Acoustic noise in MRI originates from rapid switching of magnetic field gradients, which causes strong Lorentz forces that make the gradient coils vibrate. While mechanical countermeasures such as acoustic damping materials and improved hardware design have been used to help reduce sound levels, the use of earplugs or headsets is still typically required to protect hearing. Even with these measures, gradient noise can remain a source of discomfort and impaired communication between subject and operator during scanning. Other approaches to noise reduction have focused on the pulse sequence, e.g. [1]; one method previously applied to fast-spin-echo and gradient-echo pulse sequences [2], is to minimize the harmonic content of the gradient waveforms through the use of sinusoids. Here we demonstrate the effectiveness of this method in a new silent T2*-weighted multi-echo 3D-gradient-echo sequence (SWAN, GE Healthcare) for quantitative susceptibility mapping (QSM).

The acoustic response of whole-body MRI systems is typically negligible below ~200Hz, and so, if the frequency content of the pulse sequence is kept within this range, sound levels from the scan can be substantially reduced. To achieve this, all but the weak excitation slab-select trapezoidal gradients were replaced with sinusoids, to minimize harmonic content. In order to compensate for the resulting non-uniform k-space sampling, the readout resolution was oversampled; the sampling bandwidth was accordingly increased such that the total duration of the readout was unchanged. The commonly used Kaiser-Bessel kernel was used in the gridding algorithm.

A 3.0T MRI clinical system (Discovery MR750, GE Healthcare) was used for all experiments. In a standard head phantom, the fraction of the sine lobe sampled was investigated. The sound pressure level (SPL) was recorded while varying the sine lobe fraction, at three different fields of view (FOV). The SPL was measured using a Bruel & Kjaer 2250 sound-level meter with microphone placed on the patient table ~1m from the end of the bore.

Three healthy subjects were scanned with the following two protocols detailed in the table (Figure 1): (A) standard SWAN and (B) silent SWAN. Silent SWAN was optimized to match the average TE (~20ms) of the standard acquisition. Due to the increased echo spacing (in order to maintain low frequency content of the pulse sequence), effective flow signal preservation was difficult to achieve in some blood vessels with first-order flow compensation (due to higher order motion). Thus, flow compensation was not used for silent SWAN.

The output DICOM data were processed using the MEDI software (Medimagemetric LLC [3,4]) for QSM maps. The maps were qualitatively compared and the difference within the caudate nucleus was computed from a region of interest (ROI).

Figure 2 demonstrates phantom results while varying the fraction of the sine lobe that is sampled. As the readout oversampling meets the Nyquist criterion, no significant artifact is seen for all fractions. As expected, the SPL increases with the fraction of sine-lobe sampled; an increased fraction results in decreased echo spacing, with all other parameters kept the same, and thus increased frequency content in the pulse sequence.

For the volunteer experiments, ambient SPL was 62±1 dBA. For the standard SWAN protocol (Figure 1), SPL was 104 dBA, and for the silent protocol was 73 dBA. (For lower-resolution protocols, levels were lower still, e.g., 67 dBA for 416×256×64, FOV 24x19.2 cm, with 0.5 sine-lobe fraction). QSM images from a representative subject are shown in Figure 3. The maps from silent SWAN appear slightly noisier. The mean of the susceptibility values in the caudate nucleus from standard and silent acquisitions were 49.4±8.6 parts per billion (ppb) and 49.4±15.5 ppb, and the mean paired difference was 0.1±24.1. Some larger differences were observed at the periphery and blood vessels and could be due to a combination of spatial misregistration (since no computational registration was used) and lack of flow compensation in silent SWAN.

We have demonstrated the application of sinusoidal gradients to a multi-echo gradient-echo sequence for silent QSM. The silent acquisition costs some SNR and acquisition time for the benefit of patient comfort and workflow ease. The quantitation is largely unaffected apart from the increased noise, likely due to the fewer number of echoes in the silent acquisition. The relatively weak signal in areas of flow for the silent acquisition may have contributed to some difference in susceptibility values between standard and silent acquisitions in those locations. The actual SPL is dependent on many parameters, including the fundamental gradient period and strength, and has a trade-off on the acquisition efficiency (i.e., echo spacing and number of echoes).