An efficient pipeline from medical images to numerical simulations

1. An efficient pipeline from medical images to numerical simulations Tiziano Passerini, PhD Department of Mathematics and Computer Science Emory University tiziano@mathcs.emory.edu mathcs.emory.edu/~tiziano • Computer simulations • High performance and cloud computing: evaluating the cost-effectiveness of on-demand computing facilities. • Software: LifeV (www.lifev.org) • Publications: • Slawinski, Passerini, Villa, Veneziani, Sunderam, “Experiences with target-Platform Heterogeneity in Clouds, Grids, and On Premise Resources”, Proc. IPDPS-HCW 2012. • Passerini, Slawinski, Villa, Veneziani, Sunderam, “Experiences with a computational fluid dynamics code on clouds, grids, and on-premise resources”, submitted, JPDC 2012. • Partners: • J.Slawinski, U.Villa, Dr. A.Veneziani, • Dr. V.Sunderam (Emory Math&CS) Image segmentation and morphological analysis Characterize the geometryof vascular structures. Software: VMTK (www.vmtk.org). Ongoing project: classify subject-specific morphologies of the infrarenal aorta; relevance of local morphology to the measured flow patterns. Partners: M.Piccinelli, Dr. J.Oshinski (Emory Radiology); Dr. A. Veneziani (Emory Math&CS); A.Smolensky, Dr. W.R. Taylor (Emory Cardiology). Aorta-R Iliac 145.5º Aorta-L Iliac 150º Patient data (medical images, diagnostics…) Cost vs time of a benchmark CFD simulation on different computing platforms.Puma, Ellipse and Lonestar (traditional computing clusters); EC2 and Rockhopper (on-demand HPC cloud-computing facilities). For each platform we report the number of computing cores yielding the fastest computation. T* is the average computing time across platforms. On demand facilities can be extremely cost-efficient. R Iliac-L Iliac 64.5º Left: Magnetic Resonance Angiography of the infrarenal aorta of a healthy male subject; reconstruction of the vessel lumen by means of a level set-based segmentation. Right: Centerline-based calculation of the angles between the aorta and iliac arteries at the bifurcation point. Image segmentation and morphological analysis • Numerical methods Data mining and knowledge extraction Statistical identification and classification of morphological & fluid-dynamics features of the arterial system. Publications: Passerini, Sangalli, Vantini, Piccinelli, Bacigaluppi, Antiga, Boccardi, Secchi, Veneziani, “An Integrated Statistical Investigation of Internal Carotid Arteries of Patients affected by Cerebral Aneurysms”, CVET, 2012. Partners: Dr. S.Vantini, Dr. L.Sangalli, P.Secchi (Politecnico di Milano); M.Piccinelli (Emory Radiology); Dr. A.Veneziani (Emory Math&CS) (m/s) The three plots show the values of wall shear stress (WSS), radius (R) and curvature (C) computed in the internal carotid artery of 52 patients, as functions of the spatial location along the artery (s). The features of the curves are statistically associated to the presence, position and rupture status of cerebral aneurysms in the brain. Verification and validation of the CFD methodology. Ongoing project: FDA’s benchmark (Computational Round Robin #1: Nozzle) Partners: U.Villa, Dr. A.Veneziani (Emory Math&CS); Dr. A.Quaini, Dr. S.Canic (U Houston) Hypothesis Flow in a cylindrical nozzle with conical collector and sudden expansion. Reynolds number in the throat: 3500. Direct numerical simulation with second order tetrahedral finite elements. Mesh refinement based on the nozzle diameter. Computer simulations Data mining knowledge extraction Cost of the computing resource (in $) Database WSS • Mathematical models • Geometrical multiscale models for the • cardiovascular system; • Reduced models for the dynamics of • vascular networks. • Publications: • Alastruey, Passerini, Formaggia, Peiró, “Physical • determinants of the arterial pulse waveform: • theoretical analysis and calculation using the 1-D • formulation”, J Eng Math, 2012. • Passerini, De Luca, Formaggia, Quarteroni, Veneziani, “A 3D/1D geometrical multiscale model of cerebral vasculature”, J Eng Math, 2009. • Partners: Dr. P.Blanco (LNCC, Brazil), Dr. L.Grinberg (Brown University), M.Piccinelli (Emory Radiology), Dr. A.Veneziani (Emory Math&CS). s Numerical methods R Database Store, classify, query large and complex data sets. Project: AneuriskWeb (ecm2.mathcs.emory.edu/aneuriskweb), a public, free repository of medical images, vascular geometries, hemodynamics simulations. The database can be browsed or queried based on geometric and hemodynamic features. Partners: M.Piccinelli (Emory Radiology), Dr. L.Antiga (Orobix), Dr. A.Veneziani (Emory Math&CS) Sample data from the database. Velocity streamlines in a giant aneurysm of the internal carotid artery. In red: the aneurysm neck. Total runtime (in hours) Pressure dynamics in the human physiological right arm, modeled by means of a 1D blood flow model. Numerical solver based on the LifeV library. Geometry and model parameters courtesy of Dr. P. Blanco. Mathematical models & analysis s C Acknowledgements: Support from Emory School of Medicine (Dr. W.R.Taylor, Dr. H.Samady) Support from Emory’s Strategic Initiative“Computational & Life Sciences” Support from The Brain Aneurysm Foundation (PI: Dr. A.Veneziani) s • Testing of hypotheses / Device design Find fluid dynamics parameters that can be predictors of pathological events: study the wall shear stress as a marker of endothelial dysfunction in the coronary circulation. Velocity streamlines at the systolic peak in baseline conditions and in two different simulated post-surgical scenarios. The flow distribution in the sovra-aortic branchesand flow stagnation at the aortic valve are sensitive to the cannula placement. Oscillatory shear index on the wall of the infrarenal aorta of a healthy subject in baseline conditions and after simulated left above knee amputation (inflation of a pressure cuff at the left thigh). OSI increases on the frontal and left sides of the distal arterial wall, suggesting a possible flow-induced localization of vascular diseases (e.g. aortic abdominal aneurysm). In patients with left coronary artery disease, arterial segments with endothelial dysfunction (ED) show significantly lower time average wall shear stress (TAWSS) with respect to those with normal endothelial function. A sign of ED is a reduced radius after acetylcholine injection. proximal Post treatment Pre treatment OSI (difference, Amputation - Baseline OSI (healthy, simulated amputation OSI (healthy, baseline distal TAWSS (dyn/cm2) Partners: M.Piccinelli (Emory Radiology); M.Corban, Dr. H.Samady (Emory Cardiology); Dr. A.Veneziani (Emory Math&CS); Dr. L.Timmins, Dr. D.Giddens (Georgia Tech) Radius (mm) Radius (mm) Sensitivity analysis of blood flow patterns to local morphological features (angle between bifurcating arteries) and systemic conditions (increased peripheral resistances). Partners: A.Smolensky, Dr. W.R.Taylor (Emory Cardiology); M.Piccinelli, Dr. J.Oshinski (Emory Radiology); Dr. A.Veneziani (Emory Math&CS) Parametric study on the positioning of the cannula in left ventricular assist device (LVAD) patients, and its effects on blood features. Partners: Dr. A.Veneziani (Emory Math&CS), M.Piccinelli (Emory Radiology); Dr. D.Gupta (Emory Cardiology)