MR Architects, MR System Engineers and MR Software EngineersExtraneous in-band Radio Frequency Interference (RFI) signals can mix with the MR signal of a subject creating artefacts in the image data, usually zipper like artefacts. This can happen if the RF shield/Faraday cage develops a leak or is under-specified in some way. In order to eliminate/reduce the impact of RFI, a variety of de-noising methods [1-4] have been proposed that use dedicated noise pickup coils to sense the ambient noise. The noise signal is usually multiplied by a scale and delay factor [1-2] before it is subtracted from the MR signal. In some cases direct subtraction [3-4] is used. The correction factors are obtained during a pre-scan session for reception [1-2] or transmission [4]. This abstract describes a software de-noising method which eliminates the need of a dedicated noise pickup coil, or even a physical noise sampling coil. The proposed method can be implemented both online and off-line through k-space manipulation without a pre-scan calibration.

Extraneously received in-band RFI co-exists with the subject MR signal in every k-space/channel of a multi-channel receiving (Rx) coil. If the RFI signal would be sampled independently and simultaneously with the reception of MR signal, such sampled RFI reference could be processed together with the “contaminated” k-space. In this way, the RFI component in each channel would be identified and removed. The de-noised k-space data can then be used to reconstruct a MR image without/with reduced zipper artefacts.

The RFI signals may be obtained in two ways. One way is to reuse a standard MR Rx coil as a Sniffer Coil (SC). The sniffer coil is typically placed outside the viable imaging volume in order not to pick up any MR signal. The sniffer coil can be connected to the MR receiving subsystem and works as an additional Rx coil. Both sniffer coil and Rx coil acquire their independent k-space samples. The k-space data of the Rx coil contains the information of both MR signal and RFI signal while the k-space data of the sniffer coil contains RFI that was sampled at the location of the sniffer coil. The alternative way, as demonstrated here, is to use statistical means to extract the RFI information by utilizing the multiple channel outputs of a multi-channel Rx coil. We refer to this as the Virtual Sniffer Coil (VSC) method. Each contaminated channel output contains both MR signal and RFI signal. The correlation of MR information among these channels is different from that of the RFI component. A statistical method, e.g. Principle Component Analysis (PCA), is used to separate the RFI signal cluster from the MR signal cluster.

Due to the spatial distribution of Rx coil channels, and related various path losses/time-of-arrival from the remote RFI source, the RFI signal component is not identical in each channel and different from the RFI reference of either the SC or VSC method. However, all channels are sampled simultaneously. The mapping between RFI reference and the RFI component is modeled by a complex ratio R. The magnitude and phase parts of R compensate the differences in path loss and time-of-arrival at baseband respectively.

In the outer regions of contaminated k-space, the RFI component dominates. As a first order approximation, equation (1) holds.

   [Snif]·R=[RFI_component]≈[Rx]_{outer_regions} (1)

The compensation ratio R can be calculated, using the least squares method, line-by-line in k-space, i.e. one R per TR, in equation (2).

    min ||[Rx]_{outer_regions} - [Snif]·R||₂²   (2)

De-noising experiments were performed using a Philips Multiva 1.5T MRI with a controlled external RFI source. Fig. 1 displays a single channel image in the VSC de-noising experiment from a Philips 8-channel SENSE head coil. An AM RFI with an in-band carrier frequency was introduced during the experiment. The k-space data and the estimated RFI component in this experiment are compared in Fig. 2. It is found that their triangular AM profile match well in the outer regions of k-space. Fig. 3 shows the setup of a de-noising experiment using the SC method. One coil loop is placed on the top of a phantom, which travels into the bore. Another coil loop is left outside the bore on the table and performs the role of a SC. A CW RFI was used in the SC de-noising experiment. Fig. 4 presents the result of SC de-noising method.

Two experimental setups comparing the SC and VSC methods demonstrate their de-noising effect. The de-noised images appear very close to those without a RFI signal. The de-noising performance could be improved when a more sophisticated approximation is adopted in RFI component estimation.

The SC and VSC method can be used at a clinical site, in the event that RFI occurs and a technologist is unable to resolve it in a short time. In particular, the VSC method is suitable for retrospective correction. Both methods may help technologists continue their tight MR scan schedule with satisfactory image quality.