Real-time surgical simulation by Proper Generalized Decomposition techniques

Real-time surgical simulation by Proper Generalized Decomposition techniques C. Quesada 12 , I. Alfaro 2a , D. González 2b , F. Chinesta 3c and E. Cueto 2d 1 Biomechanics & Bioengineering Laboratory (UMR CNRS 7338), Université de Technologie de Compiègne – CNRS, Sorbonne Universités, Compiègne, France 2 Aragon Institute of Engineering research, Universidad de Zaragoza, Zaragoza, Spain 3 ESI Chair, Ecole Centrale de Nantes, Nantes, France Abstract. The real-time computer-based simulation of surgery has proven to be an appealing alternative to traditional surgical simulators. Amongst other advantages, computer-based simulators provide considerable savings on time and maintenance costs, and allow trainees to practice their surgical skills in a safe environment as often as necessary. However, in spite of the current computer capabilities, computational surgery continues to be a challenging field of research. One of its major issues is the high speed at which complex problems in continuum mechanics have to be solved so that haptic interfaces can render a realistic sense of touch (generally, feedback rates of 500–1 000 Hz are required). The work here presented introduces some novel numerical methods for the interactive simulation of two usual surgical procedures: cutting and tearing of soft tissues. The common framework of these two methods is the use of the Proper Generalized Decomposition (PGD) for the generation of computational vademecums (Chinesta et al. 2013), i. e. general meta-solutions of parametric high-dimensional problems that can be evaluated at feedback rates compatible with haptic environments. In the case of cutting, computational vademecums are used jointly with XFEM-based techniques, and the computing workload is distributed into an off-line and an on-line stage. During the off-line stage, both a computational vademecum for any position of a load and the displacements produced by a set of cuts are pre-computed for the organ under consideration. Thus, during the on-line stage, the pre-computed results are properly combined together to obtain in real-time the response to the actions driven by the user (Quesada et al. 2016). Concerning tearing, a computational vademecum is obtained from a parametric equation based on continuum damage mechanics. The complexity of the model is reduced by Proper Orthogonal Decomposition (POD) techniques, and the vademecum is incorporated into an explicit incremental formulation that can be viewed as a sort of time integrator. By way of example, the cutting method is applied to the simulation of a corneal refractive surgical procedure known as radial keratotomy (Fig. 1), whereas the tearing method focuses on the simulation of laparoscopic cholecystectomy (i. e. the removal of the gallbladder) (Fig. 2). In both cases, the implemented methods show excellent performances in terms of feedback rates, and produce very realistic simulations from the visual and haptic point of view. Keywords: Proper Generalized Decomposition; computational vademecums; computational surgery; surgical simulation; real-time computing; cutting of soft tissues; tearing of soft tissues; XFEM; POD; continuum damage mechanics References Chinesta, F., Leygue, A., Bordeu, F., Aguado, J. V., Cueto, E., González, D., Alfaro, I., Ammar, A., Huerta, Corresponding author, Ph.D., E-mail: [email protected] a Ph.D., E-mail: [email protected] b Ph.D., E-mail: [email protected] c Professor, E-mail: [email protected] d Professor, E-mail: [email protected] Data-based parametric modeling of human liver anatomy for patientspecific real-time deformable models in computational surgery Nathan Lauzeral 1 , Domenico Borzacchiello 1a , Francisco Chinesta 1 , Michael Kugler 2 , Daniel George 2 , Yves Rémond 2 , Alexandre Hosttetler 3 , Elias Cueto 4 1 ICI, High Performance Computing Institute, Ecole Centrale de Nantes, France 2 iCube, Université de Strasbourg, France 3 IRCAD, France 4 I3A, Universidad de Zaragoza, Spain In this study a parametric Finite Element (FE) model of the human liver is built accounting for anatomical variability from patient to patient. An explicit parametric solution is computed at once for the whole “family” of anatomical shapes with prescribed loads and material properties using Proper Generalized Decomposition (PGD), allowing realtime computation. The shapes of human livers were reconstructed from 712 Magnetic Resonance Images acquired at IRCAD (Institut de Recherche contre les Cancers de l'Appareil Digestif, Strasbourg). All the shapes were registered to a template liver. Non-rigid registration was performed using Iterative Closest Point with Thin Plate Spline parametrization. Several dimensionality reductions techniques were applied, in a comparative way, to find the intrinsic coordinates ζ  describing a lower dimensional manifold of the liver geometries. These include Principal Component Analysis (PCA), kernel Principal Component Analysis (k-PCA), Locally Linear Embedding (LLE) and t-distributed Stochastic Neighbor Embedding (t-SNE). Finally, the livers biomechanical behavior was taken into account using a quasi-static hyper-elastic material model and PGD was used to compute numerical solutions for displacement fields 𝒖 under prescribed loads as an explicit function of the shape parameters 𝜁  . Based on the quality of segmentation, only 678 segmented liver shapes were retained for the study. The correspondence process gave good results to fit the majority of the livers. Since the database contains a large variety of anatomical shapes, we observed that linear dimensionality reduction techniques like PCA are unable to find low dimensional representations for the whole family of shapes. On the other hand, nonlinear techniques like (k-PCA, LLE, t-SNE) are more suitable to produce low dimensional mappings in terms of reduced shape parameters 𝜁  . A nonlinear greedy algorithm is used to compute a reduced order representation. The main advantage of generating an explicit solution 𝒖(𝒙,𝜻, 𝑝  , … , 𝑝  ) (with 𝒙 the space coordinate, 𝜻 the shape-parameter coordinate and (𝑝  ,… , 𝑝  ) additional coordinates that are introduced to parametrize loads and material properties), is that a fully customizable deformable organ model is readily available without the need to go through mesh generation, boundary conditions assignment and FE solution all over again for each patient. Since the solution is pre-computed offline, visualization of particular load-material-shape combination can be done practically in real time. The model can be thought of as a fully customizable virtual deformable twin of the patient’s liver. Keywords: human liver; statistical shape model; dimensionality reduction; model order reduction References Niroomandi et al. (2013), “Real-time simulation of biological soft tissues: a PGD approach.”, Int. J. Numer. Method Biomed. Eng. 29(5), 586-600. Van Der Maaten et al. (2009) "Dimensionality Reduction: a comparative.", J. Mach. Learn. Res. 10, 66-71. Corresponding author, Ph.D. Student, E-mail: nathan.lauzer[email protected]c-nantes.fr a Ph.D. , E-mail: [email protected] Source localization of uterine activity using Maximum Entropy on the Mean approach Saeed Zahran 12a , Ahmad Diab 2 , Mohamad Khalil 2 and Catherine Marque 1 1 Université de Sorbonne, Université de technologie de Compiègne, CNRS, UMR 7338 BMBI, 60200 Compiègne, France 2 Ecole d'ingénieurs et Ecole Doctorale en Sciences et Technologie, Lebanese University, Lebanon. Abstract. This study evaluates the ability of distributed source localization method to accurately estimate the location of the sources of activity, as well as their sensitivity to the spatial extent of such sources when using EHG data. We used here realistic simulations of uterine electrical activity, taking into account the 16 electrodes for recording EHG signal, and involving a realistically shaped uterus model. A Data Driven Parcellization (DDP) method [1] was used to segment the uterus surface into non-overlapping regions. Our results showed that the localization were sensitive to spatial extents of the sources (ranging from 11 𝑐𝑚  to 29 𝑐𝑚  ), and also to the number and size of the regions defining the model. Our analysis of the evolution of the real sources during contraction showed a nonlinear propagation of uterine electrical activity. Keywords: Maximum Entropy on the mean; Data Driven parcellization DDP; nonlinear propagation of uterine electrical activity Reference [1] R. A. Chowdhury, J. M. Lina, E. Kobayashi, and C. Grova, “Meg source localization of spatially extended generators of epileptic activity: comparing entropic and hierarchical bayesian approaches,” PloS one, vol. 8, no. 2, p. e55969, 2013. Fig1: Qualitative assessment. Visual analysis of source localization results together with Area Under the ROC curve (AUC) and Spatial Dispersion(SD) values for a simulated source of spatial extent se = 2, 3 and 4 at different spatial scale s=15, 10 and 8. a Ph.D. Student, E-mail: [email protected] Dynamic Viscoealstic Properties of a Fiber Composite Phantom for MRE Martina Guidetti 1 , Jacopo Romanò 1 Dieter Klatt 2 , Thomas J. Royston 2 , Dario Gastaldi 1 , Pasquale Vena* 1 1. Department of Chemistry, Materials and Chemical Engineering Giulio Natta, Politecnico di Milano, Italy; 2. Department of Bioengineering, University of Illinois at Chicago, USA Abstract: Magnetic Resonance Elastography (MRE) is a non-invasive imaging technique employed to assess biological tissue properties (typically shear stiffness) by inducing the propagation of mechanical waves in the region of interest and measuring the tissue response by means of phase contrast magnetic resonance imaging [1]. The challenge in employing MRE on muscular tissue is to characterize a non-homogeneous, viscoelastic, and anisotropic material through an inverse approach able to identify regional variation of tissue viscoelastic properties from material response to harmonic mechanical loading. Different inverse engineering techniques can be used to identify material properties either by direct local inversion of Navier's equations or by iterative methods aiming at minimizing the mismatch between the experimental results and simulation predictions. The inverse problem can be quite a formidable task for complex material behaviour involving anisotropy and viscoelasticity. The use of phantom composite materials with a-priori known mechanical properties which mimic the anisotropic frequency dependent response of skeletal muscle tissues can be a suitable strategy to set up identification methods. In this work, we propose a finite element model able to predict the macroscopic response of the fiberreinforced micro-composite used as a phantom material. A regular cross-ply micro fiber layers immersed in an isotropic PVA gel matrix has been simulated by modeling a unit cell of the composite and considering the material response as that of an infinite repetition of the representative unit cell. Known viscoelastic properties in the frequency domain have been assumed for the constituents (fibers and matrix) and periodic boundary conditions have been used to replicate the behaviour of adjacent cells not included in the model. The homogenized macroscopic constitutive model is then used to predict the response of the phantom subjected to harmonic loading. The effect of fiber density and the effect of macroscopic anisotropic material response is assessed in terms of displacement amplitude and phase angle in a frequency domain steady state dynamic finite element simulation of propagation of steady shear waves. The model will be used in the future to design optimal mechanical properties of the phantom material to be employed in MRE experiments and set up of inverse identification approaches. This study aimed to develop a model to accurately predict the acceleration of structural systems during an earthquake. The acceleration and applied force of a structure were measured at current time step and the velocity and displacement were estimated through linear integration. Keywords: Magnetic Resonance Elastography (MRE); Finite Element models; Steady State Dynamic References [1] Klatt D, Papazoglou S, Braun J, Sack I. Viscoelasticity-based MR elastography of skeletal muscle. Phys Med Biol. 2010;55(21):6445-59 Modelling the adaptation of skeletal muscles in response to isometric exercise Ekin Altan 1,2 , Leonardo Gizzi 1,2a and Oliver Röhrle 1,2b 1 Institute of Applied Mechanics (Civil Engineering), University of Stuttgart, Pfaffenwaldring 7, 70569 Stuttgart, Germany 2 Cluster of Excellence for Simulation Technology, University of Stuttgart, Pfaffenwaldring 5a, 70569 Stuttgart Germany Abstract. Exercise-induced skeletal muscle adaptation occurs over time through changes in mechanical and neural properties of the muscle targeted by exercise. In this study, a computational model describing the adaptation of skeletal muscles in response to isometric exercise is used to investigate the neural adaptation mechanisms. A systematic review of the experimental studies on isometric exercise was performed. The review focused on randomized controlled studies involving healthy, normal-weight, young (aged between 18-30) subjects, and studies which included the measurement of the changes in maximal voluntary contraction (MVC) of the target muscle. These criteria yielded a total of 56 experimental studies. When grouped based on the specific exercise type and the muscle group, it was seen that the highest number of studies (22) focused on isometric unilateral knee extension exercises in which the target muscle is the quadriceps femoris (QF). Training regimes (number of repetitions, number of sets, frequency of training sessions, rest in between sets and individual contractions, length of the individual contractions, total duration of the study), profile of the subjects (gender, age, weight, height) and changes in the MVC of QF were extracted. Data on the changes in neural properties (e.g. level of activation) and cross-sectional area and/or volume of the muscle were also extracted depending on their availability. A sensitivity analysis was performed on the extracted data to investigate which parameters influence the improvement in MVC and to what extent. The results of the sensitivity analysis provided the parameters needed for formulating the adaptation equation that describes the change in strength throughout the training period. The adaptation equation is structured as an ordinary differential equation (ODE) with respect to the exercise input. The exercise input depicts the (cumulative) number of contractions for a given exercise regime (i.e. exercise input = repetitions/set x number of sets/training session x training sessions/week x total duration of the exercise in weeks). A one-dimensional active muscle model that generates force upon the summation of individual motor unit firings was used to compute the MVC. In the model, the motor units are distributed between type I and type II. The force is generated by convoluting the twitch response of individual motor units with their respective firing times. The firing times of individual motor units are assumed to be statistically independent. The firing rate of individual motor units, the number of recruited motor units and their synchronization rate were altered in various combinations to match the improvement in MVC that is computed from the adaptation equation for a given exercise input. Keywords: isometric exercise; skeletal muscle mechanics; systematic review; adaptation of skeletal muscles  Corresponding author, Ph.D. student, E-mail: [email protected] a Ph.D., Leonardo Gizzi. E-mail: [email protected] b Professor, Oliver Röhrle, E-mail: [email protected] Carriou, V. and Boudaoud, S. and Laforet, J. and Ayachi, F.S. (2016), “Fast generation model of high density surface EMG signals in a cylindrical conductor volume”, Comput. Biol. Med., 74, 54-68. Al Harrach, M. and Carriou, V. and Boudaoud, S. and Laforet, J. and Marin, F. (2017), “Analysis of the sEMG/Force relationship using HD-sEMG technique and data fusion: A simulation study”, Comput. Biol. Med., 83, 34-47. Fig. 1: Block diagram of the neuromuscular model Object-oriented programming of optimized analytic neuromuscular model Vincent Carriou 1a , Jeremy Laforet  1 and Sofiane Boudaoud 1b 1 Sorbonne university, Universite de Technologie de Compiegne, CNRS UMR 7338 Biomechanics and Bioengineering, Centre de Recherche de Royallieu – CS 60203, Compiegne, France Abstract Each movement even the simplest is the result of complex interactions between several systems. The muscles responsible for the joint movement respond to a neural command controlled by the Central Nervous System. Then, muscles will generate several physical and chemical responses to this neural activation, i. e., calcium release, electrical activity and force generation. These muscle responses cannot be measured in experimental conditions without using an invasive protocol. Moreover, the interactions contributing to the muscle contraction cannot be experimentally determined. For this purpore, simulations can provide some new insights about the muscle contraction and a preliminary step before starting an experimental study. In this paper, we present the software structure of a neuro-electrical model (Carriou2016) describring the contraction of a muscle. We will point out the well suited object-oriented programming used to implement this model. Object-oriented programming is a programming concept based on macro structure containing data describing the object. Those objects can interact together through methods. Considering the complexity of the neuromuscular System of Systems (SoS) the first step before implementing the model is to clearly described the different systems and the interactions between them (see Fig. 1). Implementation of the model is made in Python programming language which is an interpreted language. To clarify each object and method developed in the model, a documentation is written within the model to describe how to use these features. Moreover, the Application Programming Interface (API) has been specified to simplify the use of the model through human readable definition of the model's input/outputs. One of the significant advantage of the object-oriented programming is the modularity of the computing code. In fact, this code modularity allows users to easily upgrade and extend the computing code to fit new functionalities (Al Harrach2017). Once these features considered, the management of the simulation computing time is the other challenge in muscle modeling. In the proposed model, reduction of the computing time is made at three levels. Firstly, optimization has been made on the mathematical equations describing the problem where all the calculus are considered in the Fourier domain to avoid the use of the convolution operator. Then, optimization procedures available with Python (list comprehension, code profiling, optimized libraries, etc.) were implemented in the model. Finally, thanks to all these features presented above, parallel computing has been easily integrated in the model, reducing significantly the computing time of the model. To conclude, considering the object-oriented programming for modelling the complex neuromuscular system seems to be well-suited for describing its composition and interactions (Carriou2016). Thanks to these integrated features in the model, upgrading the model to new methods or studies is feasible. Thus, a widespread use of this model in the scientific community can be considered. Keywords: Neuromuscular system; System of systems; Object-oriented programming; Skeletal muscle model; Parallel computing; Analytic model; References Corresponding author, Ph.D., E-mail: [email protected] a Ph.D. Student, E-mail: [email protected] b Ph.D., E-mail: [email protected] September 7 th 2017 SESSION ABSTRACTS Computational modelling of the dual action of parathyroid hormone in osteoporosis Silvia Trichilo 1 , Stefan Scheiner 2 and Peter Pivonka *3 1 St Vincent’s Department of Surgery, University of Melbourne, Melbourne, Australia. 2 Institute for Mechanics of Materials and Structures, Vienna University of Technology, Vienna, Austria. *3 School of Chemistry, Physics and Mechanical Engineering, Queensland University of Technology, Brisbane, Australia Abstract (500 words max). Osteoporosis is a disease characterized by long-term bone loss that occurs when bone resorption exceeds bone formation in the remodelling process. Progress has been made in the formulation of computational models of bone remodelling in order to take into account major cellcell signalling pathways and regulatory mechanisms. Many hormones exhibit different release patterns, such as continuous or intermittent, which then modulate differential cell behaviours. In the bone literature, the most prominent example is the dual action of parathyroid hormone (PTH). It has been shown that a continuous administration of PTH leads to a catabolic effect on bone remodelling. On the other hand, daily subcutaneous injections of PTH constitute an effective anabolic treatment for osteoporosis. However, current computational models of bone remodelling are not able to distinguish between these two administration patterns. The purpose of our study was to develop a computational model of bone remodelling that takes into account the dual action of PTH. The model was built using human pharmacokinetics data, then calibrated and validated with data from postmenopausal osteoporosis (PMO). The mechanism implemented in the model accounts for the anabolic effect of intermittent PTH on bone remodelling via the reduction of the apoptosis rate of active osteoblasts (OB). This mechanism involves the runt-related transcription factor 2 (Runx2) and the cAMP response elementbinding protein (CREB). Runx2 is a mediator for the transcription of survival genes, such as B-cell lymphoma 2 (Bcl-2). The action of PTH on OB apoptosis is modeled with a system of three ODEs describing the intracellular signalling components (Runx2, CREB, and Bcl-2) as a function of PTH. Changes of Bcl-2 drive the reduction of OB apoptosis rate. This action is implemented in the model via a sigmoid E max function. The effect of PTH on bone remodelling is described by coupling the intracellular model of OB apoptosis together with a bone cell population model to compute changes over time of bone cell numbers and bone matrix fraction (f bm ). The latter quantities can be linked to bone turnover markers (BTM) and bone mineral density (BMD). To reproduce the change over time of PTH concentration in plasma under different dosing regimens, a pharmacokinetic (PK) model has been developed. PMO is simulated with a rate of bone loss equal to 0.65%/year. The daily PTH subcutaneous injections are reproduced using a pharmacokinetic model. The computed apoptosis rate follows the intermittent behaviour of PTH, with an overall reduction in the daily average compared to baseline. The model shows a 3% trabecular f bm increase from baseline after the simulation of 2 years of treatment. This value is consistent with the BMD increase measured at the femur neck and distal radius. Simulation of 40 µg PTH injections lead to a higher f bm gain compared to the 20 mg dose. These results indicate that the model is capable of reproducing a dose-dependent gain in bone volume. By coupling the intracellular signalling of OB apoptosis and the bone remodelling process, our model can simulate the anabolic effect of intermittently administrated PTH for PMO treatment. Keywords: bone remodelling; bone mechanobiology; computational modelling; osteoporosis; pharmacokinetics; pharmacodynamics; Subject-specific Shoulder Muscle Attachment Region Prediction Using Statistical Shape Models: A Validity Study Asma Salhi1,2a, Valerie Burdin1,2b, Tinashe Mutsvangwa3c, Sudesh Sivarasu3d, Sylvain Brochard2,4e, Bhushan Borotikar*1,2 1Department of Image and Information processing, IMT Atlantique, Brest, France 2Laboratory for Medical Information Processing (LaTIM), INSERM, U1101, Brest, France 3Department of Biomedical Engineering, University of Cape Town, South Africa 4Department of Physical Medicine and Rehabilitation, CHRU Brest, France Abstract: Subject-specific musculoskeletal models (MSKMs) can predict accurate joint and muscle biomechanics. However, assumptions made about the input model parameters make them generic and limit their clinical utility. Shoulder MSKM faces a huge challenge of subject specificity, in particular, muscle origin/insertion sites are always almost approximated. This considerably affects the muscle force and moment arm predictions and thus cannot be effectively used for pre-surgical or for rehabilitation assessments. In this study, we present a statistical shape model (SSM) based pipeline of using the SSM’s ability to incorporate statistical variability for predicting subject-specific muscle regions in shoulder muscles. We also report the concurrent validity of the muscle origin/insertion region prediction on a randomly selected population of shoulder bones (scapula and humerus). A database of dry bone samples (27 scapulae and 28 humeri) was used for building the bone SSMs using previously published IMCP-GMM pipeline [1]. The bone SSMs were then augmented with five muscle attachment (origin/insertion) regions (Subscapularis, Supraspinatus, Infraspinatus (I-S), Teres Major (T-Maj) and Teres Minor) on both the bones. For each bone surface mesh of the database in correspondence, origin/insertion regions were identified and masked by two experts using Meshlab (http://meshlab.sourceforge.net/). Inter-rater reliability for two randomly selected muscles was quantified by comparing the area of each muscle region obtained by each expert on each bone and also by quantifying dice similarity coefficient for each muscle region. The regions were represented by subset of vertices on the bone meshes and were tracked using vertex identifiers (VIDs). Using the correspondence within the database, a subset of vertices representative of each muscle was extracted and identified on mean shape of the bone SSM, by exploring the frequency of appearance of each point (VID) in the same muscle for the whole database (Fig. 1). To avoid the overlapping between muscle attachment sites, a frequency of 60% was selected to ensure the exclusivity of each VID for a single muscle region. Subject-specific muscle attachment regions were predicted using external set of bones not used in building the SSMs. An open source toolbox SCALISMO [2] was used to create and perform functions related to augmented SSMs. Excellent intra-class correlation coefficient (ICC) for inter-rater reliability was reported for the two muscle origin/insertion regions on scapula (I-S (ICC = 0.927) and T-Maj (ICC = 0.942)), as well as, on the humerus (I-S (ICC = 0.981) and T-Maj (ICC = 0.962)). Dice coefficient ranged from 0.821 to 0.987 indicating a high region similarity between experts. Validity of region prediction was visually confirmed by the expert in ten external scapulae and eight external humeri. Excellent concurrent validity of muscle region prediction was observed based on the mean and root mean square distance measures and also based on similarity coefficient. We assumed that bone shapes can predict muscle origin/insertions regions and the excellent concurrent validity results confirm this for five major shoulder muscles on both the bones. The SSM based MSKM pipelines seems to be a good approach however, further validations are warranted on all the muscles of the shoulder complex. Keywords: shoulder biomechanics, muscle insertion prediction, SCALISMO, point correspondence microvessels as order principle to restore tissue architecture in liver regeneration. Proc. Natl. Acad. Sci. (USA), 107(23), 10371-10376. Schliess, F., Hoehme, S., Henkel, S., et. al.. 2014. Integrated metabolic spatial-temporal model for the prediction of ammonia detoxification during liver damage and regeneration. Hepatology 60 6, 2040–51. Geraldine Celliere, Ahmed Ghallab, Noemie Boissier, Stefan Hoehme, Tim Johann, Jan Hengstler, Dirk Drasdo GC, NB, DD: INRIA de Paris, France; AG, JH: IfADo, Dortmund, Germany; SH, TJ: IZBI & Inst. for Comput. Science, Univ. Leipzig, Germany Integrated vs. multi-level/multi-scale spatial temporal modeling of ammonia detoxification after drug-induced liver damage: using modeling to guide towards a new therapy approach Keywords: integrated modeling; compartment modeling; multiscale modeling; spatial-temporal modeling; agent-based modeling; liver damage; detoxification; model-guided experimentation; therapy Hyperammonemia is a severe complication after drug induced liver damage, for example resulting from overdosing acetaminophen (paracetamol), and can lead to encephalopathy and dead of the patient. A set of chemical reactions identified by Häussinger (1983) and Gebhardt (1983) has become the biological consensus model for ammonia detoxification in healthy liver. We will show how the iterative application of a pipeline consisting of confocal scanning microscopy, image analysis and modeling can be used to design predictive models of tissue regeneration and metabolism suited to guide modeling driven experimental strategies (Drasdo et. al., 2014). As an example we will present an integrated model, integrating a compartment model of ammonia detoxification and a spatial-temporal micro-architectural agent-based model of liver regeneration after drug induced liver damage, that was able to identify lack of a critical ammonia sink mechanism in the consensus reaction scheme (Schliess et. al., 2014). The finding has led to identification of a so far unrecognized ammonia sink mechanism that could be experimentally demonstrated to represent a potential therapy approach in hyperammonemia (Ghallab et. al, 2016). In a further step we redo the analysis in a full spatial temporal microarchitecture model of the smallest virtual functional micro-anatomical unit (called lobule) obtained from image analysis (Hammad et. al., 2014; Friebel et. al., 2015) whereby the detoxification reactions are executed in each individual hepatocyte. Comparison between integrated and full multiscale modeling identifies transport / reaction conditions under which critical differences between both approaches are to be expected. Drasdo, D., Hoehme, S., Hengstler, JG. How predictive quantitative modeling of tissue organization can inform liver disease pathogenesis. Journal of Hepatology, Volume 61, Issue 4, October 2014, pp 951-956. Friebel, A., Neitsch, J., Johann, T., et. al. (*shared senior authors). TiQuant: Software for tissue analysis, quantification and surface reconstruction. Bioinformatics 2015. doi: 10.1093/bioinformatics/btv346. Jun 3. Ghallab, A., Henkel, S.G., Cellière, et al. Model guided identification and therapeutic implications of an ammonia sink mechanism. J. Hepat, 64(4):860-71, doi: 10.1016/j.jhep.2015.11.018. Hammad S., Hoehme S., Friebel A., et. al. “Protocols for staining of bile canalicular and sinusoidal networks of human, mouse and pig livers, three-dimensional reconstruction and quantification of tissue microarchitecture by image processing and analysis.” Arch. of Toxicol. 88 (5) 1161-1183 (2014) Hoehme, S., Brulport, M., Bauer, A., et. al. (2010). Prediction and validation of cell alignment along References (2 max) [1] Weiss, M. et al. (2011), “A physiologically based model of hepatic ICG clearance: Interplay between sinusoidal uptake and biliary excretion”, European Journal of Pharmaceutical Sciences, vol. 44, no 3. 359-365. [2] Monolix documentation http://monolix.lixoft.com Intra-operative quantitative estimation of liver function with indocyanine green fluorescence measurements Chloe Audebert ∗ *1,2 , Anthony Daures 3 , Philippe Rizo 4 , Eric Vibert 5 and Irene E. Vignon-Clementel 1,2 1 Inria Centre de Paris, Paris, France 2 Sorbonne Universités, UPMC, Laboratoire Jacques-Louis Lions, Paris, France 3 Fluoptics company, Grenoble, France 4 CEA-LETI, Grenoble, France 5 Centre Hepato-Biliaire, Paul Brousse Hospital, INSERM U1193, Villejuif, France Abstract (500 words max). Liver transplantation and liver partial ablation are surgeries performed to treat liver diseases (cirrhosis, liver cancer). These surgery complications are related to a poor liver function, thus the evaluation intraoperatively of the hepatic function is an important clinical question. The liver function can be evaluated with blood sample analysis (for example by quantifying the level of bilirubin). Since the indocyanine green is a fluorescent dye exclusively removed from the blood by the liver cell, another method for liver function estimation is the indocyanine green clearance (plasma disappearance rate). However, these methods require time and therefore are not intra-operatively available. Besides, they only provide a combined information about possible perfusion/liver functions dysfunction. In this work, we propose a framework to evaluate the liver function intra-operatively, based on a pharmacokinetics model and its parameters identification, analyzing the indocyanine green fluorescence dynamics in the liver tissue. In [1], the measurements of liver tissue concentration of indocyanine green are fitted with a sum of two exponential functions (with two parameters) for six groups of rabbit with different treatments (colchicine, vessel occlusion…). In this work, the indocyanine green fluorescence is measured in the liver blood vessels, the liver tissue and the common bile duct, before or after liver partial ablation on pigs. The measurements enable to develop a mathematical model representing the indocyanine green processing from its removal from blood by the hepatocytes to its secretion into bile. The model focuses on a precise description of the dye processing by the liver, including the three different types of liver tissue (the sinusoids, the hepatocytes and the bile canaliculi). This model enables to precisely study the exchanges between the different liver tissues. This is a novelty compared with the previous model proposed in literature. First, the model’s outputs sensitivity to parameter is analyzed, and then the parameter estimation is performed with a population approach using Monolix software [2]. The parameters are estimated with the fluorescence measurements from pig surgeries. The link between the model’s parameters and liver function is investigated. Then, the model is adapted to clinical constraints. Since more data is available during pig surgeries, the procedure is first tried out with a subset of measurements from pig surgery that would be available in the clinics. Then, a feasibility study is performed on a few measurements from patients after liver transplantation. The first results suggest that the liver surgeries impact the exchange between blood and liver cells. This new framework may provide a new estimation of the liver function(s) intra-operatively. Keywords: Pharmacokinetics models; parameter estimation; indocyanine green; liver function estimation ∗Corresponding author, PhD. Student, E-mail: [email protected] On a Tri-Scale Multiphase Model for the Description of Perfusion coupled to Growth Effects in Human Liver Navina Waschinsky 1 , Tim Ricken 1a and Lena Lambers 1b 1 Chair of Mechanics, Structural Analysis, and Dynamics, TU Dortmund University, August-Schmidt-Str. 6, Germany The human liver regulates metabolism in a complex time depending and non-linear coupled function-perfusion-mechanism. The viability of the organ is affected by a failure in the liver structure. A common chronic disease is the excessive accumulation of fat in the tissue, known as a fatty liver. The growing fat proportion has a high impact on the perfusion of blood through the liver. A central task of the human liver is the clearance of toxic metabolites which can be found in several medications. The clearance capacity of the human liver depends on an unimpaired perfusion of the blood through the vascular system and micro circulation. Since fatty liver growth influences the perfusion we claim that a fatty liver impacts the recommended dose of daily medicines. The anatomy of the organ is characterized by a complex vascular system which changes on the different size scales from a vascular branching tree to microvessels, called sinusoids in liver lobules. To capture the interplay between fat deposition arising in the microstructure and the perfusion on the organscale it is important to couple the processes on each scale. For this we present a computational model for the human liver which is composed of three coupled submodels for the organ-, lobuleand cellscale. With the whole organ model we present the effect of growing fat vacuoles in the liver cells, which are inhomogenously distributed on organ-scale, up to the total hepatic hemodynamics. On the organscale we use a Bernoulli approach (laminar steady state) calculating the perfusion in the branching system of the vascular tree. The vascular system starts with a single branch and subdivides into smaller vessels ending up in the portal triad - the interface of the organscale to the lobulescale. With a computation of the vascular perfusion we provide information of the blood velocity and pressure as boundary condition. The lobulescale uses a homogenized mixture model for the simulation of important functionalities in the liver lobules as presented in [2] including microperfusion, growth aspects and clearance capacity. The approach uses the extended theory of porous media (eTPM) [1] to consider the biological tissue as a multi-phase structure including the liver tissue, the fat vacuoles and the blood. On the microscale we focus on metabolic processes which take place in the liver cells. Keywords: Tri-Scale Multiphase Model; eTPM; hepatic hemodynamics; growth; clearance References [1] De Boer, R. (2012), “Theory of porous media: highlights in historical development and current state.” Springer Science & Business Media. [2] Ricken, T., et al. (2015), “Modeling function perfusion behavior in liver lobules including tissue, blood, glucose, lactate and glycogen by use of a coupled two scale PDE ODE approach. Biomechanics and modeling in mechanobiology”, Biomechanics and modeling in mechanobiology, 14(3), 515-536.  Navina Waschinsky: [email protected] a Tim Ricken: [email protected] b Lena Lambers: [email protected] Modeling of Field Potential in Microlectrode Arrays and Applications in Safety Pharmacology. Fabien Raphel 1,2 , Muriel Boulakia 2,1 , Philippe Zitoun 3 , Jean-Frédéric Gerbeau * 1,2 1 Centre Inria de Paris, 2 rue Simone Iff, 75012 Paris, France 2 Laboratoire JL Lions, UMR 7598, UPMC-Sorbonne Universités, 75005 Paris, France 3 LIPZ, 78170 La Celle Saint Cloud, France Abstract The study of the electrophysiological impact of drugs, e.g. in the CiPA initiative [1], are mainly addressed by three approaches: in vitro patch clamp assays; in silico Action Potential (AP) assays; in vitro assays on human induced pluripotent stem cells (hiPSC-CMs). The present work investigates a 4 th direction: in silico assays on hiPSC-CMs. The use of microelectrode arrays (MEAs) and hiPSC-CMs allows high-throughput screening of new drugs directly on human-derived cells. But the field potential (FP) signals acquired by MEAs are difficult to analyze. We believe that an in silico approach can improve the practical use of FP signals. The presentation will include a demonstration of a web application of our FP analysis tool. Method: We developed a new strategy based on a mathematical model of MEA and an inverse problem. The model consists of the bidomain partial differential equations, coupled to a system of equations representing the ionic channels. The model provides both AP and FP, and can account for cell heterogeneities and imperfect contact with electrodes. Various devices (96-well and 6-well MEAs) and ionic models (Paci et al. for hiPSC-CM, or classical ionic models like O’Hara-Rudy or MV) were tested. An inverse problem algorithm was used to identify parameters of the ionic models from quantities measured on the FP. Results: By extracting three biomarkers from real FP data for various drugs, the proposed identification algorithm provided concentration-response curves for potassium, sodium and calcium channels. IC50 determined by simulation were in good agreement with literature values (2.8uM for Ivabradine, 114uM for Moxifloxacin). Conclusion: The forward simulations provide signals qualitatively close to those experimentally observed. They give new insights into some aspects of the FP difficult to address in real experiments, like the variability of the measurements, even in absence of variability of the cells. The inverse problem allows us to estimate the conductance of some channels and to obtain dose-response curves directly from FP measurements. Keywords: cardiac electrophysiology; safety pharmacology; hiPSC-CM; MEA; References [1] Cavero, I., Guillon, J. M., Ballet, V., Clements, M., Gerbeau, J. F., Holzgrefe, H. (2016). "Comprehensive in vitro Proarrhythmia Assay (CiPA): Pending issues for successful validation and implementation", Journal of pharmacological and toxicological methods, 81, 21-36. Study of the effect of partial hepatectomy on liver function through modeling of ammonia detoxification. Noemie Boissier ∗ 1, 2 , Stefan Hoehme 3 , Adrian Friebel 3 , Geraldine Celliere 1 ,Tim Johann 3 , Jan Hengstler 4 , Irene Vignon-Clementel 1, 2 and Dirk Drasdo 1, 2 1 INRIA Centre de recherche de Paris, 2 rue Simone Iff, CS 42112, 75589 Paris, France 2 Laboratoire Jacques Louis Lions, UPMC, Sorbonne Universités, 4 place Jussieu, 75005 Paris, France 3 Interdisciplinary Centre for Bioinformatics, Leipzig University, Härtelstraße 16-18, 04107 Leipzig, Germany 4 Department of Toxicology, Leibniz Research Centre for Working Environment and Human Factors, IfADo, Ardeystr. 67, D - 44139 Dortmund, Germany Abstract Along with the kidney, liver is one of the main detoxifying organs, with a remarkable regenerative capacity. To treat some diseases, as liver cancer or cirrhosis, the damaged part of the liver can be surgically removed. Within a few days, liver will grow back to its original weight. After surgery the liver mass is reduced which can impair its ability to detoxify the blood from ammonia. Yet ammonia concentrations in the blood above a certain threshold for too long can cause brain damage and in extreme cases death. This study aims at studying the influence of partial hepatectomy on liver function through the study of ammonia detoxification. We have developed a multi-scale model of ammonia detoxification after partial hepatectomy, taking into account its metabolism in the cells as well as the spatial inhomogeneities within the liver. From experimental confocal images of liver tissue before and after partial hepatectomy analyzed in TiQuant [1], statistically representative functional and anatomical units of the liver, liver lobules, at different time points have been generated. Blood flow is modeled as a resistive network, taking into account the effect of red blood cells.. Using a representative liver lobule permits a precise definition of boundary conditions. This results in a large algebraic system of equations. Transport of ammonia in the blood, modeled by a 1D transport PDE, is coupled with its metabolism in the cells modeled by non-linear first order ODEs, through a passive uptake of ammonia. The metabolism model inside the hepatocytes is calibrated on healthy and drug-induced-damaged liver [2]. Ammonia levels at the liver outlet are predicted, from known input levels. This study has been done for mice and an extension to pig will be investigated. This multi-scale model can be easily extended to any drug, marker or metabolite, as long as its metabolic pathways are known. Keywords: tissue functional unit; ammonia metabolism simulation; sensitivity analysis; compound transport and metabolism; liver; microcirculation References [1] Ghallab A., Cellière C., et. al. (2016), “Model guided identification and therapeutic implications of an ammonia sink mechanism.”, J Hepatol, 64(4), 860-871. [2] Hammad S., Hoehme S., Friebel A., et. al. (2014) “Protocols for staining of bile canalicular and sinusoidal networks of human, mouse and pig livers, three-dimensional reconstruction and quantification of tissue microarchitecture by image processing and analysis.”, Archives of Toxicology, 88(5), 1161-1183. ∗Corresponding author, E-mail: [email protected] Simulation of fractional flow reserve and plaque development in the coronary arteries Nenad Filipovic 1 , Arso Vukicevic 1 , Velibor Isailovic 1 , Dalibor Nikolic 2a , Zarko Milosevic 1 , Igor Saveljic 1 ,Nikola Jagic 3 , Oberdan Parodi 4 1 Faculty of Engineering, University of Kragujevac, 34000 Kragujevac, Serbia 2 BioIRC Bioengineering Research and Development Center, 34000 Kragujevac, Serbia 3 Clinical Center Kragujevac, 34000 Kragujevac, Serbia 4 National Research Council Pisa, Italy Abstract. The objective of this work was to compare computer simulation of plaque development and Fractional Flow Reserve (FFR) with standard angiography patient data. We simulated of the plaque formation with mass transport of LDL through the wall which is coupled with the Navier-Stokes equations, the Darcy equation for model blood filtration and Kedem-Katchalsky equations. Additional three reaction-diffusion equations for the inflammatory process and lesion growth model in the intima were used. Outlet boundary conditions were prescribed as inverse resistance from the corresponding diameter. Hemodynamic FFR threshold was observed for 0.80 and 0.75. We found a good correlation between real and computed FFR results on ten patients (p<0.005). Computer simulation may predict plaque development and non-invasive anatomic and functional assessment of coronary stenosis for each specific patient which may be a clear benefit for patient study. Keywords: plaque development; FFR simulation; coronary arteries References Vukicevic A.M., Stepanovic N.M., Jovicic G.R., Apostolovic S.R., Filipovic N.D. (2014), “Computer methods for follow-up study of hemodynamic and disease progression in the stented coronary artery by fusing IVUS and X-ray angiography”, Med Biol Eng Comput. 52(6):539-56. Fig. 1 Computed FFR and plaque development Corresponding author, Professor, E-mail: [email protected] a Ph.D. Student, E-mail: marko[email protected]s Fluid-structure-interaction model of Transcatheter Aortic Valve Implantation configuration: comparison with an in-vitro study Anna Maria Tango * , Andrea Ducci a and Gaetano Burriesci b UCL Mechanical Engineering, University College London, Torrington Place, London WC1E 7JE, UK Abstract: Transcatheter aortic valve implantation (TAVI) is the treatment of preference for patients who are too weak or ill to undergo major surgery. Though the clinical benefit of the treatment has been demonstrated, some post procedural complications have emerged. In particular, the occurrence of silent ischemic lesions and dementia is considerably higher than with surgical valve replacements (Kahlert et al., 2010). The source of these pathologies is still unclear, but a potential cause has been recently identified as haemodynamic perturbations downstream the valve (Ducci et al., 2013). In fact, contrary to surgical valve replacements, TAVI produces a valve-in-valve configuration which alters the flow in the aortic root, and may establish haemostatic conditions typically associated with thrombus formation. The aim of this study is to expand these findings through a numerical model, allowing to analyse to identify the thromboembolic risk in the various regions of the fluid domain, for different configurations. A fluid-structure-interaction (FSI) approach was chosen, in order to model the interaction of highly deformable structures with pulsatile fluid flows. The analyses were performed using the explicit finite element software LS-DYNA (LSTC, Livermore, CA, USA). The models include the aortic root and the prosthetic valve, in an Eulerian fluid domain. Both the structure and fluid were meshed using ICEM 17.0 (ANSYS, Inc., Canonsburg, PA, USA) and then exported to LSDYNA. An arbitrary Lagrangian–Eulerian (ALE) algorithm was used to implement the coupling between the structural and fluid elements. The numerical results were validated versus the flow patterns measured in vitro with particle image velocimetry (PIV). The study confirms that the TAVI approach has a major impact on the flow pattern, leading to the formation of zones characterised by slow flow velocities which may promote pathological conditions. Once refined, the computational model could be used to predict the haemodynamics in diseased and virtually treated conditions, providing a more appropriate tool for therapeutic planning and for the design of new improved devices. Keywords: Fluid-structure-interaction (FSI), Transcatheter aortic valve implantation (TAVI); ArbitraryLagrangian-Eulerian technique (ALE); Valsalva sinus; blood stagnation References Kahlert, P., Knipp, S.C., Schlamann, M., Thielmann, M., Al-Rashid, F., Weber, M., Johansson, U., Wendt, D., Jakob, H.G., Forsting, M., Sack, S., Erbel, R., Eggebrecht, H., (2010), “Silent and apparent cerebral ischemia after percutaneous transfemoral aortic valve implantation: a diffusion-weighted magnetic resonance imaging study”, Circulation, 121 (7), 870–878. Ducci, A., Tzamtzis, S., Mullen, M.J., Burriesci, G., (2013), “Hemodynamics in the Valsalva sinuses after transcatheter aortic valve implantation (TAVI)”, J.HeartValve , 22 (5), 688–696. ∗Corresponding author, Ph.D. Student, E-mail: [email protected] a Ph.D., E-mail: [email protected] b Ph.D., E-mail: [email protected]