Model-driven exploration of poro-viscoelasticity in human brain tissue: be careful with the parameters! Authors: Alexander Greiner, Nina Reiter, Jan Hinrichsen, Manuel P. Kainz, Gerhard Sommer, Gerhard A. Holzapfel, Paul Steinmann, Ester Comellas, Silvia Budday Published: 2024 ABSTRACT The brain is arguably the most complex human organ and modelling its mechanical behaviour has challenged researchers for decades. There is still a lack of understanding on how this multiphase tissue responds to mechanical loading and how material parameters can be reliably calibrated. While previous viscoelastic models with two relaxation times have successfully captured the response of brain tissue, the Theory of Porous Media provides a continuum mechanical framework to explore the underlying physical mechanisms, including interactions between solid matrix and free-flowing interstitial fluid. Following our previously published experimental testing protocol, here we perform finite element simulations of cyclic compression–tension loading and compression–relaxation experiments on human brain white and gray matter specimens. The solid volumetric stress proves to be a crucial factor for the overall biphasic tissue behaviour as it strongly interferes with porous effects controlled by the permeability. An inverse parameter identification reveals that poroelasticity alone is insufficient to capture the time-dependent material behaviour, but a poro-viscoelastic formulation captures the response of brain tissue well. We provide valuable insights into the individual contributions of viscous and porous effects. However, due to the strong coupling between porous, viscous, and volumetric effects, additional experiments are required to reliably determine all material parameters. Introduction Despite decades of research, the human brain still poses exciting challenges for researchers from various fields. More recently, there is increasing interest in the role of mechanical signals for brain development [1–3], injury [4–6] and disease [7–10]. Modelling based on the theory of nonlinear continuum mechanics proves a valuable tool to computationally test hypotheses that complement experimental findings [11], to understand processes in the brain under physiological and pathological conditions [12] and to assist diagnosis and treatment of neurological disorders through personalized predictions [13–15]. Depending on the application, mechanical models for human brain tissue need to cover a wide range of time and length scales. Its highly heterogeneous, region-dependent microstructure relates to viscoelastic effects [16] and cannot be neglected for predictions on the organ scale [17]. Viscoelastic models with two relaxation times have been successful in capturing the time-dependent mechanical response of brain tissue under various loading conditions [11,18]. However, free-flowing interstitial fluid occupies a large fraction of the brain volume and contributes to the biomechanical response of human brain tissue through poroelastic effects [19–21]. For some applications, e.g. drug delivery in the brain during cancer treatment, it is essential to model the porous properties of brain tissue explicitly [22,23]. Existing poroelastic models are tailored to particular applications, e.g. tissue fracture [24], decompressive craniotomy [25], tumour growth and treatment [13], hydrocephalus [26] or drug delivery [22]. Models that treat brain tissue as a biphasic poro-viscoelastic material either focus on a specific experimental setup [27] or incorporate important analytical simplifications [28–30]. To our knowledge, the model described by [31] and the formulation proposed by our group [32] are the only approaches to date with the potential to capture the wide range of characteristics observed in the response of brain tissue under different biomechanical loading scenarios by modelling the brain as a poro-viscoelastic material. Our versatile poro-viscoelastic model provides the possibility to describe and explore the underlying physical mechanisms within a biphasic material during mechanical loading, but identifying the associated model parameters becomes increasingly difficult with increasing model complexity. Even for poroelastic models (without accounting for viscoelastic effects) and without the specific application to brain tissue, the meaning of—and the correlation between—the individual model parameters seems poorly understood. In particular, parameters that control the volumetric response of the solid constituent of the biphasic material are chosen and interpreted in various fashions. For example, the shear modulus and the first Lamé parameter are often obtained through direct conversion from the Young’s modulus and the Poisson’s ratio values of brain tissue extracted from literature [22,23,33]. Pierce et al. [34] account for Lamé’s first parameter as a stress-like material parameter, which in the case of isochoric deformation of the solid matrix degenerates to a non-physical (positive) penalty parameter used to enforce incompressibility. Lucci et al. [35] mention that volumetric moduli penalize volumetric changes in the solid skeleton and acknowledge that their estimation is difficult. Indeed, due to the strong interaction between several parameters within a poro-viscoelastic formulation, currently available experimental data is not yet comprehensive enough to find unique material parameters. In this study, we first carefully assess the physical meaning of poroelastic parameters in a poro-viscoelastic model based on the Theory of Porous Media. Thereby, we specifically analyse the correlation between volumetric constraints on the solid constituent and the porous effects in the overall biphasic tissue. We further discuss the size-dependency of poroelastic relaxation behaviour in a fully nonlinear setting. We then provide the basis to identify the free model parameters based on an inverse approach using the finite element method and experimental data from cyclic and stress relaxation experiments on human brain tissue from two different regions: visual cortex (grey matter) and corona radiata (white matter). We find that poroelasticity alone can capture important mechanical characteristics of human brain tissue, but is not sufficient to capture the highly hysteretic material response. Finally, we provide an outlook towards a reliable quantification of poro-viscoelastic material parameter sets in the future. Material and methods We apply our nonlinear poro-viscoelastic model [32] based on the Theory of Porous Media [36] to two different brain regions, i.e. the corona radiata (white matter) and the visual cortex (grey matter). Experimental data As a reference, we use experimental data from large-strain compression and tension experiments performed on two cylindrical samples with a radius r=4mm, one extracted from the human visual cortex (grey matter) and one from the corona radiata (white matter), as illustrated in figure 1. The human brain tissue was extracted from a body donor (male, 71) who had given his written consent to donate his body to research and stored in artificial cerebrospinal fluid [38] until testing. The study was additionally approved by the Ethics Committee of Friedrich-Alexander-University Erlangen-Nürnberg, Germany, with the approval number 405_18 B. For details regarding the sample preparation and the experimental setup, we refer to [38]. We first applied three cycles of compression and tension with a loading velocity of 40μms−1, and minimum and maximum overall vertical stretches of 0.85 and 1.15, corresponding to 15% strain in compression and tension, respectively (figure 1a). Subsequently, we performed a stress relaxation test, first in compression then in tension, at a maximum stretch of 0.85 and 1.15, respectively, with a loading velocity of 100μms−1 and a holding period of 300s (figure 1b). We recorded the corresponding force in the direction of loading and determined the nominal stress as the force divided by the undeformed cross-sectional area of the specimen. The recorded data is subsequently preprocessed using a moving average filter and the Ramer–Douglas–Peucker algorithm [39], where the latter reduces the number of data points. To be able to perform tensile testing, we fully fixed the specimens to the upper and lower specimen holders using sandpaper and superglue. We tested the specimens fully submerged in phosphate buffered saline solution at 37°C. The pictures in figure 1a show a deformed specimen from the cortex at 15% compression (left) and 15% tension (right). Figure 1c demonstrates the lateral retraction of the deformation during long-term relaxation of 2 h for a corona radiata specimen extracted from a body donor (male, 57) who had given his written consent to donate his body to research. The final diameter dt=0.98d0 after relaxation is smaller than the initial diameter d0 at the beginning of the relaxation period, measured with Fiji ImageJ [37]. (a) Total nominal stress response during cyclic loading in compression and tension for samples extracted from the human visual cortex and corona radiata, respectively. In addition, we show a representative deformed specimen from the cortex and its corresponding finite element model. (b) Experimental results for compression relaxation and tension relaxation, normalized by the peak stress. (c) Lateral retraction of the deformation during long-term compression relaxation of 2 h for a corona radiata specimen. The final diameter dt=0.98d0 after relaxation is smaller than the initial diameter d0 at the beginning of the relaxation period, measured with Fiji ImageJ [37]. Continuum kinematics The biphasic brain tissue consists of a viscoelastic solid, representing the network of cells and blood vessels embedded within the extracellular matrix, fully saturated by free-flowing interstitial fluid. The individual solid and fluid constituents are assumed to be incompressible, while the overall compressibility of the biphasic material is captured by changing the solid and fluid volume fractions nS and nF, respectively, subjected to the saturation condition nS+nF=1. Following the Theory of Porous Media, the constituent deformation map reads x=χS(XS,t)=χF(XF,t) and indicates that the material constituents originate from different reference positions XS and XF at time t0, but occupy the same spatial position 𝒙 in the current configuration at time t (figure 2). We obtain the displacement of the solid component Kinematics of a biphasic material body within the context of the Theory of Porous Media [36]. Material particles of the solid and fluid components (PS and PF, respectively) originate from different reference positions in the material configuration B0 at initial time t0, but occupy the same spatial position in the current configuration Bt at time t. Bτ refers to the configuration at time τ>t. The deformation gradient 𝑭S is multiplicatively split into volumetric 𝑭Svol and isochoric 𝑭~S contributions. The isochoric part is decomposed into viscous 𝑭~Sv and elastic 𝑭~Se parts. The seepage velocity 𝒘F=𝒗F−𝒗S=∂𝝌F/∂t−∂𝝌S/∂t describes the motion of the fluid with respect to the deforming solid. The rheological schematic indicates the solid-fluid interaction via a ‘porous' element depending on the pore pressure p and the initial intrinsic permeability K0; μ∞, μ1 and α∞, α1 denote the Ogden shear moduli and nonlinearity parameters for the equilibrium and non-equilibrium part, respectively, while λ* is the first Lamé parameter and η the solid viscosity. and its material deformation gradient Importantly, the Jacobian of the solid deformation gradient JS=detFS>n0SS describes volumetric changes of the whole biphasic material, i.e. it also captures volumetric changes due to pore fluid flow. Once the Jacobian approaches the initial solid volume fraction JS⟶n0SS, the point of compaction is reached, all pores are closed and the incompressibility constraint of the solid component prevents any further volume deformations [36]. Governing equations We assume quasi-static loading conditions and neglect body forces and external tractions such that the weak form of the linear momentum balance in the reference configuration B0 reads The constitutive equation of the solid component renders the Kirchhoff stress tensor 𝝉, dV0S refers to the volume elements of the biphasic material in the reference configuration of the solid and δ𝒖S are the solid displacement test functions. Since we do not prescribe fluid flow across the boundaries, the nonstationary, time-dependent mass balance equation reduces to The constitutive equation of the fluid provides the volume-weighted seepage velocity 𝒘=nF𝒘F, J˙S denotes the material time derivative of the Jacobian and δp are the pore pressure test functions. Constitutive equations We perform a multiplicative decomposition of the solid deformation gradient into elastic and viscous parts, i.e. 𝑭S=𝑭Se⋅𝑭Sv [40]. Similar to previous studies, we assume that the viscous contribution is purely isochoric, based on the premise that the volumetric response of brain tissue is primarily governed by fluid flow captured through poroelasticity. From a biophysical perspective, this modelling choice is motivated by the fact that the viscoelastic response is intended to solely represent the behaviour of the solid matrix, consisting of the extracellular matrix and the network of cells, inside which fluid is mostly trapped in the physiological state. We thus expect a purely isochoric viscous contribution, and the volumetric-isochoric decomposition in figure 2 is recovered through 𝑭S=𝑭Se,vol⋅𝑭Se~⋅𝑭Sv~=𝑭Svol⋅𝑭~S. Then, the solid ‘extra’ stress 𝝉ES is split into (full) equilibrium (eq) and isochoric non-equilibrium (neq) parts and an additional volumetric (vol) contribution: The fluid exerts a hydrostatic stress (−pJS1) on the solid and 1 denotes the second-order unit tensor. Note that setting 𝝉Eneq=0 in equation (2.5) reduces our nonlinear poro-viscoelastic model to a nonlinear poroelastic model. Based on previous studies [11,20,38], we choose a one-term Ogden model for the equilibrium and non-equilibrium parts. The equilibrium part of the Kirchhoff stress tensor is and depends on the principal stretches λS,a and the eigenvectors 𝒏S,a of the left Cauchy–Green strain tensor 𝒃S=𝑭S⋅𝑭S⊤=∑a=13λS,a2𝒏S,a⊗𝒏S,a. The constitutive parameters are the equilibrium Ogden shear modulus μ∞ and the nonlinearity parameter α∞. We formulate the non-equilibrium Kirchhoff stress tensor in terms of the isochoric elastic principal stretches λ~S,ae=[JSe]−1/3λS,ae, the eigenvectors 𝒏S,ae of the elastic part of the left Cauchy–Green strain tensor 𝒃Se=𝑭Se⋅(𝑭Se)⊤=∑a=13[λS,ae]2𝒏S,ae⊗𝒏S,ae, the non-equilibrium Ogden shear modulus μ1 and the nonlinearity parameter α1. To ensure thermodynamical consistency, we assume isotropy and introduce an evolution equation where L𝒗S denotes the Lie derivative along the velocity field of the solid motion and η is the solid viscosity, such that we a priori satisfy non-negative viscous dissipation power, i.e. The volumetric Kirchhoff stress contribution [36] completes the definition of the solid stress tensor (2.5) and introduces the first Lamé parameter λ* of the solid component and the volume fraction of the solid component with respect to the solid reference configuration at the initial time, n0SS. We compute the volume-weighted seepage velocity of the fluid with a Darcy-like law [41] according to where μFR is the effective shear viscosity of the pore fluid and 𝑲0S=K01 is the initial intrinsic permeability tensor, which we assume is isotropic. The porous dissipation power is and will always be non-negative, given that μFR and K0 are necessarily positive, n0SS∈(0,1) and JS>n0SS. Numerical setup The open source finite element library deal.ii [42] provides the numerical framework to reconstruct our large-strain cyclic loading and stress relaxation experiments introduced in §2.1. We discretize a quarter of our cylindrical specimens with 384 full integration Q2P1 elements, i.e. quadratic shape functions for the solid displacements, linear shape functions for the pore pressure and third-order Gaussian quadrature. We have performed a mesh refinement study to ensure that the simulation results were independent of the chosen mesh. The geometry dimensions are specimen specific for the visual cortex (r=4mm, h=3.4mm) and the corona radiata (r=4mm, h=5.0mm). The solid volume fraction is n0SS=0.75 and the fluid shear viscosity μFR=0.89Pa⋅s [32,43]. The degrees of freedom at the bottom surface are fixed in space and a vertical displacement in the z-direction is applied to the top surface while being fixed in the x-y-plane (figure 1a). We apply symmetric boundary conditions to the flat lateral surfaces and only the cylinder hull is drained, i.e., fluid can only leave the sample through the cylinder hull and not through the spatially fixed (glued) top and bottom surfaces. Greiner et al. [43] describe the data analysis of the quantities (stresses, dissipation power, etc.) that we use to visualize and compare the finite element results. In §3.2, we perform cyclic loading simulations with three cycles up to 15% strain in compression and tension at a constant strain rate of 0.01s−1. For the compression relaxation simulations in §3.3, we apply an almost instantaneous load of 15% compressive strain with a strain rate of 2.5s−1 and study the effect of different specimen radii r={4,6,8}mm. Finally, we use an inverse parameter identification algorithm (trust region reflective algorithm), as introduced in [38], to identify the best fitting material parameters to reproduce our experimental results introduced in §2.1 and §3.4. Results The volumetric stress contribution Figure 3 shows the volumetric stress response as a function of the Jacobian of the solid deformation gradient JS for an initial solid volume fraction of n0SS=0.75 and three different values of the first Lamé parameter λ*={102,103,104}Pa. The additional term 𝝉Evol in equation (2.5) primarily ensures incompressibility of the solid component once the compaction point is reached, i.e. 𝝉Evol⟶−∞ when all fluid has left the biphasic material and JS⟶n0SS. Therefore, the domain JS1, the resistance of the solid matrix against volumetric expansion becomes smaller than the fluid flow resistance due to the permeability. This way, the fluid contribution to the total stress reduces while maintaining a large lateral displacement. In contrast, high values of λ* impede volumetric changes leading to less fluid flow and a more homogeneous distribution of volumetric deformations, see figure 11 (right). Deformation state after 60s of poroelastic stress relaxation in compression with Ogden parameters μ∞=−43.8Pa and α∞=−12.76 [38]. Effect of λ*={102,104}Pa on the local volumetric changes for an intrinsic permeability K0=10−8mm2 and a specimen radius of r=4mm. Region-dependent inverse parameter identification To see how our insights obtained from the parameter studies in the previous sections transfer to real experimental data introduced in §2.1, we apply an inverse parameter identification scheme to identify the best fitting material parameters for two regions of the human brain: the grey matter visual cortex and the white matter corona radiata. For details regarding the inverse parameter identification algorithm, we refer to [38]. Figure 12 shows the results of the inverse parameter identification for cyclic compression–tension with a purely poroelastic material model for the visual cortex and the corona radiata. For each brain region, we perform three fits with fixed values for λ*={102,103,104}Pa to determine three unknown material parameters: the Ogden shear modulus μ∞, the Ogden nonlinearity parameter α∞, and the isotropic initial intrinsic permeability K0. In addition, we run one fit with λfit* as a fourth parameter that is identified by the optimization algorithm. In accordance with our observations in §3.2, the highest values for λ*=104Pa lead to the highest model prediction errors, indicated by a root mean square error (RMSE) of 165Pa for the visual cortex and 94Pa the corona radiata, respectively. For λ*=104Pa, the volumetric constraints on the biphasic material suppress the actual porous behaviour and prevent almost any hysteresis. Decreasing λ* improves the quality of the fits significantly and our poroelastic model proves its ability to capture important mechanical characteristics of brain tissue, i.e. compression–tension asymmetry and—to some extent—hysteresis. The optimization algorithm identifies an even lower λfit*=10Pa as the best possible choice. We note that this is the lower limit of admissible values we set for λ* to avoid numerical problems. Nevertheless, the quality of the fit does not change noticeably—the curves for λfit*=10Pa and λ*=102Pa almost coincide (see also RMSE in tables 1 and 2)—indicating that we approach the limits of the poroelastic formulation. Table 1. Inverse poroelastic parameter identification for the visual cortex. Parameters (μ∞, α∞, K0) are identified for four different first Lamé parameters λ*={102,103,104}Pa. λfit*=10Pa shows the result with λ* as an additional parameter for the optimization algorithm.visual cortex μ∞(Pa) α∞(−) K0(mm2) RMSE(Pa) λfit∗=10Pa −1180 −2.38 1.27×10−7 84 λ∗=102Pa −1120 −2.43 1.21×10−7 88 λ∗=103Pa −714 −3.00 8.46×10−8 115 λ∗=104Pa −347 −3.74 2.50×10−8 165 Inverse poroelastic parameter identification for the visual cortex. Parameters (μ∞, α∞, K0) are identified for four different first Lamé parameters λ*={102,103,104}Pa. λfit*=10Pa shows the result with λ* as an additional parameter for the optimization algorithm. Table 2. Inverse poroelastic parameter identification for the corona radiata. Parameters (μ∞, α∞, K0) are identified for four different first Lamé parameters λ*={102,103,104}Pa. λfit*=10Pa shows the result with λ* as an additional parameter for the optimization algorithm.corona radiata μ∞(Pa) α∞(−) K0(mm2) RMSE(Pa) λfit∗=10Pa −498 −3.52 1.89×10−7 57 λ∗=102Pa −457 -3.67 1.73×10−7 60 λ∗=103Pa −256 −4.93 9.95×10−8 76 λ∗=104Pa −266 -3.38 4.01×10−8 94 Inverse poroelastic parameter identification for the corona radiata. Parameters (μ∞, α∞, K0) are identified for four different first Lamé parameters λ*={102,103,104}Pa. λfit*=10Pa shows the result with λ* as an additional parameter for the optimization algorithm. Inverse poroelastic parameter identification for the visual cortex (left) and the corona radiata (right). Parameters (μ∞, α∞, K0) are identified for four different first Lamé parameters λ*={λfit*,102,103,104}Pa (see tables 1 and 2). λfit*=10Pa shows the result with λ* as an additional parameter for the optimization algorithm. Tables 1 and 2 summarize the identified material parameters for cyclic compression–tension and the purely poroelastic case. Both the Ogden shear modulus μ∞ and the initial intrinsic permeability K0 decrease with increasing first Lamé parameter, while the nonlinearity slightly increases. This highlights that λ* influences the solid and the fluid behaviour within the biphasic material. The permeability is slightly lower for the visual cortex than for the corona radiata but in the same order of magnitude. As the purely poroelastic model is not able to capture the highly hysteretic response of brain tissue, we add a viscoelastic element to the solid part of our model (§2.4). Figure 13 shows the results of our inverse parameter identification for cyclic compression–tension for the poro-viscoelastic case. Again, we identify the best fitting material parameters for three different first Lamé parameters λ*={102,103,104}Pa. The optimization algorithm determines the equilibrium and non-equilibrium Ogden parameters μ∞,α∞,μ1,α1, the solid viscosity η, and the initial intrinsic permeability K0. For both brain regions, the poro-viscoelastic model improves the quality of the fits significantly compared to the purely poroelastic model. Interestingly, the fits almost coincide for different values of the first Lamé parameter λ*, indicating that viscoelastic effects dominate the fitted material response. Inverse poro-viscoelastic parameter identification for the visual cortex and the corona radiata. Parameters (μ∞, α∞, μ1, α1, η, K0) are identified for three different first Lamé parameters λ*={102,103,104}Pa. Tables 3 and 4 display the best fitting poro-viscoelastic material parameters for cyclic compression-tension. The equilibrium Ogden parameters for the visual cortex are similar to those of the purely poroelastic model. The corresponding non-equilibrium parameters increase the material nonlinearity and indicate a significant strain-rate dependency of the tissue. The solid viscosity decreases with increasing λ*. The equilibrium Ogden parameters for the corona radiata differ slightly from those for the purely poroelastic model. For all λ*, the equilibrium Ogden shear modulus μ∞ decreases to similar values and the nonlinearity parameter α∞ increases. The solid viscosity decreases with increasing λ*. For both brain regions, the initial intrinsic permeability K0 is similar to that of the poroelastic model for low λ*, while it changes by several orders of magnitude for high λ*. This further demonstrates that high values of λ* suppress porous effects, such that the permeability loses its influence on the biphasic material response and might be chosen almost arbitrarily. Table 3. Inverse poro-viscoelastic parameter identification for the visual cortex. Parameters (μ∞, α∞, μ1, α1, η, K0) are identified for three different first Lamé parameters λ*={102,103,104}Pa.visual cortex μ∞(Pa) α∞(−) μ1(Pa) α1(−) η(Pa⋅s) K0(mm2) RMSE (Pa) λ∗=102Pa −1010 −2.78 -1170 −6.31 5380 1.86×10−7 45 λ∗=103Pa −683 −3.18 −791 -10.7 5230 1.52×10−7 52 λ∗=104Pa −404 -2.94 −1040 -8.85 3300 2.28×10−10 49 Inverse poro-viscoelastic parameter identification for the visual cortex. Parameters (μ∞, α∞, μ1, α1, η, K0) are identified for three different first Lamé parameters λ*={102,103,104}Pa. Table 4. Inverse poro-viscoelastic parameter identification for the corona radiata. Parameters (μ∞, α∞, μ1, α1, η, K0) are identified for three different first Lamé parameters λ*={102,103,104}Pa.corona radiata μ∞(Pa) α∞(−) μ1(Pa) α1(−) η(Pa⋅s) K0(mm2) RMSE (Pa) λ∗=102Pa −173 -5.57 −810 -3.70 3220 5.75×10−9 32 λ∗=103Pa -149 −5.96 -546 −6.18 3120 2.87×10−9 32 λ∗=104Pa −181 -4.17 −2970 -1.46 2650 7.57×10−4 38 Inverse poro-viscoelastic parameter identification for the corona radiata. Parameters (μ∞, α∞, μ1, α1, η, K0) are identified for three different first Lamé parameters λ*={102,103,104}Pa. So far, our results suggest that a single compression-tension cycle is insufficient to reliably quantify poro-viscoelastic material parameters. Therefore, we apply our inverse parameter identification scheme to larger experimental datasets. We fit the experimental response for three subsequent compression–tension loading cycles, followed by compression and tension relaxation. Thereby, we provide additional information on the conditioning behaviour and include two different displacement rates, i.e. 40μms−1 (cyclic loading) and 100μms−1 (relaxation test). Figures 14 and 15 show the fitted material response of the visual cortex and corona radiata, respectively, for λ*=102Pa. Table 5 shows the corresponding best fitting material parameters. Compared to the poro-viscoelastic parameters in table 3, the equilibrium and non-equilibrium shear moduli decrease significantly while the corresponding nonlinearity parameters increase for the visual cortex. The solid viscosity increases slightly and the permeability decreases by two orders of magnitude. For reduced permeability, the fluid takes longer to flow out of the sample, indicating that it controls the long-term material response. We observe similar but less pronounced trends for the corona radiata in comparison with table 4. Inverse poro-viscoelastic parameter identification for the visual cortex. Three loading cycles (yellow), followed by compression (orange) and tension (red) relaxation. Parameters (μ∞, α∞, μ1, α1, η, K0) are identified for λ*=102Pa. Inverse poro-viscoelastic parameter identification for the corona radiata. Three loading cycles (yellow), followed by compression (orange) and tension (red) relaxation. Parameters (μ∞, α∞, μ1, α1, η, K0) are identified for λ*=102Pa. Table 5. Inverse poro-viscoelastic parameter identification for the visual cortex (VC) and the corona radiata (CR). Three loading cycles, followed by compression and tension relaxation. Parameters (μ∞, α∞, μ1, α1, η, K0) are identified for λ*=102Pa.region λ∗(Pa) μ∞(Pa) α∞(−) μ1(Pa) α1(−) η(Pa⋅s) K0(mm2) RMSE (Pa)VC 102 -30.7 −11.9 -45.9 −14.4 6770 4.18×10−9 62CR 102 −81.0 -7.60 −494 -1.53 3860 5.16×10−9 46 Inverse poro-viscoelastic parameter identification for the visual cortex (VC) and the corona radiata (CR). Three loading cycles, followed by compression and tension relaxation. Parameters (μ∞, α∞, μ1, α1, η, K0) are identified for λ*=102Pa. The overall quality of the fit is very satisfactory for such a highly complex material as human brain tissue. We almost perfectly capture the material behaviour under compression and during the relaxation tests for the visual cortex (figure 14). But, we underestimate the stress under tension—especially in the first loading cycle—and cannot reproduce the conditioning behaviour under tension. An almost perfect fit is observed for the corona radiata under tensile loading and the relaxation behaviour is captured reasonably well, while we slightly underestimate the material response under compression (figure 15). All in all, while our poro-viscoelastic model captures the cyclic loading and relaxation behaviour reasonably well, for the identified set of parameters, the model cannot reproduce the conditioning behaviour. Discussion In this study, we have investigated the physical meaning of poroelastic parameters and their interplay with viscous effects in the context of a poro-viscoelastic model following the Theory of Porous Media. Based on our analyses, we have identified a set of model parameters for two different regions of the human brain, visual cortex (gray matter) and corona radiata (white matter), through an inverse approach using the finite element method and experimental data from cyclic and stress relaxation loadings. The identified parameters hint towards differences between the individual contributions of viscous and porous effects in gray and white matter, which can also explain the large variation of corresponding measured properties in the literature that highly depend on the experimental setup and loading conditions. Finally, we have highlighted critical points for the reliable quantification of poro-viscoelastic material parameter sets in the future. The role of the parameter λ* The first Lamé parameter λ* plays a crucial role in the model-predicted overall behaviour of biphasic human brain tissue. Decreasing λ* increases the sensitivity of the model to the permeability K0, while large values of λ* constrain volumetric changes of the biphasic material and lead to diminishing porous effects, less deformation-dependency of the fluid response (see equation 2.11), less model nonlinearity and a more homogeneous material behaviour. From a physical perspective, the permeability should significantly affect the material resistance to compressive loading (e.g. when squeezing a sponge). The lower the permeability, the more fluid remains ‘trapped’ within the tissue, which generates higher hydrostatic (i.e. volumetric) stress, and hence, larger total nominal stresses. Our results show that high values of λ*, around one order of magnitude higher than the classical shear modulus, constrain fluid flow within the tissue and thus prevent physically reasonable behaviour, including almost any porous dissipation during cyclic loading (compare figure 4, right). Reducing λ* allows larger volumetric deformations and fluid flow. This introduces a significant time dependence of the poroelastic material response (see deformation states in figure 6) controlled by the permeability. Therefore, the choice of λ* should be made with caution and under consideration of the expected overall material stiffness and compressibility. In comparison to other material model parameters in the literature, the first Lamé parameter has a qualitatively similar effect to the Poisson’s ratio ν. Therefore, we can relate our findings to [44] mentioning less volume change, less interstitial fluid movement and less fluid flow-dependent viscoelasticity in articular cartilage for ν→0.5, i.e. high λ*. In addition, [45], observe a decreasing fluid contribution with increasing Poisson’s ratio during rat brain indentation. In contrast [31], set a rather high Poisson’s ratio, ν=0.49. Be aware that we assume the elastic stress tensor 𝝉Eeq has both volumetric and isochoric contributions (see equation 2.6), which implies that the equilibrium shear modulus μ∞ may influence part of the volumetric response of the tissue. However, our simulations to date suggest that the volumetric stress response is primarily driven by 𝝉Evol and pJS1, hence we focus on the parameter λ*. A deeper understanding of these potential, hidden interactions could affect how we interpret the influence of the parameters. Note that the choice of the first Lamé parameter λ* might become less crucial under different loading conditions. For example, the parameter sensitivity study performed by [23] found that λ* was not determinant. However, their model considers a triphasic material (solid skeleton, interstitial fluid and blood) with mass-production terms added to the mass balance equation that represent a series of cellular mechanisms. Therefore, the parameters controlling these added terms have a larger influence on the results. Another case is a perfusion setup, where primarily an externally applied fluid pressure drives the fluid flow through the tissue and deformation-driven fluid flow becomes secondary, i.e. no significant material deformation occurs. But, this would require a comparably stiff solid matrix in combination with rather high permeabilities, which is certainly not the case for the ultrasoft human brain tissue. Poroelastic effects during stress relaxation experiments Due to the high loading rate, the biphasic poroelastic material responds like an incompressible elastic material during the loading phase of the relaxation simulations. This results in almost identical stress-strain states before the actual relaxation process starts, independent of the choice of the first Lamé parameter λ* and the initial intrinsic permeability K0. In this initial state, the resistance of the biphasic material to volumetric changes is almost exclusively determined by the fluid constituent, which bears the pressure load. During the relaxation, two processes occur simultaneously. Due to the pressure exerted by the solid on the fluid, fluid starts flowing out of the specimen with a seepage velocity regulated by the intrinsic permeability. At the same time, as fluid content is reduced in the biphasic material, the pressure load on the fluid is gradually transferred to the solid volumetric stresses. This process is governed by the extension function (see equation 2.10), and accelerates with increasing λ*. The larger λ*, the less volumetric changes are allowed, and the earlier the pressure load on the fluid transfers to the solid volumetric stresses. Increasing λ* leads to higher equilibrium stresses and less porous dissipation, i.e. relaxation, while the permeability controls the shape of the curve, i.e. the relaxation times (compare figure 7). Increasing the specimen radius while keeping the specimen height constant induces larger volumetric deformations and the effects of porous dissipation and corresponding relaxation increase in importance. In contrast to other studies [46], the time until equilibrium is reached does not appear to increase with increasing radius. In fact, for high values of λ*, the relaxation behaviour is not affected by the radius, while the shape of the relaxation curve changes with the radius for lower values of λ* (figure 8). This could be related to the high degree of nonlinearity within our model, as [47] report on much higher stress relaxation in rat brain tissue than predicted by a linear biphasic theory. Larger deformation leads to higher solid stresses and higher pore pressure. Subsequently, larger pressure gradients induce higher seepage velocities that compensate for the larger distances the fluid has to move through in a larger specimen. Importantly, the first Lamé parameter λ* not only affects the equilibrium stress during relaxation experiments, but also the temporal evolution of the specimen geometry (figure 9). For high values of λ*, the ‘bulging out’ of the sample during compression (when the specimen is glued to the upper and lower specimen holder) remains, even after a holding time of 600 s, as λ* constrains volumetric changes due to fluid flow. For low values of λ*, more fluid can flow out of the sample, so that the sample almost forms back to a cylindrical shape during the holding time. Note that these effects are also strongly coupled to the deformation-dependent permeability, which decreases with decreasing specimen volume. Incorporating a precise camera setup to capture the specimen geometry and deformation during relaxation experiments can in the future help to choose an appropriate λ* value for poroelastic materials (see also [48]). But, due to the relatively long timescales involved, it remains to be seen whether effects like tissue degradation or swelling need to be taken into account to obtain reliable results. Poro-viscoelastic material parameters for human brain tissue While a purely poroelastic model is unable to capture the highly hysteretic response of brain tissue (compare figure 12), we show that a poro-viscoelastic material model can capture the response of human brain tissue during both cyclic loading and stress relaxation experiments in compression and tension (see figures 14 and 15). Two coupled processes control the time-dependent tissue response: viscous effects are responsible for the short-term relaxation and porous effects take over for the long-term relaxation behaviour. Similar observations have been previously reported for calve white matter [27] and articular cartilage [44]. As discussed in the previous subsections, high values of λ* suppress porous effects, such that the permeability loses its influence on the biphasic material response and might be chosen almost arbitrarily during an inverse parameter identification. Therefore, we purposely chose a low value for λ* to identify all remaining model parameters. We obtained an initial intrinsic permeability K0 of 4.18×10−9mm2 for the visual cortex and 5.16×10−9mm2 for the corona radiata. Table 6 summarizes permeability measures of brain tissue that have been used previously and shows that our identified permeabilities are within a reasonable range. Overall, we observe slightly lower permeabilities for grey matter (visual cortex) than white matter (corona radiata). Differences in permeability alter the biphasic material response to different loading rates: for very small and intermediate loading rates, more fluid flows through white matter tissue and is squeezed out than in grey matter. As a result, white matter appears softer than grey matter under slow and intermediate loading conditions. For extremely fast loading in turn, white matter becomes stiffer as now trapped fluid contributes to the stiffness and the difference between grey and white matter becomes less pronounced or even inverted. For example, during magnetic resonance elastography at very high frequencies, corona radiata has been reported to be stiffer than the cortex [56,57]. Table 6. Values for brain tissue permeability used throughout the literature for different tissue types: white matter (WM), grey matter (GM) and mixed matter (MM). Note that experiments typically provide hydraulic permeability, which can be directly related to the intrinsic permeability used in our model. Units have been converted under the assumption of μFR=10−3Pa⋅s (effective shear viscosity of water at 20∘C) and γFR=104Nm−3 (specific weight of water).studytissue typepermeability K0 in mm2[27]calve, WM 4.08×10−9 [19]human, WM 2.4×10−11 Greiner et al. (present study)human, WM 5.16×10−9 Greiner et al. (present study)human, GM 4.18×10−9 [47]rat, MM 3.56×10−6…2.22×10−12 [31]bovine 1.57×10−9 [49]ovine, WM 0.7…2.0×10−10 [50]human, WM 1.6×10−8 [50]human, GM 1.6×10−10 [30]human 2.19×10−9 [51]human 1.4×10−8 [52]rat, MM 6.4×10−12 [53]sheep, WM 0.43…1.71×10−12 [45]rat, 3 regions 1.2…5.5×10−10 [54]ovine, WM 1.3×10−8…2.0×10−9 [55]human, WM 6.5×10−9 Values for brain tissue permeability used throughout the literature for different tissue types: white matter (WM), grey matter (GM) and mixed matter (MM). Note that experiments typically provide hydraulic permeability, which can be directly related to the intrinsic permeability used in our model. Units have been converted under the assumption of μFR=10−3Pa⋅s (effective shear viscosity of water at 20∘C) and γFR=104Nm−3 (specific weight of water). Equally as important as the porous effect is the second time-dependent process controlled by the solid’s viscosity. Our fits indicate a 75% higher viscosity for gray matter than white matter, i.e. faster relaxation for the corona radiata. Again, this could explain why white matter appears softer for slow loading rates and stiffer than gray matter for high loading rates. In addition, not only the loading rate but also the loading magnitude can impede the comparability between different experimental setups. For example, for the corona radiata, our fit suggests a distinctively lower nonlinearity of the non-equilibrium part in combination with a higher non-equilibrium shear modulus. This indicates a stiffer response in the small-strain regime compared to the visual cortex. Limitations and future perspectives We note that our inversely identified parameters are not yet unique. For example, the porous and viscous timescales could switch such that the porous effects dominate the short-term relaxation and the viscous effects control the long-term relaxation behaviour. This would still allow us to satisfactorily fit the material response, but the underlying physical material behaviour would change completely, i.e. large and rapid volumetric changes would occur. The non-uniqueness resides on the model complexity and shows that additional experimental data are required to reliably calibrate all poro-viscoelastic material parameters. The strong coupling between porous, viscous and volumetric effects in conjunction with the ultrasoft and fragile nature of human brain tissue poses immense challenges to future experimental setups. Specific perfusion experiments that precisely trigger and measure fluid flow through the brain on the tissue scale, combined with tracking local deformation states, appear most promising to identify the first Lamé parameter λ* and the permeability K0. Thereby, the aim should be to keep the overall deformation as small as possible—while maintaining a measurable and stable fluid flow—and to reduce the loading rate such that viscous effects can be neglected. With these parameters at hand, complementary large-strain cyclic loading and relaxation tests can provide the missing viscous material properties and a digital image correlation system could serve as a validation for the local deformation states. Kainz et al. [58] showed that brain tissue-mimicking materials with very similar mechanical properties provide the great opportunity to design and calibrate such new experimental setups while adhering to ethical principles.