Citation: Hamed Azizollahi, Marion Darbas, Mohamadou M. Diallo, Abdellatif El Badia, Stephanie Lohrengel. EEG in neonates: Forward modeling and sensitivity analysis with respect to variations of the conductivity[J]. Mathematical Biosciences and Engineering, 2018, 15(4): 905-932. doi: 10.3934/mbe.2018041
[1] | Saadullah Farooq Abbasi, Qammer Hussain Abbasi, Faisal Saeed, Norah Saleh Alghamdi . A convolutional neural network-based decision support system for neonatal quiet sleep detection. Mathematical Biosciences and Engineering, 2023, 20(9): 17018-17036. doi: 10.3934/mbe.2023759 |
[2] | Krzysztof Fujarewicz, Krzysztof Łakomiec . Adjoint sensitivity analysis of a tumor growth model and its application to spatiotemporal radiotherapy optimization. Mathematical Biosciences and Engineering, 2016, 13(6): 1131-1142. doi: 10.3934/mbe.2016034 |
[3] | Sakorn Mekruksavanich, Wikanda Phaphan, Anuchit Jitpattanakul . Epileptic seizure detection in EEG signals via an enhanced hybrid CNN with an integrated attention mechanism. Mathematical Biosciences and Engineering, 2025, 22(1): 73-105. doi: 10.3934/mbe.2025004 |
[4] | Süleyman Cengizci, Aslıhan Dursun Cengizci, Ömür Uğur . A mathematical model for human-to-human transmission of COVID-19: a case study for Turkey's data. Mathematical Biosciences and Engineering, 2021, 18(6): 9787-9805. doi: 10.3934/mbe.2021480 |
[5] | DongMei Li, Rui-xue Zhang, Qian Xie, Qi Wang . Mathematical model for treatment of neonatal hyperbilirubinemia. Mathematical Biosciences and Engineering, 2021, 18(6): 8758-8782. doi: 10.3934/mbe.2021432 |
[6] | Ning Huang, Zhengtao Xi, Yingying Jiao, Yudong Zhang, Zhuqing Jiao, Xiaona Li . Multi-modal feature fusion with multi-head self-attention for epileptic EEG signals. Mathematical Biosciences and Engineering, 2024, 21(8): 6918-6935. doi: 10.3934/mbe.2024304 |
[7] | Yujie Kang, Wenjie Li, Jidong Lv, Ling Zou, Haifeng Shi, Wenjia Liu . Exploring brain dysfunction in IBD: A study of EEG-fMRI source imaging based on empirical mode diagram decomposition. Mathematical Biosciences and Engineering, 2025, 22(4): 962-987. doi: 10.3934/mbe.2025035 |
[8] | Alberto M. Gambaruto, João Janela, Alexandra Moura, Adélia Sequeira . Sensitivity of hemodynamics in a patient specific cerebral aneurysm to vascular geometry and blood rheology. Mathematical Biosciences and Engineering, 2011, 8(2): 409-423. doi: 10.3934/mbe.2011.8.409 |
[9] | Zizhuo Wu, Qingshan She, Zhelong Hou, Zhenyu Li, Kun Tian, Yuliang Ma . Multi-source online transfer algorithm based on source domain selection for EEG classification. Mathematical Biosciences and Engineering, 2023, 20(3): 4560-4573. doi: 10.3934/mbe.2023211 |
[10] | Tianfang Hou, Guijie Lan, Sanling Yuan, Tonghua Zhang . Threshold dynamics of a stochastic SIHR epidemic model of COVID-19 with general population-size dependent contact rate. Mathematical Biosciences and Engineering, 2022, 19(4): 4217-4236. doi: 10.3934/mbe.2022195 |
Electroencephalography (EEG) is a non-invasive functional brain imaging technique. EEG measures the electrical activity of the brain recorded by electrodes on the scalp and, more precisely, the voltage potential fluctuations between different cortical regions on the scalp. The recorded electrical activity is the synchronous activity of a large number of neighboring well-oriented neurons in the cerebral cortex beneath the skull.
The important goal of functional brain imaging using EEG is to localize cerebral sources generating measured EEG signals. The measurements provide valuable information about the sources that are at the origin of physological and pathological activities of the brain. In particular, EEG is one of the main diagnostic tests in presurgical evaluation for refractory epilepsy. Mathematical models are required to relate neural sources to EEG measurements. The accuracy of the EEG source reconstruction relies heavily on the accuracy of the associated forward model [1] which consists in computing the potential on the scalp for a given electrical source located in the brain.
On the one hand, the spherical multi-layer head model has gathered much interest from the beginning of EEG source analysis since an asymptotic formula for the potential is available [32,42]. On the other hand, realistic head models obtained from segmentation of magnetic resonance imaging (MRI) are able to take into account the geometrical complexity of the different tissues and their specific electrical characteristics. Several source models have been developed as e.g. partial integration, the St. Venant model or the subtraction approach [12,40,5,31,2]. For the numerical resolution of the forward problem, both boundary elements and 3D finite elements are commonly used [21,20,26,27]. In the head model for adults, the effect of anisotropy as well as the uncertainty in the tissue conductivities have been investigated [39]. All the aforementioned results are concerned with head models for adults or elderly children. In this paper, we are interested in an EEG model for neonates. The study is motivated by clinical questions. Few studies of neonates [35,18] exist due to the difficulties in creating realistic head models in neonates. Two characteristics are inherent to the neonate. The first one is that neonates show higher skull conductivities than children or adults [33]. The second one is the presence of fontanels in the skull. Fontanels are in the process of ossification and possess different electrical properties in comparison to the skull (cf. Figure 1.1). Clinicians wonder if it would be realistic to use the same EEG forward modeling for adults, early children and neonates. The underlying question is thus to quantify the impact of fontanels and the effects of uncertainty in the skull and fontanel conductivity on the values of the electric potential on the scalp. This investigation could be used directly for the development of realistic source localization methods from neonatal EEG.
From this motivation, the present work aims at proposing a mathematical EEG model for neonates and its numerical validation for both the multi-layer spherical head model and a realistic neonate head model. It may be noticed that the inhomogeneity of the skull conductivity prevents the application of boundary element methods which are currently used in commercial software for EEG source localization.
In order to answer clinical questions, we compare the proposed EEG model (with fontanels) and the one for adults (without fontanels). In neurophysiology studies, comparisons are commonly done via the computation of two error functionals, respectively the RDM (Relative Difference Measure) and the MAG (MAGnification factor). In this paper, we introduce an additional analysis tool to investigate the sensitivity of the potential solution of the problem with respect to the variations in the skull and fontanel conductivities. From a mathematical point of view, sensitivity is the directional derivative of the solution with respect to conductivity. It allows to analyze small variations in the tissue conductivities which currently occur from one patient to the other. Sensitivity analysis with respect to the conductivity can be employed to gather information about optimal design of the electrodes for EEG head caps [4,38]. In the present paper, sensitivity analysis is at the service of studying the effect of uncertainty in the knowledge of skull and fontanel conductivity on EEG measurements. Indeed, these parameters are not well known in the special case of neonates, and values of different orders of magnitude are proposed in the literature. We define in a rigorous way the mathematical setting of the sensitivity equation. Furthermore, the numerical discussion is performed on a realistic head model [3] obtained recently by the Inserm U1105 at Amiens' hospital. This model takes into account the fontanels and is more precise than previous ones [18].
The paper is organized as follows. In Section 2, we derive the EEG forward model. In Section 3, we present the subtraction approach to deal with the singularity of the source term and prove an existence and uniqueness result of a weak solution. In Section 4, we present a sensitivity analysis of the potential with respect to the conductivity. Section 5 is devoted to the numerical part: discretization, convergence analysis and various simulations. We discuss the validation of the EEG model, the impact of fontanels and the sensitivity of the potential with respect to a perturbation of the fontanel and/or skull conductivity.
In the low frequency range under consideration in EEG measurements, the electromagnetic field satisfies the quasi-static Maxwell equations where the time derivatives are neglected [22,17]. In terms of the electric field
∇⋅(εE)=ρ, | (2.1a) |
curlE=0, | (2.1b) |
curlH=J, | (2.1c) |
∇⋅(μH)=0. | (2.1d) |
Here,
J=σE+Js, | (2.2) |
where
E=∇u. | (2.3) |
Now, consider a bounded regular domain
−∇⋅(σ∇u)=∇⋅Js. | (2.4) |
The source model of neural activity can be described by a sum of
Sm≠Sp∀m≠pandqm≠0∀m,p∈{1,...,M}. | (2.5) |
The current density
Js=M∑m=1qmδSm |
where
Fdef=∇⋅Js=M∑m=1qm⋅∇δSm. | (2.6) |
In the typical multi-layer head model, we distinguish three to five layers for the brain (containing gray and white matters, cerebrospinal fluid (CSF)), skull, and scalp. Therefore, consider a partition of
¯Ω=L⋃i=1¯Ωi andΩi∩Ωj=∅ ∀i≠j. |
We assume the subdomains to be nested which means that
¯Ωi∩¯Ωj=∅ ∀j≠i−1,i,i+1. |
Then, for
This configuration includes the classical spherical model of three concentric spheres representing brain, skull and scalp (see Figure 2.1). Now, let
By considering that no electric current can flow out of the skull, the electric potential
{−∇⋅(σ∇u)=F,in Ω,σ∂nu=0,on Γ∞, | (2.7) |
where the source term
[u]|Γi=0 onΓi (i=1,…,L−1), | (2.8a) |
[σ∂nu]|Γi=0 onΓi (i=1,…,L−1). | (2.8b) |
Here and below, if
Problem (2.7) enters within the framework of standard elliptic problems, and its analysis is essentially covered by the classical results in variational theory. However, a careful formulation of these results (existence, uniqueness and regularity) in the context of a piecewise regular conductivity is of great importance. Indeed, the interest of studying the direct problem is twofold: on the one hand, these results are necessary to the sensitivity analysis with respect to the conductivity (Section 4), on the other, they allow the validation of the implementation of the method (Section 5.2).
Notice that a direct variational formulation of (2.7) in
The principle of the subtraction method is to decompose the potential
u=˜u+w, with ˜u=M∑m=1˜um. | (3.1) |
In order to define the singular potentials
σ1|Vm≡cm∈R,for anym∈{1,…,M}. | (3.2) |
Furthermore, since the locations
The singular potential
−cmΔ˜um=qm⋅∇δSm in R3, | (3.3) |
where
˜um(x)=14πcmqm⋅(x−Sm)|x−Sm|3, ∀x∈R3∖{Sm}. | (3.4) |
We see that the potential
In order to identify the problem satisfied by
∇⋅(σ∇˜um)=∇⋅((σ−cm)∇˜um)+cmΔ˜um. |
Indeed, both terms on the right hand side of the above identity are well defined as distributions on
−∇⋅(σ∇w)=−∇⋅(σ∇(u−˜u))=F+M∑m=1∇⋅((σ−cm)∇˜um)+cmΔ˜um on Ω. |
It follows from the definition of
{−∇⋅(σ∇w)=M∑m=1∇⋅((σ−cm)∇˜um)in Ω,σ∂nw=−σ∂n˜u,on Γ∞, | (3.5) |
Problem (3.5) can be reformulated on each subdomain
∇⋅((σ1−cm)∇˜um)=(σ1−cm)Δ˜um+∇(σ1−cm)⋅∇˜um. | (3.6) |
We thus get
−∇⋅(σ1∇w)=M∑m=1∇(σ1−cm)⋅∇˜um=∇σ1⋅∇˜u. |
On
{−∇⋅(σi∇w)=∇σi⋅∇˜u,in Ωi(i=1,…,L),σ∂nw=−σ∂n˜u,on Γ∞, | (3.7) |
with transmission conditions
[w]|Γi=0 onΓi,, | (3.8a) |
[σ∂nw]|Γi=(σi+1−σi)∂n˜u onΓi, | (3.8b) |
for
In this section, we give a variational formulation of problem (3.7)-(3.8) for the auxiliary function
L∑i=1∫Ωiσi∇w⋅∇vdx=L−1∑i=1∫Γi(σi+1−σi)∂n˜uvds−∫Γ∞σL∂n˜uvds+L∑i=1∫Ωi(∇σi⋅∇˜u)vdx. | (3.9) |
Since
∫Ωσ∇w⋅∇vdx=M∑m=1(∫Ω(cm−σ)∇˜um⋅∇vdx−∫Γ∞cm∂n˜umvds) | (3.10) |
which is the variational formulation of the boundary value problem (3.5). In the following, we focus on formulation (3.10). We introduce the bilinear form
a(w,v)=∫Ωσ∇w⋅∇vdx, | (3.11) |
as well as the linear form
l(v)=M∑m=1(∫Ω(cm−σ)∇˜um⋅∇vdx−∫Γ∞cm∂n˜umvds). | (3.12) |
Notice that
l(1)=M∑m=1∫Γ∞cm∂n˜um(x)ds=0. | (3.13) |
Condition (3.13) follows from the following classical lemma given in [16] that we recall for the convenience of the reader.
Lemma 1. Let
∫Γ∞cm∂n˜umds=0 ∀m=1,…,M. | (3.14) |
Note that a solution to (3.10) is unique only up to an additive constant. To this end, we introduce the following subspace of
V={v∈H1(Ω)|∫Ωvdx=0,} | (3.15) |
on which the Poincaré-Wirtinger inequality holds true,
‖v‖0,Ω≤CP‖∇u‖L2(Ω) ∀v∈V. | (3.16) |
In the sequel, we write
Theorem 1. Let
Find \,\,\,\,\,\,w\in V \,\,\,\,\,\,such\,\,\,\,that\,\,\,\,\,\,a(w, v) = l(v), \,\,\,\forall v\in H^1(\Omega) | (3.17) |
has exactly one solution
\| w \|_{H^1(\Omega)} \lesssim \sum\limits_{m = 1}^M \left( \| \nabla \widetilde u_m \|_{L^2(\Omega\setminus\mathcal{V}_m)} + \| \partial_\boldsymbol{n}\widetilde u_m \|_{L^2({\Gamma _\infty })} \right). | (3.18) |
Proof. It follows from standard arguments in variational theory that the bilinear form
|l(v)|\lesssim \sum\limits_{m = 1}^M\left(\|\nabla \widetilde u_m\|_{L^2(\Omega\backslash\mathcal{V}_m)}+C_T\|\partial_\boldsymbol{n}\widetilde u_m\|_{L^2({\Gamma _\infty })}\right)\|v\|_{H^1(\Omega)}, | (3.19) |
where
a(w, v) = l(v)\ \forall v\in V. |
Next, let
l(v) = l(v-v_\Omega) = a(w, v-v_\Omega) = a(w, v) |
which proves that
Remark 2. Notice that the linear form
The following theorem states the global
Theorem 3. In addition to the assumptions of Theorem 1, assume that
\| w \|_{H^2(\Omega_i)} \lesssim \sum\limits_{m = 1}^M \left( \| \nabla\widetilde u_m \|_{H^1(\Omega_i\setminus\mathcal{V}_m)} + \|\partial_\boldsymbol{n}\widetilde u_m\|_{H^1({\Gamma _\infty })}\right). | (3.20) |
The proof of Theorem 3 relies on standard techniques for elliptic partial differential equations. Indeed, we may notice that on each
-\nabla\cdot( \sigma_i\nabla w) = f_i |
with
f_i \stackrel{\rm def}{ = } \sum\limits_{m = 1}^M \nabla\cdot( (\sigma_i - c_m)\nabla\widetilde u_m) = \sum\limits_{m = 1}^M \nabla(\sigma_i-c_m)\cdot\nabla\widetilde u_m. |
According to the assumptions on
Sensitivity indicates the behavior of the potential when there is a slight variation of physical parameters. Here, we are interested in the sensitivity with respect to conductivity. This permits to measure the effects of uncertainty in the skull and fontanel conductivity on the model. Mathematically, a rigorous way to describe sensitivity is given by Gâteaux differentiability which expresses a weak concept of derivative.
Definition 4. Let
D_\mu F(\sigma) = \lim\limits_{h\to 0} \frac{F(\sigma+h\mu) -F(\sigma)}{h} |
if the limit exists. If
Now, let
\mathcal{P} = \left\{ {{\sigma\in L^\infty(\Omega)}|{\sigma_{|\mathcal{V}_m}\equiv {\rm const.}\, \forall m = 1, \ldots, M}} \right\} |
as well as the (open) subset
\mathcal{P}_\text{adm} = \left\{ {{\sigma\in \mathcal{P}}|{ \sigma_{\min} < \sigma < \sigma_{\max} }} \right\} |
of admissible conductivities. Here,
\widetilde u_m(\boldsymbol{x}, \sigma) = \frac{1}{4\pi c_m}\boldsymbol{q}_m\cdot\frac{(\boldsymbol{x}-S_m)}{|\boldsymbol{x}-S_m|^3} | (4.1) |
is the solution of the Poisson equation
-c_m\Delta\widetilde u_m(\cdot, \sigma) = \boldsymbol{q}_m\cdot\nabla \delta_{S_m} \ {\rm in}\ \mathbb{R}^3 |
where
\widetilde u_m(\boldsymbol{x}, \sigma+ h\mu) = \frac{1}{4\pi(c_m+h p_m)} \boldsymbol{q}_m\cdot\frac{(\boldsymbol{x}-S_m)}{|\boldsymbol{x}-S_m|^3}, | (4.2) |
where
-(c_m+h p_m)\Delta\widetilde u_m(\cdot, \sigma+h \mu) = \boldsymbol{q}_m\cdot\nabla \delta_{S_m}\ {\rm in}\ \mathbb{R}^3. |
The following proposition states that
Proposition 1. Let
\label{Dumtilde} D_\mu \widetilde u_m(\cdot, \sigma) = -\frac{p_m}{c_m} \widetilde u_m(\cdot, \sigma) | (4.3) |
with
Proof. A straightforward computation of the differential quotient yields
\label{um1tilde} \frac{\widetilde u_m(\boldsymbol{x}, \sigma+h\mu) - \widetilde u_m(\boldsymbol{x}, \sigma)}{h} = -\frac{p_m}{c_m+hp_m} \widetilde u_m(\boldsymbol{x}, \sigma), \ \forall \boldsymbol{x} \neq S_m | (4.4) |
and (4.3) follows. The right-hand side of (4.3) is obviously linear and continuous in
Theorem 5. Let
\int_{\Omega}\sigma\nabla w^1\cdot\nabla v\, d\boldsymbol{x} = -\int_{\Omega}\mu\nabla w\cdot\nabla v\, d\boldsymbol{x} \label{Vsens}\\ \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,+\sum\limits_{m = 1}^M\left(\int_{\Omega}(c_m-\sigma)\nabla \widetilde u_m^1\cdot\nabla v\, d\boldsymbol{x} -\int_{{\Gamma _\infty }}c_m\partial_{\boldsymbol{n}}\widetilde u_m^1v\, d\boldsymbol{x}\right)\nonumber \\\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, +\sum\limits_{m = 1}^M\left(\int_{\Omega}(p_m-\mu)\nabla \widetilde u_m \cdot\nabla v\, d\boldsymbol{x} -\int_{{\Gamma _\infty }}p_m\partial_{\boldsymbol{n}}\widetilde u_m v\, d\boldsymbol{x}\right), \nonumber | (4.5) |
for all
The sensitivity equation (4.5) is obtained by varying the conductivity parameter
-\nabla \cdot(\sigma\nabla w) = \nabla \cdot((\sigma-c_m)\nabla \widetilde u_m) |
and consider a perturbation of
-\nabla \cdot((\sigma+h\mu)\nabla w_h) = \nabla \cdot((\sigma+h\mu-c_m-hp_m)\nabla \widetilde u_{m, h}) |
where
-\nabla \cdot \left(\sigma\nabla \left( \frac{w_h-w}{h} \right)\right) = \nabla \cdot(\mu\nabla w_h) + \nabla \cdot\left((\sigma-c_m)\nabla \left( \frac{\widetilde u_{m, h}-\widetilde u_m}{h} \right)\right) |
+ \nabla \cdot((\mu-p_m)\nabla \widetilde u_{m, h}). |
At the limit
-\nabla \cdot(\sigma\nabla w^1) = \nabla \cdot(\mu\nabla w) + \nabla \cdot((\sigma-c_m)\nabla \widetilde u_m^1) + \nabla \cdot((\mu-p_m)\widetilde u_m). |
The boundary equation can be obtained in a similar way which yields finally the variational formulation (4.5). The full proof of Theorem 5 with a mathematical justification of the sense in which the limit has to be taken is given in Appendix A.
Remark 6. Assume that the variation of the conductivity in a given direction
w(\cdot, \sigma+h\mu) - w(\cdot, \sigma) = u(\cdot, \sigma+h\mu) - \tilde{u}(\cdot, \sigma_1) - (u(\cdot, \sigma) + \tilde{u}(\cdot, \sigma_1)) = u(\cdot, \sigma+h\mu) -u(\cdot, \sigma). |
Thus, the Gâteaux derivative
In this section, we address the discretization of problem (3.17). Boundary Element Methods (BEMs) have been very popular for a long time since they allow to reduce the three-dimensional problem to only boundary integral equations on the interfaces between the different tissues, see, for example [26,27,28] and references therein. However, BEMs are restricted to a piecewise constant conductivity representing in EEG the standard model of homogeneous tissues in an adult's head. Here, because of fontanels, BEMs are thus prohibited, and we propose a discretization by means of three-dimensional Lagrange Finite Elements (FE) which are able to handle complex geometries.
Problem (3.5), even if it seems to be rather standard, does not fit rigorously the standard assumptions in finite element analysis. For the convenience of the reader who could be less familiar with advanced finite element techniques, we address in this section the following points.
The first one is concerned with the discretization of the computational domain. With regard to the application in EEG and conforming to the assumptions in Section 2, the different subdomains
Further, we have to deal with the singularity of the source term. In [40], error estimates have been stated for the subtraction approach in the case of a single subdomain and constant conductivity.
The aim of this section is to give a precise description of the discrete problem as well as a rigorous error analysis for the complex setting of the multi-layer model with variable and piecewise regular conductivity. From a practical point of view, this is an important issue for the validation of the numerical implementation.
Throughout this subsection, we assume that the regularity assumptions of Theorem 3 are fulfilled. Consider a family
\mathcal{T}_{h, i} = \left\{ {{T\in\mathcal{T}_h}{\mathcal{N}_h\cap T \subset\overline{\Omega}_i}\ \forall i = 1, \ldots, L. } \right\} |
Then, let
\Omega_{i, h} = \bigcup\limits_{T\in\mathcal{T}_{h, i}} T\ \mbox{and}\ \Gamma_{i, h} = \mathcal{N}_h\cap\Gamma_i. |
The following conditions state that
\bigcup\limits_{i = 1}^L \Omega_{i, h} = \Omega_h | (5.1a) |
\Gamma_{i, h} = \Omega_{i, h}\cap\Omega_{i+1, h}\ \forall i = 1, \ldots, L-1. \label{hyp:mesh2} | (5.1b) |
These assumptions guarantee that no element has nodes in the interior of two different subdomains. In order to formulate the discrete problem, we need to extend the functions
\overline{\Omega_i}\subset \widetilde \Omega_{i} \ \mbox{and}\ \Omega_{i, h}\subset \widetilde \Omega_{i} . |
Without loss of generality, we may assume that
On
X_h = \left\{ {{v_h\in\mathcal{C}^0(\overline{\Omega_h})}{v_{h|T}\in\mathbb{P}_1(T)\ \forall T\in\mathcal{T}_h}} \right\} | (5.2) |
where
Then the discrete problem reads: find
a_h(w_h, v_h) = l_h(v_h)\ \forall v_h\in V_h | (5.3) |
with
a_h(w_h, v_h) = \sum\limits_{i = 1}^L \int_{\Omega_{i, h}} \widetilde \sigma \nabla w_h\cdot\nabla v_h\, d\boldsymbol{x}\ \mbox{and}\ l_h(v_h) = \sum\limits_{i = 1}^L \int_{\Omega_{i, h}} \tilde{\boldsymbol{F}_i}\cdot\nabla v_h + \int_{{\Gamma _\infty }h} \tilde{g} v_h \, ds | (5.4) |
where
\tilde{\boldsymbol{F}_i} = \sum\limits_{m = 1}^M (c_m-\widetilde \sigma )\nabla \widetilde u_m \ \mbox{on}\ \widetilde \Omega_{i} \ \mbox{and}\ \tilde{g} = - \sum\limits_{m = 1}^M c_m\partial_\boldsymbol{n}\widetilde u_m \ \mbox{on}\ {\Gamma _\infty }h. |
Notice that
As for the continuous problem, existence and uniqueness of the solution of (5.3) follow from Lax-Milgram's theorem since
Definition 7. For a family of discrete problems (5.3), the bilinear forms
a_h(v_h, v_h) \ge \alpha \| v_h \|_{H^1(\Omega_h)}^2\ \forall v_h\in V_h. | (5.5) |
Notice that the family
Theorem 8. Consider a regular family of meshes
\forall K\in \bigcup\limits_h \mathcal{T}_h, \ \frac{h}{h_K} \le C_{\rm inv}. | (5.6) |
Let
Consider the family of discrete problems (5.3) and assume that the associated bilinear forms
Then, the following error estimate holds true,
\sum\limits_{i = 1}^L \| \tilde{w}_i-w_h \|_{H^1(\Omega_{i, h})} \lesssim h \sum\limits_{m = 1}^M \left( \| \nabla \widetilde u_m\|_{H^1(\Omega\setminus\mathcal{V}_m)} + \| \partial_\boldsymbol{n}\widetilde u_m\|_{H^1({\Gamma _\infty })} \right) \\ \, \, \, \, \, \, \, \, \, + h^{3/2} \sum\limits_{m = 1}^M \left( \| \partial_\boldsymbol{n}\widetilde u_m \|_{W^{1, \infty}({\Gamma _\infty }h)} + \| \nabla\widetilde u_m \|_{H^1({\Gamma _\infty }h)} \right). \nonumber | (5.7) |
The error estimate for the approximate solution
The result of Theorem 8 relies on an abstract error estimate which states that the discretization error may be estimated by the interpolation error with respect to
Such an abstract error estimate has been obtained in [8] in the case of one single domain and the result carries over to our configuration of multiple subdomains taking into account the extensions of the conductivity
In this paragraph, we address the numerical validation of the subtraction approach. From our motivation of studying EEG in neonates, special attention is paid to the presence of fontanels in the skull. As mentioned above, fontanels are tissues that are in the process of ossification. They are taken into account in our model through the definition of an appropriate variable skull conductivity.
Let us consider a three-layer spherical head model (see Figure 2.1) representing the brain, skull and scalp with respective radii
We consider a family of three tetrahedral meshes with decreasing mesh size
Mesh | Nodes | Tetrahedra | Boundary nodes | ||
Two criteria are commonly used in the numerical validation of EEG models [40]. The first, called the Relative Difference Measure (RDM), is computed as follows
\label{RDM} \mbox{RDM} : = \left\| \dfrac{u_h}{\| u_h\|_{L^2(\Gamma_{\!\infty, h})}} - \dfrac{u_{\tiny{\mbox{ref}}} }{\| u_{\tiny{\mbox{ref}}} \|_{L^2(\Gamma_{\!\infty, h})}} \right\|_{L^2(\Gamma_{\!\infty, h})}. | (5.8) |
The second is the magnification factor (MAG) which is defined by
\label{MAG} \mbox{MAG}: = \left| 1 - \dfrac{\| u_h\|_{L^2(\Gamma_{\!\infty, h})}}{\|u_{\tiny{\mbox{ref}}}\|_{L^2(\Gamma_{\!\infty, h})}} \right|. | (5.9) |
The RDM and MAG are error functionals with respect to a reference solution
Next, we take into account the main fontanel, i.e. the anterior fontanel situated between the frontal and parietal bones (cf Figure 1.1). The inclusion of the main fontanel in the three-layer spherical model is performed by the following function defined in the subdomain
\sigma_2(\boldsymbol{x}) = \sigma_{skull}+(\sigma_{\!f}-\sigma_{skull})g(\boldsymbol{x}). | (5.10) |
The parameter
g(\boldsymbol{x}) = e^{-\alpha (x_1^2 + x_2^2)} | (5.11) |
depending on the parameter
The numerical validation is performed for different configurations. Notice that no analytical solution is available for the spherical model with fontanels. Numerical solutions are therefore compared with a numerical reference one,
\| u_h - u_{\tiny{\mbox{ref}}} \|^2_{H^1(\Omega)} \stackrel{\tiny \mbox{def}}{ = } \| w_h - w_{\tiny{\mbox{ref}}} \|^2_{H^1(\Omega)}. |
Figure 5.3 shows two convergence curves in logarithmic scale of the relative error in the
We report in Figure 5.4 the RDM and MAG coefficients for different source positions and mesh sizes. Conclusions are the same as those obtained for the spherical head model without fontanels. The factors RDM and MAG keep under 1.5% and 0.5% respectively for all meshes and eccentricities, and decrease with the mesh size. This validates the subtraction method in the case of the spherical head model with the anterior fontanel.
One notices in Figure 5.1 and Figure 5.4 that the error increases with the eccentricity. Indeed, a careful analysis of the right hand side in the error estimate (5.7) in the case where
\delta = \max\limits_{m = 1:M} {\rm dist}(S_m, \Gamma_1) |
which is related to the eccentricity
e = 1 - \frac{\delta}{r_1}. |
This deterioration in the precision of the simulated data is inherent to the numerical method and dependent on the mesh that is used in the simulations.
In this section, we investigate numerically the effects of fontanels and their conductivity on the electric potential measured at the scalp with following question in mind: is it important or not to consider fontanels in the EEG model in neonates?
We consider a realistic head model of a healthy full-term newborn obtained from coregistration of MR and CT images of the Amiens' hospital database (see Figure 6.1). The diameter of the computational domain is about 12cm. Obtaining a realistic model was necessary for our numerical study. Nevertheless, neonatal cerebral image segmentation is a challenging task due to inherent difficulties such as low signal to noise ratio and partial volume effects. Segmentation of adult MR images has been extensively addressed by many researchers, and several methods for automatic segmentation have been developed in the case of healthy adults. However, due to insufficient anatomical similarity between neonatal and adult's images, these methods generally fail to segment accurately cerebral images of neonates. Extracting cranial bones and fontanels from neonatal MR images is also difficult and even unreliable in some cases. In this study, one has used the coregistration of MR and CT images of one healthy male neonate of 42 weeks gestational age at the time of the scan to construct a realistic volume conductor head model containing five different newborn's head compartments: brain (
Mesh | Nodes | Tetrahedra | Boundary faces | ||
108 669 | 590 878 | 55 660 |
We compare different EEG forward models. We use a panel of conductivities which are found in literature (e.g. [35,30,18,3]). To each couple
The different conductivity values are fixed to
We study numerically how a slight variation of the conductivity affects EEG measurements. We are interested in analyzing conductivity perturbations in neonatal skull including fontanels.
The sensitivity
Figure 6.4 compares the sensitivity of two sources of same moment
The sensitivity analysis of this section provides information on those scalp areas on which the electrical potential is affected by small variations of the fontanel's conductivity. These variations appear naturally from one patient to the other and even for the same patient at different ages. The results corroborate the conclusions of the previous section that the fontanel and skull conductivity values impact the EEG forward solution. This effect seems to be localized near the fontanels and depends on the eccentricity and orientation of the dipolar sources.
In this paper, we have proposed an EEG forward problem for neonates, that amounts to solve an elliptic problem with a singular source term in a an inhomogeneous medium. The considered domain is made up of disjoint coated regions, the source term is a finite linear combination of dipoles and the conductivities are functions of the position in each region. This last assumption allows to take into account the presence of fontanels in the skull and their ossification process. In this context, a rigorous mathematical framework has been proposed. In order to deal with the singularity of the source term, we apply a subtraction method and give a proof for existence and uniqueness of the weak solution in the context of variable conductivities. Convergence estimates are obtained for standard finite elements of type P1.
Numerical validations were performed. Firstly, the numerical validation has been carried out for the three-layer spherical head model with or without fontanels. For a given source term, the error of the regular part of the solution has been computed in the
Nevertheless, in all tested configurations the error keeps lower than 1.5% even for the coarsest mesh and the most eccentric source position. Higher quadrature rules could help to overcome the influence of the singular behavior of the source term.
Secondly, with the motivation of answering clinical questions, we have studied the influence of fontanels and uncertainty in fontanel and skull conductivities on the electrical potential values at the scalp using a realistic neonatal head model. This model is provided by Inserm U1105 (Amiens' hospital) database. The RDM and MAG factors between models with or without fontanels depend on the ratio of the fontanel conductivity over the skull conductivity. The difference is more pronounced for a higher ratio. With regard to EEG source reconstruction, these numerical observations attest the importance of considering the presence of the fontanels in the EEG forward model for neonates. It shows also that uncertain conductivity values impact the EEG forward solution. These conclusions are confirmed by a mathematical and numerical sensitivity analysis. Furthermore, this study provides useful information on areas where the electrical potential is the most sensitive to a variation of conductivity. The support of the sensitivity function depends on the source characteristics (position and moment) and is localized to an area above the fontanels.
Our findings are comparable with those of Azizollahi et al. [3] who studied the influence of different tissue conductivities including the fontanels. Their approach was based on a model of distributed sources. We are actually working together with the group of GRAMFC at Amiens' hospital for a detailed study of these new aspects.
The analysis of the EEG forward model is an essential preliminary step to the resolution of the corresponding inverse source problem. In the head model of adults, theoretical and numerical results exist [9,13,14,16,29,10,4]. For the neonatal head model, identifiability and stability results for the inverse EEG source problem have been obtained in parallel to the present work [11]. A numerical study that should help to understand the impact of the presence of fontanels on the source reconstruction in neonates, especially for epileptogenic sources, is underway.
The present work has been realized as part of the MIFAC project receiving financial support from the region Picardie (now Hauts-de-France).
Proof of Theorem 5. Let
\int_{\Omega}\sigma\nabla w^1\cdot\nabla v\, d\boldsymbol{x} = -\int_{\Omega}\mu\nabla w\cdot\nabla v\, d\boldsymbol{x} \label{Vsens2}\\ \, \, \, \, \, \, +\sum\limits_{m = 1}^M\left(\int_{\Omega}(c_m-\sigma)\nabla \widetilde u_m^1\cdot\nabla v\, d\boldsymbol{x} -\int_{{\Gamma _\infty }}c_m\partial_{\boldsymbol{n}}\widetilde u_m^1v\, d\boldsymbol{x}\right)\nonumber \\ \, \, \, \, \, \, +\sum\limits_{m = 1}^M\left(\int_{\Omega}(p_m-\mu)\nabla \widetilde u_m \cdot\nabla v\, d\boldsymbol{x} -\int_{{\Gamma _\infty }}p_m\partial_{\boldsymbol{n}}\widetilde u_m v\, d\boldsymbol{x}\right), \nonumber | (A.1) |
for all
First of all, notice that
Then, in order to investigate the Gâteaux derivative of
In the sequel we shall omit the dependence of
w \stackrel{\rm def}{ = } w(\cdot, \sigma) \ {\rm and}\ w_h \stackrel{\rm def}{ = } w(\cdot, \sigma+h\mu) |
as well as
\widetilde u_m \stackrel{\rm def}{ = } \widetilde u_m(\cdot, \sigma)\ {\rm and}\ \widetilde u_{m, h}\stackrel{\rm def}{ = } \widetilde u_m(\cdot, \sigma+h\mu) |
Now, recall that
\label{Vsigh} \begin{array}{lll} \displaystyle \int_{\Omega}(\sigma+h\mu)\nabla w_h\cdot\nabla v\, d\boldsymbol{x}& = &\displaystyle \sum\limits_{m = 1}^M (\int_{\Omega}\left((c_m+h p_m)-(\sigma+h\mu)\right)\nabla \widetilde u_{m, h}\cdot\nabla v\, d\boldsymbol{x}\\ &-& \displaystyle \int_{{\Gamma _\infty }} (c_m+h p_m) \partial_{\boldsymbol{n}}\widetilde u_{m, h} v \, ds), \end{array} | (A.2) |
and
\int_{\Omega} \sigma\nabla w\cdot\nabla v\, d\boldsymbol{x} = \sum\limits_{m = 1}^M\left(\int_{\Omega}(c_m- \sigma)\nabla \widetilde u_m \cdot\nabla v\, d\boldsymbol{x} -\int_{{\Gamma _\infty }}c_m\partial_{\boldsymbol{n}}\widetilde u_m v \, ds \right), | (A.3) |
for all
\| w_h \|_{H^1(\Omega)} \lesssim \sum\limits_{m = 1}^M \left( \| \nabla \widetilde u_{m, h} \|_{L^2(\Omega\setminus\mathcal{V}_m)} + \| \partial_\boldsymbol{n}\widetilde u_{m, h}\|_{L^2({\Gamma _\infty })} \right). | (A.4) |
In order to identify the Gâteaux derivative of
w_h^1 = \, \frac{w_h - w}{h} \mbox { and } \, \widetilde u^1_{m, h} = \frac{\widetilde u_{m, h} - \widetilde u_m}{h}. |
Subtracting (A.3) from (A.2) and dividing by
\label{limwh1} \int_{\Omega} \sigma\nabla w_h^1 \cdot\nabla v\, d\boldsymbol{x} = - \int_{\Omega} \mu \nabla w_h \cdot\nabla v\, d\boldsymbol{x} \\ \, \, \, \, \, \, + \sum\limits_{m = 1}^M \left(\int_{\Omega} (c_m- \sigma)\nabla \widetilde u^1_{m, h} \cdot\nabla v\, d\boldsymbol{x} - \int_{{\Gamma _\infty }} c_m\partial_{\boldsymbol{n}} \widetilde u^1_{m, h} \, v \, ds \right) \nonumber\\ \, \, \, \, \, \, + \sum\limits_{m = 1}^M \left(\int_{\Omega} (p_m- \mu) \nabla \widetilde u_{m, h}\cdot\nabla v\, d\boldsymbol{x} - \int_{{\Gamma _\infty }} p_m \partial_{\boldsymbol{n}} \widetilde u_{m, h} v \, ds\right). \nonumber | (A.5) |
We compare the above formulation for the differential quotient
\label{wh1_w1} \int_{\Omega} \sigma\nabla \left( w_h^1 -w^1 \right) \cdot\nabla v\, d\boldsymbol{x} = - \int_{\Omega} \mu \nabla \left( w_h-w \right) \cdot\nabla v\, d\boldsymbol{x}\\ \, \, \, \, \, \, \, \, \, \, \, + \sum\limits_{m = 1}^M \left(\int_{\Omega} (c_m- \sigma)\nabla \left( \widetilde u^1_{m, h} -\widetilde u^1_m \right) \cdot\nabla v\, d\boldsymbol{x} - \int_{{\Gamma _\infty }} c_m\partial_{\boldsymbol{n}} \left( \widetilde u^1_{m, h}-\widetilde u^1_m\right) \, v \, ds \right) \nonumber\\ \, \, \, \, \, \, \, \, \, \, \, + \sum\limits_{m = 1}^M \left(\int_{\Omega} (p_m- \mu) \nabla \left( \widetilde u_{m, h} -\widetilde u_m\right)\cdot\nabla v\, d\boldsymbol{x} - \int_{{\Gamma _\infty }} p_m \partial_{\boldsymbol{n}} \left( \widetilde u_{m, h} -\widetilde u_m\right)\, v \, ds\right). \nonumber | (A.6) |
Notice that the integrals over
\label{estim_wh1w1} \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \| w_h^1 -w^1\|_{H^1(\Omega)}\\ \, \, \, \, \lesssim \| \nabla ( w_h-w )\|_{L^2(\Omega)}\nonumber \\ \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, + \sum\limits_{m = 1}^M \left( \left\|\nabla \left( \widetilde u^1_{m, h} -\widetilde u^1_m \right) \right\|_{L^2(\Omega\setminus\mathcal{V}_m)} + \left\| \partial_\boldsymbol{n} \left( \widetilde u^1_{m, h} -\widetilde u^1_m \right) \right\|_{L^2({\Gamma _\infty })} \right) \nonumber\\ \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, + \sum\limits_{m = 1}^M \left( \left\|\nabla \left( \widetilde u_{m, h} -\widetilde u_m \right) \right\|_{L^2(\Omega\setminus\mathcal{V}_m)} + \left\| \partial_\boldsymbol{n} \left( \widetilde u_{m, h} -\widetilde u_m \right) \right\|_{L^2({\Gamma _\infty })} \right) \nonumber | (A.7) |
The second and third term in the right-hand side of (A.7) are of order
\label{estim_whw} \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, h^{-1} \| w_h -w\|_{H^1(\Omega)} \nonumber\\ \, \, \, \, \, \, \, \, \, \, \, \lesssim \|\nabla w_h\|_{L^2(\Omega)} + \sum\limits_{m = 1}^M \left( \left\|\nabla \widetilde u^1_{m, h} \right\|_{L^2(\Omega\setminus\mathcal{V}_m)} + \left\| \partial_\boldsymbol{n} \widetilde u^1_{m, h} \right\|_{L^2({\Gamma _\infty })} \right) \\ \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, \, + \sum\limits_{m = 1}^M \left( \left\|\nabla \widetilde u_{m, h} \right\|_{L^2(\Omega\setminus\mathcal{V}_m)} + \left\| \partial_\boldsymbol{n} \widetilde u_{m, h} \right\|_{L^2({\Gamma _\infty })} \right). \nonumber | (A.8) |
The right hand side in (A.8) is obviously bounded when
\| w_h^1 -w^1\|_{H^1(\Omega)} \lesssim h \sum\limits_{m = 1}^M \left( \left\|\nabla \widetilde u_m \right\|_{L^2(\Omega\setminus\mathcal{V}_m)} + \left\| \partial_\boldsymbol{n} \widetilde u_m \right\|_{L^2({\Gamma _\infty })} \right) . |
This proves the strong convergence of the sequence
It remains to show that
In order to prove that the linear application
\|w^1\|_{H^1(\Omega)} \lesssim \|\mu\|_{\infty}\|\nabla w\|_{L^2(\Omega)} +\sum\limits_{m = 1}^M |\mu(S_m)| \left( \|\nabla \widetilde u_m\|_{L^2(\Omega\setminus\mathcal{V}_m)} + \|\partial_{\boldsymbol{n}}\widetilde u_m\|_{L^2({\Gamma _\infty })} \right) |
using again that
Lemma 2. Let
\left \|\nabla \left( \widetilde u_{m, h}-\widetilde u_m\right)\right \|_{L^2(\Omega\setminus\mathcal{V}_m)} \lesssim h||\nabla \widetilde u_m||_{L^2(\Omega\setminus\mathcal{V}_m)}, \label{estim1:a}\\ | (A.9a) |
\left \|\partial_{\boldsymbol{n}}\left( \widetilde u_{m, h}-\widetilde u_m\right)\right \|_{L^2({\Gamma _\infty })} \lesssim h||\partial_{\boldsymbol{n}}\widetilde u_m||_{L^2({\Gamma _\infty })}. \label{estim1:b} | (A.9b) |
Further, let
\left \|\nabla \left( \widetilde u_{m, h}^1-\widetilde u_m^1 \right)\right \|_{L^2(\Omega\setminus\mathcal{V}_m)} \lesssim h||\nabla \widetilde u_m||_{L^2(\Omega\setminus\mathcal{V}_m)}, \label{estim2:a}\\ | (A.10a) |
\left \|\partial_{\boldsymbol{n}}\left(\widetilde u_{m, h}^1-\widetilde u_m^1\right)\right \|_{L^2({\Gamma _\infty })} \lesssim h||\nabla \widetilde u_m||_{L^2({\Gamma _\infty })}. | (A.10b) |
Proof. From the definition of
\widetilde u_{m, h}(\boldsymbol{x}) - \widetilde u_m(\boldsymbol{x}) = \frac{1}{4\pi} \left( \frac{1}{c_m+hp_m} - \frac{1}{c_m} \right) \frac{\boldsymbol{q}_m\cdot(\boldsymbol{x}-S_m)}{|\boldsymbol{x}-S_m|^3} = -\frac{hp_m}{c_m+hp_m} \widetilde u_m(\boldsymbol{x}). | (A.11) |
Integration of the gradient of the above expression over
Next, recall that
\widetilde u_{m, h}^1 - \widetilde u_m^1 = \left( -\frac{p_m}{c_m+hp_m} + \frac{p_m}{c_m} \right) \widetilde u_m = h \frac{p_m^2}{c_m(c_m+hp_m)}\widetilde u_m. | (A.12) |
The boundedness of
[1] | [ Z. Akalin Acar,S. Makeig, Effects of Forward Model Errors on EEG Source Localization, Brain Topogrography, 26 (2013): 378-396. |
[2] | [ A. Alonso-Rodriguez,J. Camano,R. Rodriguez,A. Valli, Assessment of two approximation methods for the inverse problem of electroencephalography, Int. J. of Numerical Analysis and Modeling, 13 (2016): 587-609. |
[3] | [ H. Azizollahi,A. Aarabi,F. Wallois, Effects of uncertainty in head tissue conductivity and complexity on EEG forward modeling in neonates, Hum. Brain Ma, 37 (2016): 3604-3622. |
[4] | [ H. T. Banks,D. Rubio,N. Saintier, Optimal design for parameter estimation in EEG problems in a 3D multilayered domain, Mathematical Biosciences and Engineering, 12 (2015): 739-760. |
[5] | [ M. Bauer,S. Pursiainen,J. Vorwerk,H. Köstler,C. H. Wolters, Comparison Study for Whitney (Raviart-Thomas)-Type Source Models in Finite-Element-Method-Based EEG Forward Modeling, IEEE Trans. Biomed. Eng., 62 (2015): 2648-2656. |
[6] | [ J. Borggaard and V. L. Nunes, Fréchet Sensitivity Analysis for Partial Differential Equations with Distributed Parameters, American Control Conference, San Francisco, 2011. |
[7] | [ H. Brezis, Functional Analysis, Sobolev Spaces And Partial Differential Equations, Universitext. Springer, New York, 2011. |
[8] | [ P. G. Ciarlet, The Finite Element Method for Elliptic Problems, North Holland, New York, 1978. |
[9] | [ M. Clerc,J. Kybic, Cortical mapping by Laplace-Cauchy transmission using a boundary element method, Journal on Inverse Problems, 23 (2007): 2589-2601. |
[10] | [ M. Clerc, J. Leblond, J. -P. Marmorat and T. Papadopoulo, Source localization using rational approximation on plane sections, Inverse Problems, 28 (2012), 055018, 24 pp. |
[11] | [ M. Darbas, M. M. Diallo, A. El Badia and S. Lohrengel, An inverse dipole source problem in inhomogeneous media: application to the EEG source localization in neonates, in preparation. |
[12] | [ A. El Badia,T. Ha Duong, An inverse source problem in potential analysis, Inverse Problems, 16 (2000): 651-663. |
[13] | [ A. El Badia,M. Farah, Identification of dipole sources in an elliptic equation from boundary measurements, J. Inv. Ill-Posed Problems, 14 (2006): 331-353. |
[14] | [ A. El Badia and M. Farah, A stable recovering of dipole sources from partial boundary measurements, Inverse Problems, 26 (2010), 115006, 24pp. |
[15] | [ Q. Fang and D. A. Boas, Tetrahedral mesh generation from volumetric binary and grayscale images, EEE International Symposium on Biomedical Imaging: From Nano to Macro, (2009), SBI? 09. Boston, Massachusetts, USA, 1142–1145. |
[16] | [ M. Farah, Problémes Inverses de Sources et Lien avec l'Electro-encéphalo-graphie, Thése de doctorat, Université de Technologie de Compiégne, 2007. |
[17] | [ O. Faugeras, F. Clément, R. Deriche, R. Keriven, T. Papadopoulo, J. Roberts, T. Viéville, F. Devernay, J. Gomes, G. Hermosillo, P. Kornprobst and D. Lingrand, The Inverse EEG and MEG Problems: The Adjoint State Approach I: The Continuous Case ,Rapport de recherche, 1999. |
[18] | [ P. Gargiulo,P. Belfiore,E. A. Friogeirsson,S. Vanhalato,C. Ramon, The effect of fontanel on scalp EEG potentials in the neonate, Clin. Neurophysiol, 126 (2015): 1703-1710. |
[19] | [ D. Gilbarg and N. S. Trudinger, Elliptic Partial Differential Equations of Second Order Springer, Berlin, 1977. |
[20] | [ R. Grech, T. Cassar, J. Muscat, K. P. Camilleri, S. G. Fabri, M. Zervakis, P. Xanthopoulos, V. Sakkalis and B. Vanrumste, Review on solving the inverse problem in EEG source analysis J. NeuorEng. Rehabil. , 5 (2008). |
[21] | [ H. Hallez, B. Vanrumste, R. Grech, J. Muscat, W. De Clercq, A. Vergult, Y. d'Asseler, K. P. Camilleri, S. G. Fabri, S. Van Huffel and I. Lemahieu, Review on solving the forward problem in EEG source analysis J. NeuorEng. Rehabil., 4 (2007). |
[22] | [ M. Hämäläinen,R. Hari,J. Ilmoniemi,J. Knuutila,O. V. Lounasmaa, Magnetoencephalography-theory, instrumentation, and applications to noninvasive studies of the working human brain, Rev. Mod. Phys., 65 (1993): 413-497. |
[23] | [ F. Hecht, O. Pironneau, A. Le Hyaric and K. Ohtsuka, FreeFem++ Manual, 2014. |
[24] | [ E. Hernández,R. Rodríguez, Finite element Approximation of Spectral Problems with Neumann Boundary Conditions on curved domains, Math. Comp., 72 (2002): 1099-1115. |
[25] | [ R. Kress, Linear Integral Equations, Second edition, Applied Mathematical Sciences 82, Spinger-Verlag, 1999. |
[26] | [ J. Kybic,M. Clerc,T. Abboud,O. Faugeras,R. Keriven,T. Papadopoulo, A common formalism for the integral formulations of the forward EEG Problem, IEEE Transactions on Medical Imaging, 24 (2005): 12-28. |
[27] | [ J. Kybic,M. Clerc,O. Faugeras,R. Keriven,T. Papadopoulo, Fast multipole acceleration of the MEG/EEG boundary element method, Physics in Medicine and Biology, 50 (2005): 4695-4710. |
[28] | [ J. Kybic,M. Clerc,T. Abboud,O. Faugeras,R. Keriven,T. Papadopoulo, Generalized head models for MEG/EEG: Boundary element method beyond nested volumes, Phys. Med. Biol., 51 (2006): 1333-1346. |
[29] | [ J. Leblond, Identifiability properties for inverse problems in EEG data processing and medical engineering, with observability and optimization issues, Acta Applicandae Mathematicae, 135 (2015): 175-190. |
[30] | [ S. Lew,D. D. Silva,M. Choe,P. Ellen Grant,Y. Okada,C. H. Wolters,M. S. Hämäläinen, Effects of sutures and fontanels on MEG and EEG source analysis in a realistic infant head model, Neuroimage, 76 (2013): 282-293. |
[31] | [ T. Medani,D. Lautru,D. Schwartz,Z. Ren,G. Sou, FEM method for the EEG forward problem and improvement based on modification of the saint venant's method, Progress In Electromagnetic Research, 153 (2015): 11-22. |
[32] | [ J. C. de Munck,M. J. Peters, A fast method to compute the potential in the multisphere model, IEEE Trans. Biomed. Eng., 40 (1993): 1166-1174. |
[33] | [ Odabaee,A. Tokariev,S. Layeghy,M. Mesbah,P. B. Colditz,C. Ramon,S. Vanhatalo, Neonatal EEG at scalp is focal and implies high skull conductivity in realistic neonatal head models, NeuroImage, 96 (2014): 73-80. |
[34] | [ P. A. Raviart and J. M. Thomas, Introduction á l'Analyse Numérique des Equations aux Dérivées Partielles, Masson, Paris, 1983. |
[35] | [ N. Roche-Labarbe,A. Aarabi,G. Kongolo,C. Gondry-Jouet,M. Dümpelmann,R. Grebe,F. Wallois, High-resolution electroencephalography and source localization in neonates, Human Brain Mapping, 29 (2008): 167-76. |
[36] | [ C. Rorden,L. Bonilha,J. Fridriksson,B. Bender,H. O. Karnath, Age-specific CT and MRI templates for spatial normalization, NeuroImage, 61 (2012): 957-965. |
[37] | [ M. Schneider, A multistage process for computing virtual dipole sources of EEG discharges from surface information, IEEE Trans. on Biomed. Eng., 19, 1-19. |
[38] | [ M. I. Troparevsky, D. Rubio and N. Saintier, Sensitivity analysis for the EEG forward problem Frontiers in Computational Neuroscience, 4 (2010), p138. |
[39] | [ J. Vorwerk,J. H. Cho,S. Rampp,H. Hamer,T. T. Knösche,C. H. Wolters, A guideline for head volume conductor modeling in EEG and MEG, NeuroImage, 100 (2014): 590-607. |
[40] | [ C. H. Wolters,H. Köstler,C. Möller,J. Härdtlein,L. Grasedyck,W. Hackbusch, Numerical mathematics of the subtraction approach for the modeling of a current dipole in EEG source reconstruction using finite element head models, SIAM J. Sci. Comput., 30 (2007): 24-45. |
[41] | [ C. H. Wolters,H. Köstler,C. Möller,J. Härdtlein,A. Anwander, Numerical approaches for dipole modeling in finite element method based source analysis, Int. Congress Ser., 1300 (2007): 189-192. |
[42] | [ Z. Zhang, A fast method to compute surface potentials generated by dipoles within multilayer anisotropic spheres, Phys. Med. Biol., 40 (1995): 335-349. |
1. | Valdas Noreika, Stanimira Georgieva, Sam Wass, Victoria Leong, 14 challenges and their solutions for conducting social neuroscience and longitudinal EEG research with infants, 2020, 58, 01636383, 101393, 10.1016/j.infbeh.2019.101393 | |
2. | Marios Antonakakis, Sophie Schrader, Andreas Wollbrink, Robert Oostenveld, Stefan Rampp, Jens Haueisen, Carsten H. Wolters, The effect of stimulation type, head modeling, and combined EEG and MEG on the source reconstruction of the somatosensory P20/N20 component, 2019, 40, 1065-9471, 5011, 10.1002/hbm.24754 | |
3. | S Schrader, M Antonakakis, S Rampp, C Engwer, C H Wolters, A novel method for calibrating head models to account for variability in conductivity and its evaluation in a sphere model, 2020, 65, 1361-6560, 245043, 10.1088/1361-6560/abc5aa | |
4. | Marios Antonakakis, Sophie Schrader, Ümit Aydin, Asad Khan, Joachim Gross, Michalis Zervakis, Stefan Rampp, Carsten H. Wolters, Inter-Subject Variability of Skull Conductivity and Thickness in Calibrated Realistic Head Models, 2020, 223, 10538119, 117353, 10.1016/j.neuroimage.2020.117353 | |
5. | Marion Darbas, Mohamadou Malal Diallo, Abdellatif El Badia, Stephanie Lohrengel, An inverse dipole source problem in inhomogeneous media: Application to the EEG source localization in neonates, 2019, 27, 0928-0219, 255, 10.1515/jiip-2017-0120 | |
6. | Marion Darbas, Stephanie Lohrengel, Review on Mathematical Modelling of Electroencephalography (EEG), 2019, 121, 0012-0456, 3, 10.1365/s13291-018-0183-z | |
7. | Marion Darbas, Jérémy Heleine, Stephanie Lohrengel, Sensitivity analysis for 3D Maxwell's equations and its use in the resolution of an inverse medium problem at fixed frequency, 2020, 28, 1741-5977, 459, 10.1080/17415977.2019.1588896 | |
8. | Hongguang Xi, Jianzhong Su, A harmonic function method for EEG source reconstruction, 2022, 30, 2688-1594, 492, 10.3934/era.2022026 | |
9. | Carsten H. Wolters, Marios Antonakakis, Asad Khan, Maria Carla Piastra, Johannes Vorwerk, 2021, Chapter 11, 978-1-0716-1212-5, 153, 10.1007/978-1-0716-1213-2_11 | |
10. | Ming Meng, Luyang Dai, Qingshan She, Yuliang Ma, Wanzeng Kong, Crossing time windows optimization based on mutual information for hybrid BCI, 2021, 18, 1551-0018, 7919, 10.3934/mbe.2021392 | |
11. | Stephanie Lohrengel, Mahdi Mahmoudzadeh, Farah Oumri, Stéphanie Salmon, Fabrice Wallois, A homogenized cerebrospinal fluid model for diffuse optical tomography in the neonatal head, 2022, 38, 2040-7939, 10.1002/cnm.3538 | |
12. | Eya Bejaoui, Faker Ben Belgacem, Singularity extraction for elliptic equations with coefficients with jumps and Dirac sources, 2023, 18758576, 1, 10.3233/ASY-221824 | |
13. | Tim Erdbrügger, Andreas Westhoff, Malte Höltershinken, Jan-Ole Radecke, Yvonne Buschermöhle, Alena Buyx, Fabrice Wallois, Sampsa Pursiainen, Joachim Gross, Rebekka Lencer, Christian Engwer, Carsten Wolters, CutFEM forward modeling for EEG source analysis, 2023, 17, 1662-5161, 10.3389/fnhum.2023.1216758 | |
14. | M Darbas, J Leblond, J-P Marmorat, P-H Tournier, Numerical resolution of the inverse source problem for EEG using the quasi-reversibility method, 2023, 39, 0266-5611, 115003, 10.1088/1361-6420/acf9c6 | |
15. | Rose Anne D. Alas, Arrianne Crystal T. Velasco, A Sensitivity-Based Algorithm Approach in Reconstructing Images in Electrical Impedance Tomography, 2024, 12, 2169-3536, 146560, 10.1109/ACCESS.2024.3473616 | |
16. | Malte B. Höltershinken, Pia Lange, Tim Erdbrügger, Yvonne Buschermöhle, Fabrice Wallois, Alena Buyx, Sampsa Pursiainen, Johannes Vorwerk, Christian Engwer, Carsten H. Wolters, The Local Subtraction Approach for EEG and MEG Forward Modeling, 2025, 47, 1064-8275, B160, 10.1137/23M1582874 | |
17. | M. Darbas, S. Lohrengel, B. Sulis, A mathematical model for coregistered data from electroencephalography and diffusive optical tomography, 2025, 20, 0973-5348, 4, 10.1051/mmnp/2025001 |
Mesh | Nodes | Tetrahedra | Boundary nodes | ||
Mesh | Nodes | Tetrahedra | Boundary faces | ||
108 669 | 590 878 | 55 660 |
Mesh | Nodes | Tetrahedra | Boundary nodes | ||
Mesh | Nodes | Tetrahedra | Boundary faces | ||
108 669 | 590 878 | 55 660 |