Magnetic Field Gradient Artifact Reduction on Ecg for Improved Triggering

INTRODUCTION: Due to heart motion, cardiac MRI is made difficult and image acquisitions have to be synchronized with heart activity to suppress cardiac motion artifacts. Electrocardiogram (ECG) is therefore the state-of-the-art signal [1], each MRI acquisition being launched after a fixed delay following a QRS complex detection. The complex MRI environment highly distorts ECG signals, due to the presence of a high static magnetic field and mainly to fast switching Magnetic Field Gradients (MFG). Many hardware developments have been suggested to reduce these undesirable effects, but MFG artifacts remain an open problem and require specific signal processing methods. Two avenues of research have been explored: (a) first solution has been based on the development of MR specific QRS detectors. A well-known solution has taken advantage of the vectocardiogram (VCG) representation of heart activity [2]. (b) Second class of methods consists in MFG artifact suppression. MFG artifacts are the responses of the MFG through a linear time invariant (LTI) system, modeled as three Finite Impulse Response (FIR) filters, one for each direction of the MFG [3]. The MFG artifact suppression relies on the estimation of these three FIR filters and can be achieved with adaptive filtering [4]. One major drawback of this method is that the ECG is considered as noise in the modeling, adaptive filtering tends to cancel all contributions as soon as MFG are played. ECG signal can thus be altered when MFG artifacts overlap QRS complexes. In order to overcome this limitation, a new MFG artifact suppression method, which takes the ECG signal into account during the FIR filter estimation, is presented. METHOD: ECG Model: An accurate ECG denoising has recently been proposed [5], by using an ECG model. ECG is considered as a pseudo-periodic signal, where each ECG cycle can be modeled as a sum of five Gaussian functions. ECG denoising consists in an online estimation of the model parameters, which can be performed with Bayesian filtering. This filtering technique requires the observation of two signals, the ECG signal and the cardiac phase, which is created after the QRS detection. A linear phase between –π and π is then assigned between two consecutives R waves. The Bayesian filtering can thus only be performed with at least one cardiac cycle delay. ECG+MFG model: ECG acquired during MRI is mainly distorted by MFG artifacts, these artifacts have been modeled as outputs of three FIR filters. Artifact reduction is currently performed by estimating these filters while considering ECG signal as noise. The herein presented method suggests a new model by merging the MFG artifact and ECG models. This new modeling can be written in a state-space formulation, as in [5], where the FIR filters are integrated in the state vector and the observation equation takes the MFG artifacts into account. A recursive estimation of all model parameters can be performed thanks to nonlinear Bayesian filtering, namely Extended Kalman Filter (EKF) (Fig. 1). As the observation of the cardiac phase is required for the parameter estimation, the ECG signal estimate cannot be directly used for triggering purpose. In order to deal with the severe time constraints of triggering, the MFG artifact suppression can be applied on a semi-online way, by filtering the ECG signal with the FIR filter parameters estimated during the previous cardiac cycle. Let assume that at time n the last EKF update has been performed at time k, (meaning the estimation has an n-k+1 delay). The ECG can be processed by suppressing the MFG artifact estimation computed by filtering MFG signals with FIR filters estimated at time k as: where zn is the denoised ECG, hk the FIR filter estimated at time k and gj,n, the j MFG signal at time n (Fig. 2).