Outline for the New ASME Standard for the Verification and Validation of Computational Fluid Dynamics (CFD) Simulation in Medical Devices Marc Horner, Ph.D. ANSYS, Inc. 1
CFD Sub-group Members Marc Horner, ANSYS, Inc. (Lead) Dawn Bardot, HeartFlow Jeff Bodner, Medtronic Ricky Chow, Lake Region Medical Kristian Debus, CD-Adapco* Prasanna Hariharan, US FDA Vinod Kaimal, Materialise* 2
V&V20 Conceptual Model 3
V&V20 Conceptual Model Geometry via imaging has it own uncertainty Sometimes hard to estimate, sometimes in another species Not given by most published literature, some inputs are just guesses 4
Errors Involved in Geometry Generation Detection and Characterization of ICAs w/ MRA: Comparison of Volume Rendering and Max Intensity Projection Algorithms 5 Mallouhi et al. AJR. 2003; 180: 55 64.
Errors Involved in Geometry Generation 6 Mallouhi et al. AJR. 2003; 180: 55 64.
Errors Involved in Geometry Generation 7 Chow. Unpublished.
Errors Involved in Geometry Generation CFD Analysis in an Anatomically Realistic Coronary Artery Model 8 Goubergrits et al. Int J Cardiovascular Imaging. 24: 411 421.
Errors Involved in Geometry Generation CFD Analysis in an Anatomically Realistic Coronary Artery Model 0.65 +/- 0.79 Pa 0.68 +/- 0.92 Pa 0.67 +/- 0.88 Pa 9 Goubergrits et al. Int J Cardiovascular Imaging. 24: 411 421.
Potential Example Problems 1. Nozzle model (focus on validation) - Based on the FDA Round Robin - https://fdacfd.nci.nih.gov/interlab_study_1_nozzle 2. Blood damage modeling (initial focus on verification) - Use analytical solutions to verify implementation of hemolysis index equation - Possible extension to validation (Prasanna checking with Rich Malinauskis @ FDA) 3. Drug delivery to the back of the eye (focus on verification) - Have an analytical solution for drug delivery to the retina from a centrally placed bolus 10
Example 1 FDA CPI Project (Improving the use of CFD to Evaluate Device Performance and Blood Damage Safety) Interlaboratory evaluations to assess use of CFD in a series of device models (open to all participants) Create a data repository of results for public use Contribute to standards and CFD guidance documents PHASE 1: Simple nozzle model PHASE 2: Blood pump model 12 mm 4 mm 12 mm rotor design Outlet 11 CFD modelers: 28 groups from 6 countries Validation of velocity fields using quantitative flow visualization techniques: 3 labs Blood damage testing of the device: 3 labs Inlet Opening Summer 2012 blood testing
Comparison of CFD and Experimental Results Results for Axial Velocity Component along the Centerline Lines: CFD Symbols: Experiment Laminar Flow (Re = 500) Turbulent Flow (Re = 3500) 12
Data Repository for FDA s CFD Round Robin Access to experimental data as a function of Reynolds number (and eventually flow direction) Access to all publications and presentations 13 https://fdacfd.nci.nih.gov/ Contact Information sandy.stewart@fda.hhs.gov richard.malinauskas@fda.hhs.gov
Example 2 - Blood Damage Modeling ( Hb) Hb (%) 3.62E 5 2.416 t 0.785 exp 14
Verification Study HI = constant Substituting HI' = HI 1/ 15 HI ' A 1/ / We can now set t = t and integrate: HI ' A 1/ / t t Then transform back to HI before reporting the damage at the outlet. true value
Hemolysis Index Sample Case - Planar Couette Flow 2.50E-06 2.00E-06 shear strain rate (du/dy) viscosity shearing stress density 90 m^2/s 0.0035 kg/m/s 0.315 Pa 1050 kg/m^3 Analytical CFX-Eulerian Fluent-Eulerian 1.50E-06 Fluent-Lagrangian 1.00E-06 5.00E-07 0.00E+00 0 0.001 0.002 0.003 0.004 0.005 0.006 Y (m) 16
Example 3 - Drug Delivery to the Retina Drug delivery to an idealized version of the vitreous region of a rabbit eye Model includes: Infinite drug sink Diffusion Choroid sink Partitioning sclera vitreous choroid retina sclera C sclera /C choroid = 0.33 17 vitreous C choroid /C retina = 3.0 drug bolus Missel and Horner, ARVO 2007
Verification Study ANSYS Fluent results compared to analytical solution generated in FlexPDE 18
APPENDIX 19
Proposed Outline - Introduction Introduction to Validation Methodology General Objective and Scope Errors and Uncertainties: o Should be estimated not just for velocity or temperature but also for quantities more relevant to device safety and performance (ex: shear stress, hemolysis index, thermal dose) Example for Validation Nomenclature and Approach Validation Approach Overview of the unique difficulties related to Biomedical Models Overview of Subsequent Sections References 20
Proposed Outline - Code Verification and Solution Verification Code Verification and Solution Verification General Introduction Code Verification Solution Verification (e.g. drug delivery to the eye) Code and Solution Verification in a customized environment User-defined functions (e.g. heat source for hyperthermia) User-defined variables (e.g. hemolysis index) Special Considerations Final Comment References 21
Proposed Outline - Effect of Input Parameter Uncertainty on Simulation Uncertainty Effect of Input Parameter Uncertainty on Simulation Uncertainty Introduction Sensitivity Coefficient (Local) Method for Parameter Uncertainty Propagation Sampling (Global) Methods for Parameter Uncertainty Propagation Importance Factors Special Considerations including guidance on dealing with inputs obtained from literature Final Comment on Parameter Uncertainty References 22
Proposed Outline - Effect of Geometry Uncertainty on Simulation Uncertainty Effect of Geometry Uncertainty on Simulation Uncertainty - Tolerances in making medical devices (e.g. how sharpness affects blood damage) - Errors involved in extracting geometries from different imaging modalities - Accounting for patient-to-patient variability in flow geometry 23
Proposed Outline - Uncertainty of an Experimental Result Uncertainty of an Experimental Result Overview Experimental Uncertainty Analysis Uncertainty of Validation Experiment Summary References 24
Proposed Outline - Evaluation of Validation Uncertainty Evaluation of Validation Uncertainty Overview Guidelines for validation using data obtained from animal experiment or bench-top Theme: the model is only validated for the system/species for which validation data exists. Example: a model validated against canine data is valid for canines. A model validated against a rubber vein is valid for a rubber vein. Saying any more than that is the subject of prediction per the final section of the standard. Guidelines for estimating the uncertainty of geometry-generation via segmentation of imaging data Estimating uval When the Experimental Value, D, of the Validation Variable is Directly Measured (Case 1) Estimating uval When the Experimental Value, D, of the Validation Variable is Determined From a Data Reduction Equation (Cases 2 and 3) Estimating uval When the Experimental Value, D, of the Validation Variable is Determined From a Data Reduction Equation That Itself Is a Model (Case 4) Assumptions and Issues References 25
Proposed Outline - Prediction Beyond Validated Range Prediction beyond validated range 26
Proposed Outline - Examples Examples 1. Nozzle model (focus on validation) - Based on the FDA Round Robin - https://fdacfd.nci.nih.gov/interlab_study_1_nozzle 2. Blood damage modeling (initial focus on verification) - Use analytical solutions to verify implementation of hemolysis index equation - Possible extension to validation (Prasanna checking with Rich Malinauskis @ FDA) 3. Drug delivery to the back of the eye (focus on verification) - Have an analytical solution for drug delivery to the retina from a centrally placed bolus 27