Paper Submission & Registration
8th Dutch Bio-Medical Engineering Conference
12:30   Heart - III
Chair: Bart Verkerke
15 mins
Image-based computational fluid dynamics in the fontan circulations to assess the hepatic flow distribution
Seline F.S. van der Woude, Friso M. Rijnberg, Hans C. van Assen, Joe F. Juffermans, Mark G. Hazekamp, Monique R.M. Jongbloed, Sasa Kenjeres, Hildo J. Lamb, Jos J.M. Westenberg, Arno A.W. Roest, Jolanda J. Wentzel
Abstract: Background Children born with a single ventricle heart defect undergo a series of surgeries resulting in the Fontan circulation allowing for passive blood flow from the body to the lungs without interaction of the heart. A balanced hepatic blood flow distribution (HFD) to the lungs is crucial to prevent malformations of the lung tissue. Currently, HFD is quantified by tracking Fontan tunnel flow to the lungs (Figure 1), assuming hepatic venous (HV) flow to be uniformly distributed within the Fontan tunnel (conventional method). We used computational fluid dynamics (CFD) in models that included hepatic veins to 1) assess mixing of HV flow within the Fontan tunnel and 2) quantify the HFD by tracking HV flow directly from the hepatic veins (novel method) based on CFD in comparison with the conventional method. Methods Patient-specific, time-resolved 3D models of 15 Fontan geometries were generated, including the hepatic veins using MR images. Unstructured polyhedral meshes with 4 prism layers at the wall were generated in ICEM with an element size ranging from 0.4 and 0.5 mm. PC-MRI flow measurements served as in- and output boundary conditions. 7 cardiac cycles were simulated with 1000 time steps per cardiac cycle using ANSYS Fluent (v17.1). The last 4 cardiac cycles were used for post-processing. Mixing of HV within the tunnel, on a scale between 0 (no mixing) and 1 (perfect mixing), was assessed at a caudal and cranial cross section in the Fontan tunnel. HFD was quantified by tracking particles from the caudal (HFDcaudal) and cranial (HFDcranial) cross section within the tunnel and from the hepatic veins (HFDHV) (Figure 1). Results HV flow was non-uniformly distributed at both the caudal (mean mixing 0.66±0.13) and cranial (mean 0.79±0.11) level within the Fontan tunnel. On a cohort-level, differences in HFD between conventional and novel methods were significant but small; HFDHV (51.0±20.6%) versus HFDcaudal (48.2±21.9%, p=0.033) or HFDcranial (48.0±21.9%, p=0.044). However, individual absolute differences of 8.2-14.9% in HFD were observed in 4/15 patients. Conclusion HV flow is non-uniformly distributed within the Fontan conduit. Substantial individual inaccuracies in HFD quantification were observed with potential clinical impact.
15 mins
Ex-vivo study of myocardial deformation imaging using multi-perspective ultrafast ultrasound
Peilu Liu, Hein de Hoop, Hans-Martin Schwab, Richard Lopata
Abstract: The heart is a complex organ with high levels of deformation occurring in different directions. It is difficult to measure the strains in vital parts of the heart precisely when only using a single probe, even at a high frame rate. Therefore, we proposed a multi-probe 2D ultrasound (US) cardiac strain imaging system followed by a novel method to automatically compound strain data acquired simultaneously from the dual probes. The aim is to improve the measurement of myocardial deformation in all directions within a 2D image plane. In an ex-vivo beating porcine heart experiment (PhysioHeart, LifeTec, NL), parasternal long axis, short axis and four-chamber apical views of the left ventricle were acquired with two phased array probes (P4-2V, Verasonics). The dual probes were attached to an arch that ensures they were exactly in the same imaging plane, and allows to adjust the relative angle between the probes. First, probe 1 (P1) was positioned at four-chamber apical view and probe 2 (P2) was rotated in steps of 15˚ from parasternal long axis view to apical view. Next, P1 was positioned to acquire the parasternal short axis view while P2 was rotated every 15˚ from 30˚ to 75˚ with respect to P1. All US data were acquired at a frame rate of 170 FPS, using spherical wave imaging under 11 angles. RF-based strain imaging was performed using the same segmentation mesh for both datasets after automatic registration of the dual probe image sets. Global and regional strains of the two probes were calculated respectively. A strain compounding mask was created to automatically fuse the optimal strain data from both probes. In the short axis view, compounded radial (RS) and circumferential (CS) strains have all been improved in every different relative angle compared to single probes. For the largest angle (75˚), mean drift error MDE (RS: - 38.7%, CS: -35.3%) and strain variability (SV) (RS: -36.8%, CS: -37.1%) decrease significantly, and strain curves reveal less noise for each region. Long axis view strains show similar results. Future work will include 3D multi-probe imaging and first-in-men studies.
15 mins
Robust estimator of the cardiorespiratory coupling in the presence of abnormal beats
John Morales, Pascal Borzée, Dries Testelmans, Bertien Buyse, Sabine van Huffel, Carolina Varon
Abstract: Introduction: In general, methods for non-invasive quantification of the cardiorespiratory coupling combine information from heart rate variability (HRV) and respiratory signals. Abnormal beats alter the reliability of the HRV and thus hinder the evaluation of the cardiorespiratory coupling. This is particularly problematic for methods based on autoregressive models due to their sensitivity to outliers. An alternative approach based on robust regression is proposed in this abstract. Methods: A vector y containing the HRV, and a matrix X with an embedding of the respiratory signal, are constructed. X and y train a weighted Least Squares Support Vector Machines regression model which is used to predict y. Afterwards, the residuals of the prediction (e) are derived. The cardiorespiratory coupling is characterized by η=IQR(e)/IQR(y), where IQR(*) refers to the interquartile range. η measures how good the prediction is. The prediction, in turn, is better for signals with a stronger coupling. Materials: Respiratory and ECG signals from polysomnographic studies of 100 sleep apnea patients were used. Both, HRV and respiration were segmented into 5-minutes epochs. The respiratory component of the HRV, Px, was derived using subspace projections. The linear coupling between HRV and respiration was confirmed with a surrogate analysis applied to Px. Ten ranges of Px were defined to split the segments into 10 groups. Afterwards, the epochs were contaminated with random numbers of simulated ectopic beats and the coupling was evaluated before (Px, η) and after (P_x^c, ηc) contamination. Results and conclusion: Despite significant differences (Kruskall-Wallis, p<0.05) between η and η_x^c before and after contamination, the trends observed in the parameters suggest that this parameter is more robust compared to Px. The trends with Px are affected when more than 2 ectopic beats are present. In contrast, these are maintained when up to 9 ectopic beats are allowed using η. As future work, the application of the method in data with real irregularities needs to be evaluated.
15 mins
Heart rate estimation from motion-contaminated functional near infrared spectroscopy signals
Naser Hakimi, M. Sofía Sappia, Jörn M. Horschig, Willy J. N. M. Colier
Abstract: Monitoring heart rate (HR) is a non-invasive diagnostic method to investigate physiological stability provided by the autonomic nervous system. Elaborate analysis of HR patterns provides an early and sensitive indicator of compromised health. The gold standard of HR monitoring is analysing interbeat intervals determined by means of electrocardiogram. However, this traditional method has several limitations, e.g., movement limitation and skin irritation. Functional near infrared spectroscopy (fNIRS) measures hemodynamics which are not only composed of the brain hemodynamic response, but also include several physiological information. The most prominent physiological information in fNIRS signals are pulsatile fluctuations caused by heartbeats. This information allows for the estimation of HR from fNIRS signals. It is, however, necessary to estimate HR from signals contaminated by motion artefacts to allow subject movement in portable and continuous monitoring. In this study, we propose a novel algorithm for estimating the HR of subjects undertaking physical exercise. The proposed algorithm has three main stages. First, a normalised least mean squares adaptive filter is used to reduce the motion interference from fNIRS signals, using the simultaneously recorded accelerometer signals in three-dimensional space. Second, the filter outputs are combined using Hilbert transform to reconstruct a motion-free fNIRS signal with enhanced spectrum. Third, the reconstructed motion-free signal is used to estimate HR by finding the maximum peak in the median spectrum obtained by implementing multiple autoregressive models. To the best of our knowledge, the proposed algorithm is the first algorithm addressing the motion artifact issue when estimating HR from fNIRS signals. Monitoring HR besides the brain hemodynamic response provided by fNIRS can be desirable for the development of home-based healthcare systems because of mobility, comfort, and cost-effectiveness of fNIRS. In addition, HR estimation from fNIRS during intense physical movement has particular applications in sports monitoring to provide athletes with crucial information about their heart and brain fitness to manage their training.

end %-->