Filtering of pathological ventricular rhythms during MRI scanning

Filtering of pathological ventricular rhythms during MRI scanning

Julien Oster

, Matthieu Geist

, Olivier Pietquin

, Gari Clifford

1Institute of Biomedical Engineering, Department of Engineering Science, University of Oxford, UK,

2IMS Research Group, Sup´elec, Metz, France.

Introduction

I Electrocardiogram (ECG) acquired for 1. Patient monitoring¹.

2. MRI acquisition synchronization with heart activity².

I More and more applications need accurate ECG analysis during MRI:

. High or Ultra-High field cardiac applications³, . Interventional MRI⁴,

. Intra-Cardiac Electrophysiolgy guided by real-time MRI⁵, . Cardiac Stress-tests during MRI⁶.

Figure 1: Example of ECG distortion by the MHD effect.

I ECG is highly distorted by the static magnetic field (B₀). The flow of charged fluid (blood) inside a magnetic field induces the creation of an electrical field as a consequence of the Lorentz Force (F = qv ∧ B). This phenomenon is known as the MagnetoHydroDynamic (MHD) effect⁷.

Objective

Development of a technique for the suppression of the MHD effect and accurate analysis of ECG signal, especially during pathological ventricular repolarization.

Methods

I Data:

. Artificial ECG signals⁸, with pathological

ventricular repolarization

I QT elongation

I T wave inversion

. Modeling of the MHD

based on blood flow MRI measurements⁹.

. Real noise data from the

NSTDB^10,11,12. ^{Figure 2:} Flowchart of the model-based filtering.

I Bayesian filtering, applied for denoising and source separation, derived from a set of equations, evolution (1) and observation (2):















θ_k = (θ_k−1 + ωδ) mod 2π

W_k = − X

i

δω∆θ_i,k−1^E b^E_i,k−1²

G

α^E_i,k−1, ∆θ_i,k−1^E , b^E_i,k−1

+ W_k−1 + η_W,k ψ_i,k^E = ψ_i,k−1^E + ε_ψ,i

,

with ψ_i,k^E ∈ {α^E_i,k, b^E_i,k, ξ_i,k^E } and W_k ∈ {P_k, Q_k, T_k, M_k} and (1)











ϕ_k = θ_k + v_1,k

s_k = P_k + Q_k + T_k + M_k + v_2,k s_k = M_k + X

G

α^E_i,k, ∆θ_i,k^E , b^E_i,k

+ v_3,k

, (2)

. θ_k the angular position,

. ω = 2π/RR the angular speed, . δ the sampling period,

. ∆θ_i,k−1 = (θ_k−1 − ξ_i,k−1) . G(a, b, c) = a exp(− b²

2c²) . s_k is the ECG signal.

. ϕ_k is an artificial piecewise linear

phase signal. ^{Figure 3:} Flowchart of Bayesian filtering.

. P, Q, T and M represent the P wave, the QRS complex, the T wave and the MHD effect respectively,

.

α^E_i,k, ∆θ_i,k^E , b^E_i , ξ_i,k^E

are the Gaussian parameters for the ECG, .

α^M_i,k, ∆θ_i,k^M, b^M_i , ξ_i,k^M

those for the MHD effect.

I Automatic ECG annotations with ECGpuwave¹³.

Results

Figure 4: Denoising of the simulated acquisition of an ECG during MRI, with a prolonged QT (left) and T wave inversion (right) (blue: raw ECG with MHD noise, green: original (clean) ECG without MHD, red: denoised ECG).

Figure 5: Annotations of the T wave for the T wave inversion example.

I T wave inversion detected 14 cycles after the event.

I Prolongation of the QT interval can be detected almost immediately.

I QT slightly over-estimated due to the presence of residual noise.

I Mean absolute difference is 34.3ms ± 25.9 over the whole segment (only 22.1ms ± 14.3 before the QT prolongation and 59.5ms ± 26.2 after).

I Over-estimation within human error¹⁴ before the elongation, but slightly larger than human error after the QT prolongation.

Figure 6: Estimation of the QT interval on clean and denoised ECG by the Ecgpuwave software.

Discussion

I Limitations:

. Time of convergence for the filter ⇒ missed transient events ?

. Over-estimation of QT segment ⇒ use Bayesian filter for extracting fiducial points.

. Artificial data do not model changes in blood flow ⇒ application of the technique on real data (induced ischemia).

I Promising technique lays foundations for accurate ECG analysis in hostile environment such as during MRI scanning.

References

1. F.G. Shellock, Magnetic Resonance Procedures: Health effect and safety, 2001.

2. A.D Scott, et al., Radiology, 250:331–351, 2009.

3. T. Niendorf, et al., Eur. Radiol., 20(12):2806–2816, 2010.

4. R.J. Lederman, Circulation, 112(19):3009–3017, 2005.

5. M. Jekic, et al., Journ. of Cardiov. Magn. Res., 10(1):3, 2008.

6. M. Gutberlet, et al., Journ. of Cardiov. Magn. Res., 14(Suppl 1):205, 2012.

7. J.R. Keltner, et al., Magn. Res. Med., 16:139–149, 1990.

8. G.D. Clifford, et al., Phys. Meas., 31:595–609, 2010.

9. J. Oster, et al., In Proc. of the annual meeting of the ISMRM, 2012.

10. G.B. Moody, et al.. Computers in Cardiology, 11(3):381–384, 1984.

11. A. Goldberger, et al., Circulation, 101:215–220, 2000.

12. G.D. Clifford, et al., Phys. Meas., accepted for publication, Aug/Sep 2012.

13. R. Jan´e, et al., In Computers in Cardiology, 295–298, 1997.

14. I. Christov, et al., BioMedical Engineering OnLine, 5(1):31, 2006.

Acknowledgement

I This study has been funded through a Newton International Fellowship (round 2010 by the Royal Academy of Engineering, Grant: 93/914/N/K/EST/DD-PF/tkg/4004642).

http://www.ibme.ox.ac.uk [email protected]