Learning Patient-Specific Lumped Models for Interactive Coronary Blood Flow Simulations
We propose a parametric lumped model (LM) for fast patient-specific computational fluid dynamic simulations of blood flow in elongated vessel networks to alleviate the computational burden of 3D finite element (FE) simulations. We learn the coefficients b
- PDF / 1,760,464 Bytes
- 9 Pages / 439.363 x 666.131 pts Page_size
- 116 Downloads / 206 Views
Philips Research, Hamburg, Germany GRAD, CT, Philips Healthcare, Haifa, Israel Clinical Science, CT, Philips Healthcare, Cleveland, Ohio, USA 2
3
Abstract. We propose a parametric lumped model (LM) for fast patientspecific computational fluid dynamic simulations of blood flow in elongated vessel networks to alleviate the computational burden of 3D finite element (FE) simulations. We learn the coefficients balancing the local nonlinear hydraulic effects from a training set of precomputed FE simulations. Our LM yields pressure predictions accurate up to 2.76mmHg on 35 coronary trees obtained from 32 coronary computed tomography angiograms. We also observe a very good predictive performance on a validation set of 59 physiological measurements suggesting that FE simulations can be replaced by our LM. As LM predictions can be computed extremely fast, our approach paves the way to use a personalised interactive biophysical model with realtime feedback in clinical practice. Keywords: CCTA, coronary blood flow, lumped parameter biophysical simulation, patient specific model.
1
Introduction
Fractional flow reserve (FFR) based on invasive coronary angiography is the goldstandard for the assessment of the functional impact of a lesion, thus helping
Fig. 1. From image to simulation. The heart and its coronary arteries are (automatically) segmented from a CTA scan yielding a tree representation of centerline points and polygonal cross-sections (area encoded as color) used to conduct a patient-specific blood flow simulation (FFR encoded as color). c Springer International Publishing Switzerland 2015 N. Navab et al. (Eds.): MICCAI 2015, Part II, LNCS 9350, pp. 433–441, 2015. DOI: 10.1007/978-3-319-24571-3_52
434
H. Nickisch et al. LCX
D1
M1
D11
D2
In/LAD
ϕ(f )
LAD
Septal
Fig. 2. Parametric nonlinear lumped model with n = 21 elements and m = 15 nodes including ground. Based on the centerline representation, we set up a lumped model with nonlinear resistances. The black boxes indicate inflow and outflow boundary conditions. The white tubes representing tree segment transfer functions ϕ(f ) are composed of a series of linear and nonlinear resistance elements reflecting both the local vessel geometry and hydraulic effects.
with clinical decisions for revascularization [14]. More recently, patient-specific simulations of physiologic information from the anatomic CCTA data have been proposed [7,16]. These computational fluid dynamics models are based on 3D finite element (FE) Navier-Stokes simulations, and are challenging both in terms of computation and complexity. Computation alone can be accelerated by performing them on GPUs, and complexity can be cut down by reduced order models e.g. [3,6,5], reuse of precomputations or lattice Boltzmann [10] methods. Operating a simulation pipeline from image to prediction (see Figure 1) in a failproof way is challenging as FE computations are sensitive to the quality of the underlying mesh. Besides complexity reduction, simpler and more robust models could be more appropriate for statistical r
Data Loading...
