjacent regions apart from the targeted regions.
The similar scenario can be found in the Fig. 9(b) and (c), corresponding to the GPe and GPi neuronal populations, respectively,
apart from the more uctuations on the elevating phases of SD. More importantly, the larger threshold differences for the side
effects of 1.5 mA can be found for the GPe and GPi, where the optimal stimulating amplitudes for the spherical and the steering
stimulation paradigms are SA = 1.5 and 3 mA corresponding to the GPe, and SA = 2.5 and 4 mA corresponding to the GPi,
respectively. In addition, for the spherical stimulation mode GPe and GPi respectively achieve the larger values of SD on the
smaller optimal stimulation amplitudes, i.e., SA = 1.5 mA (Fig. 9(b)) and 2.5 mA (Fig. 9(c)), compared to the case of STN where
SA = 3 mA. For the directional steering stimulation paradigms, STN, GPe and GPi achieve the comparable values of SD on the
respective optimal stimulation amplitudes, i.e., SA = 4 (Fig. 9(a)), SA = 3 (Fig. 9(b)) and SA = 4 (Fig. 9(c)), respectively. Also,
with the SA increasing large enough, e.g., more than SA = 6.5, 4 and 4.5 respectively corresponding to the Fig. 9(a)(c), without
exception, the SD of all the STN, GPe and GPi neuronal populations reaches once again into the steady parkinsonian states.
3.4. Local eld potential (LFP) and dominant frequency (DF) for the STN neuronal population
The novel 32-contact electrode cannot only perform directional steering stimulation through selecting and activating the
specically located contacts, but can also record the electronic signals from the local structures of interest within the stimulation
targets and touched by specic contacts [1]. This provides the approach to record the local eld potentials (LFP) for the application
of adaptive feedback control and investigate the local physiological and pathological activities of neuronal populations. Isyn in
Eq. (1) represents the internal synaptic coupling within the neuronal population, and Idbs denotes an external stimulation. The
neurons interact via synapses by means of the PSP sp in Eq. (17) which is triggered by a spike of neuron p and modeled by the
Eq. (18). The normalized sum of the PSPs sp (t) describes the collective synaptic activity of the neuronal population, which yields
the local eld potential (LFP) [87,88]. In particular, STN LFP has its synaptic origin and is locally generated [87,88]. As seen from
the Fig. 4 there are no connections between STN neurons and the origin of STN LFP is in the pallido-subthalamic (i.e., GPe STN)
synaptic transmission. What is more, the model LFP is measured at the same site where stimulation is also applied, hence the
k
be the inhibitory synaptic input from GPes to the kth
stimulation current Idbs is should be added to LFP of STN cells. Let IGPeST
N
k
STN neuron and IdbsST N be the DBS current to the kth STN neuron, hence the model kth STN LFP can be expressed as follow,
k
k1
k+1
k
LF P k (ST N ) = IGPeST
N w (IGPeST N + IGPeST N ) + IdbsST N ,

(26)

Table 4
All the parameters used for the applied currents, stimulation currents and the cortical sensorimotor input currents administrated on the various neurons.
Applied currents: Iapp (STN,GPe,GPi)
Cells
Conditions

TC

STN

GPe

GPi

Healthy
Parkinsonian

33 A/cm2
23 A/cm2

21 A/cm2
8 A/cm2

22 A/cm2
16 A/cm2

Stimulation currents: IStim (STN,GPe,GPi)

Tp
IStim
[0.5, 8mA]
1000/130 ms

v1

v2

0.06 ms

1.001

1.0001

Cortical input currents: ISMC (TH)

SMC
ISMC
1000/ (1/CV2 , AR/A)ms
3.5 A/cm2

SMC
5 ms

CV
0.2

AR
14 Hz

234

D. Fan et al. / Commun Nonlinear Sci Numer Simulat 36 (2016) 219237

where w (= 0.4 in this paper) is the weight representing the effect of remote synaptic activity and the negative sign indicates
the inhibitory function from GPes. The LFP can be used to detect synchronization where low-amplitude LFP oscillation is the
characteristic of desynchronized dynamics [87,88]. In particular, the STN neurons among the STN neuronal populations with
coupling- and stimulation-free regime can independently re at different times.

Fig. 10. (a) Local eld potential (LFP) (mV) and (b) dominant frequency (DF) (Hz) simulated with the 320 STN neurons around 32 contacts of the novel electrode
[1]. Each LFP (a) or DF (b) is averaged by every 10 STN neurons around each contact. Totally, 32 LFPs (a) or DFs (b) have been numerically computed, per depth
and across all directions of the 32-contact electrode, and under the conditions of (according to top-to-bottom and left-to-right ): anteriorly, posteriorly, laterally,
medially and spherically steering stimulations. (For interpretation of the references to color in this gure legend, the reader is referred to the web version of this
article.)

D. Fan et al. / Commun Nonlinear Sci Numer Simulat 36 (2016) 219237

235

In addition, the power spectral analysis is used to estimate the dominant frequency (DF) of neural oscillations [89,90]. To do
this, the power spectral density is rstly obtained from the time series of each STN cell by using the fast Fourier transform (FFT).
Then, the maximum peak frequency is dened as the DF of neural oscillations. Therefore, 32 DFs are calculated with each one
being an average for all the 10 STN cells around each contact.
As shown in Fig. 9, directional steering stimulation obtained the optimal desynchronizing effect on around SA = 4 mA. Hence,
based on the modied RT computational model [50,53], we here present the LFP of STN neuronal population under the ve
different directional steering stimulations with the stimulation amplitude SA = 4 mA and numerically computed over per depth
and across all directions of the 32-contact electrode. This can reect the spatiotemporal information of LFP for the targeted STN
neuronal population touched by these 32 located contacts. Result shows that directional stimulation can induce the large LFP
differences between stimulation sites and their adjacent areas, as shown in the Fig. 10(a). Also, clear boundaries can be found
to distinguish the stimulation regions and their adjacent areas. The STN neurons receive the inhibitory synaptic input from
GPe, hence the STN LFP on the adjacent areas without stimulation has the negative but absolutely large values which implies
synchronizing dynamics. Whereas, the stimulations can drive the STN LFP into the low-amplitude oscillations around zero, which
corresponding to the weakened synchronization or desynchronization dynamics. Moreover, based on the experiment in [1], the
spectral powers of all simultaneous LFP recordings averaged per depth across all directions of the 32-contact electrode have been
studied, which shows that the evident power intensity lies in the 1340 Hz of beta frequency range and represents the regional
differences in pathological activity within the target nucleus, STN. Therefore, we also has simulated the dominant frequency
(DF) with being averaged by each subpopulation composed of 10 STN neurons around each contact. As shown in Fig. 10(b),
under various stimulation paradigms, the DF basically exhibits the beta (1340 Hz) oscillations, which are consistent with the
experimental results.
4. Conclusion
Deep brain stimulation (DBS) has increasingly become the conventional therapeutic method for the various neurodegenerative disorders such as Parkinsons disease, static tremor, and so on. Even though many conventional DBS strategies have been
adopted to explore the mechanisms underlying the effect of DBS, it is still unmanageable to modulate the stimulating parameters
applied in the DBS surgery. In addition, conventional DBS can yield a spherical eld which can inuence both the stimulation
targets and their adjacent non-target ganglia, hence inducing unsatisfactory treatment results. Recent experimental evidence for
patients with Parkinsons disease has shown that directional steering STN-DBS by a novel 32-contact electrode can effectively
control the DBS direction and improve the therapeutic effect compared to the conventional stimulation. More importantly, compared to the conventional spherical DBS, the novel 32-contact electrode can enlarge the therapeutic windows which can make
sure the safety and effectiveness of DBS therapy, and increase the choice of stimulation parameters.
To give a computational evidence for the novel directional steering DBS and also indirectly verify the effectiveness of existing
computational models related to the basal gangliathalamocortical motor circuits, based on the previous modied model [53],
originated from RT model [50], we computationally investigated the newest experimental ndings [1] with the stimulation
parameters as used in the experimental process [1]. In addition, parkinsonian symptoms are mainly characterized by the low
thalamic reliability and high-frequency bursting synchronous oscillations, so in this paper we considered the error index (EI) (see
Eq. (24)) of thalamus and the standard deviation (SD) (see Eq. (25)) of STN, GPi and GPe cells as the assessments of directional
DBS effect. Results have shown that directional steering stimulation is superior to the conventional spherical stimulation eld in
reducing the failure of thalamic transmission for the information from SMC and improving the desynchronization of neuronal
populations. In particular, the results show that directional steering stimulations can increase the optimal stimulation intensity
of desynchronization by more than 1 mA compared to the spherical case, which is consistent with the experimental result
that there exists at least one steering direction that can allow increasing the threshold of side effects by 1 mA. Furthermore,
we have recorded the local current signal by activating specic contacts on the 32-contact electrode to investigate the local
pathologic neuronal activity. Mainly, we have simulated the local eld potentials (LFP) and dominant frequency (DF) through
all 32 contacts of electrode yielding the spatiotemporal information of pathological neuronal behavior. We have found that
directional steering stimulation can induce the low-amplitude LFP which implies the occurrence of desynchronizing regime.
The distribution of DF in the 1340 Hz of beta frequency range has also been found, which is consistent with the experimental
records.
However, as seen in Fig. 8, the model considered in this text can still not completely model the evolution of EI for the thalamus consistent with the experiment, even though it can be qualitatively shown that directional steering stimulation is superior to the spherical case in improving the reliability of thalamus. Our computational investigation is just the tentative exploration for the mechanism underling the directional steering DBS. Further research is still needed to clarify more details of the
mechanism for the directional DBS and improve the computational model related to the basal gangliathalamocortical motor
circuits.
Acknowledgments
This research was supported by the National Science Foundation of China (Grants 11325208, 11572015 and 11172017), and by
the Innovation Foundation of BUAA for Ph.D Graduates.

236

D. Fan et al. / Commun Nonlinear Sci Numer Simulat 36 (2016) 219237

References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
[66]
[67]
[68]
[69]
[70]
[71]
[72]
[73]
[74]
[75]
[76]

Contarino MF, Bour LJ, Verhagen R, Lourens MA, de Bie RM, van den Munckhof P, et al. Neurology 2014;83:1163.
St George RJ, Nutt JG, Burchiel KJ, Horak FB. Neurology 2010;75:1292.
Grill WM, Snyder AN, Miocinovic S. Neuroreport 2004;15:1137.
Khn AA, Kempf F, Brcke C, Doyle LG, Martinez-Torres I, Pogosyan A, et al. J Neurosci 2008;28:6165.
Rosin B, Slovik M, Mitelman R, Rivlin-Etzion M, Haber SN, Israel Z, et al. Neuron 2011;72:370.
Zhang HH, Wang QY, Chen GR. Chaos 2014;24:033134.
Little S, Pogosyan A, Neal S, Zavala B, Zrinzo L, Hariz M. Ann Neurol 2013;74:449.
Fan DG, Wang QY. J Theor Biol 2015;370:157.
Tass PA. Biol Cybern 2003;89:81.
Lysyansky B, Popovych OV, Tass PA. Front Neuroeng 2013;6:5.
Popovych OV, Tass PA. Front Neurol 2014;5:268.
Adamchic I, Hauptmann C, Barnikol UB, Pawelczyk N, Popovych O, Barnikol TT, et al. Mov Disord 2014;29:1679.
Tass PA, Qin L, Hauptmann C, Dovero S, Bezard E, Boraud T, et al. Ann Neurol 2012;72:816.
Tass PA, Silchenko AN, Hauptmann C, Barnikol UB, Speckmann EJ. Phys Rev E 2009;80:011902.
Hauptmann C, Popovych O, Tass PA. Biol Cybern 2005;93:463.
Batista CAS, Lopes SR, Viana RL, Batista AM. Neural Netw 2010;23:114.
Batista CAS, Viana RL, Ferrari FAS, Lopes SR, Batista AM, Coninck JCP. Phys Rev E 2013;87:042713.
Beuter A, Vasilakos K. Chaos 1995;5:35.
Obeso JA. Trends Neurosci 2000;23:S8.
Albada SJV, Robinson PA. J Theor Biol 2009;257:642.
Bar-Gad I, Morris G, Bergman H. Prog Neurobiol 2003;71:439.
Garcia L, DAlessandro G, Bioulac B, Hammond C. Trends Neurosci 2005;28:209.
Janssen MLF, Duits AA, Tourai AM, Ackermans L, Leentjes AFG, van Kranen-Mastenbroek V, et al. Stereotact Funct Neurosurg 2014;92:381.
Pahwa R, Wilkinson S, Smith D, Lyons K, Miyawaki E, Wc K. Neurology 1997;49:249.
Volkmann J, Sturm V, Weiss P, Kappler J, Voges J, Koulousakis A. Ann Neurol 1998;44:953.
Ogura M. Stimulation of globus pallidus. Springer International Publishing; 2015.
Shih LC, Lafaver K, Lim C, Papavassiliou E, Tarsy D. Parkinsonism Relat Disord 2013;19:676.
Miyagi Y. Thalamic stimulation for Parkinsons disease: clinical studies on DBS. Springer International Publishing; 2015.
Okun MS, Zeilman PR. Parkinsons disease: guide to deep brain stimulation therapy. National Parkinson Foundation; 2014.
Pa A, Camera F, Apollonio F, DInzeo G, Liberti M. Front Comput Neurosci 2015;9:00002.
Plaha P, Ben-Shlomo Y, Patel NK, Gill SS. Brain 2006;129:1732.
Tommasi G, Krack P, Fraix V, Bas JFL, Chabardes S, Benabid AL, et al. J Neurol Neurosurg Psychiatry 2008;79:813.
Hariz M. Parkinsonism Relat Disord 2014;20:S192.
Martens HCF, Toader E, Decr MMJ, Anderson DJ, Vetter R, Kipke DR, et al. Clin Neurophysiol 2011;122:558.
Toader E, Decr MMJ, Martens HCF. Proceedings of the 2010 annual conference of IEEE on Engineering in Medicine Biology Society 2010:2061.
Firszt JB, Koch DB, Downing M, Litvak L. Otol Neurotol 2007;28:629.
Chaturvedi A. Development of accurate computational models for patient-specic deep brain stimulation. Case Western Reserve University; 2012.
Magara A, Bhler R, Moser D, Kowalski M, Pourtehrani P, Jeanmonod D. J Ther Ultrasound 2014;2:11.
Dallapiazza R, Khaled M, Eames M, Snell J, Lopes MB, Wintermark M, et al. Stereotact Funct Neurosurg 2015;93:140.
Liu C, Wang J, Yu H, Deng B, Wei X, Li H. Commun Nonlinear Sci Numer Simul 2015;28:10.
Modolo J, Mosekilde E, Beuter A. J Physiol Paris 2007;101:56.
Nahvi A, Alireza F, Hemmati S. J Mech Med Biol 2012;12:1240004.
Viana RL, Batista AM, Batista CAS, De Pontes JCA, Silva FADS, Lopes SR. Commun Nonlinear Sci Numer Simul 2012;17:2924.
Rubchinsky LL, Park C, Worth RM. Nonlinear Dyn 2012;68:329.
Schettino L, Rowat P, Erp EV, Song D, Poizner H. Applications of Nonlinear Dynamics Understanding Complex Systems. In: Nonlinear DDE analysis of
repetitive hand movements in parkinsons disease. Springer Berlin Heidelberg; 2009. p. 421.
Cai GL, Jiang SQ, Cai SM, Tian LX. Nonlinear Dyn 2015;80:503.
Beuter A, Edwards R, Titcombe MS. Data analysis and mathematical modeling of human tremor. New York: Springer; 2010.
Qin YM, Wang J, Men C, Chan WL, Wei XL, Deng B. Commun Nonlinear Sci Numer Simul 2013;18:2775.
Hauptmann C, Popovych O, Tass PA. Biol Cybern 2005;93:463.
Rubin JE, Terman D. J Comput Neurosci 2004;16:211.
Pascual A, Modolo J, Beuter A. J Integr Neurosci 2006;5:541.
Pirini M, Rocchi L, Sensi M, Chiari L. J Comput Neurosci 2009;26:91.
So RQ, Kent AR, Grill WM. J Comput Neurosci 2012;32:499.
Feng XJ, Shea-Brown E, Greenwald B, Kosut R, Rabitz H. J Comput Neurosci 2007;23:265.
Terman D, Rubin JE, Yew AC, Wilson CJ. J Neurosci 2002;22:2963.
Gorzelic P, Schiff SJ, Sinha A. J Neural Eng 2013;10:026016.
Steigerwald F, Petter M, Herzog J, Pinsker M, Kopper F, Mehdorn H, et al. J Neurophysiol 2008;100:2515.
Boraud T, Bezard E, Guehl D, Bioulac B, Gross C. Brain Res 1998;787:157.
Hammond C, Bergman H, Brown P. Trends Neurosci 2007;30:357.
Ferrari FA, Viana RL, Lopes SR, Stoop R. Neural Netw 2015;66:107.
Chu JU, Jeong MJ, Song KI, Lee HC, Kim J, Kin YJ, et al. Neurosci Res 2012;72:324.
Kubota S, Rubin JE, Robotics J. Netw Artif Life 2014;1:49.
Dovzhenok A, Park C, Worth RM, Rubchinsky LL. PloS ONE 2013;8:e58264.
Goldberg JA, Rokni U, Boraud T, Vaadia E, bergman H. J Neurosci 2004;24:6003.
George JS, Strunk J, Mak-McCully R, Houser M, Poizner H, Aron AR. NeuroImage Clin 2013;3:261.
Joundi RA, Brittain JS, Green AL, Aziz TZ, Brown P, Jenkinson N. Clin. Neurophysiol 2013;124:565.
Woerd ESt, Oostenveld R, de Lange FP, Praamstra P. NeuroImage 2014;100:507.
Smirnov DA, Barnikol UB, Barnikol TT, Bezruchko BP, Hauptmann C, Bhrle C, et al. Europhys Lett 2008;83:20003.
Hahn PJ, McIntyre CC. J Comput Neurosci 2010;28:425.
Levy R, Ashby P, Hutchison WD, Lang AE, Lozano AM, Dostrovsky JO. Brain 2002;125:1196.
Wilson CL, Cash D, Galley K, Chapman H, Lacey MG, Stanford IM. Neuroscience 2006;143:565.
Bergman H, Feingold A, Nini A, Raz A, Slovin H, Abeles M, et al. Trends Neurosci 1998;21:32.
Guo Y, Rubin JE, McIntyre CC, Vitek JL, Terman D. J Neurophysiol 2008;99:1477.
Sohal VS, Huntsman M, Huguenard JR. J Neurosci 2000;20:1735.
Sohal VS, Huguenard JR. Neurocomput 2002;44:653.
Rinzel J. Fed Proc 1985;44:2944.

D. Fan et al. / Commun Nonlinear Sci Numer Simulat 36 (2016) 219237

[77]
[78]
[79]
[80]
[81]
[82]
[83]
[84]
[85]
[86]
[87]
[88]
[89]
[90]

Dasanayake IS, Li JS. Syst Control Lett 2015;75:124.

Shendkar C, Lenka PK, Biswas A, Kumar R, Mahadevappa M. Healthc Technol Lett 2015;2:129.
Thurgood BK, Warren DJ, Ledbetter NM, Clark GA, Harrison RR. IEEE Trans Biomed Circuits Syst 2009;3:405.
Cappaert NL, Ramekers D, Martens HC, Wadman WJ. Int J Neural Syst 2013;23:1250031.
Johnson MD, McIntyre CC. J Neurophysiol 2008;100:2549.
Smith Y, Hazrati LN, Parent A. J Comp Neurol 1990;294:306.
Parent A, Smith Y, Filion M, Dumas J. Neurosci Lett 1989;96:140.
Deniau JM, Degos B, Bosch C, Maurice N. Eur J Neurosci 2010;32:1080.
Taylor PN, Baier G. J Comput Neurosci 2011;31:679.
Gao Z, Hu B, Hu G. Phys Rev E 2001;65:016209.
Popovych OV, Tass PA. Front Hum Neurosci 2012;6.
Park C, Worth RM, Rubchinsky LL. Phys Rev E 2011;83:042901.
Li W, Yang C. Chin J Med Instrum 2015;39:79.
Nakatani Y, Kumagai K, Minami K, Nakano M, Inoue H, Oshima S. Heart Rhythm 2015;12:257.

237