Image quality and subject experience of quiet T1-weighted 7-T brain imaging using a silent gradient coil – European Radiology Experimental

Study population

To achieve the most MRI ‘naïve’ experience, five healthy subjects (aged 21–29 years; three females) with no previous 7-T MRI experience were included. This study was approved by the local ethics committee and in compliance with national legislation and the Declaration of Helsinki; all subjects provided written informed consent.

Experimental setup

Subjects were imaged with both the quiet and conventional sequence during the same MRI examination. MRI examinations were performed with the silent gradient coil (Futura Composites, Heerhugowaard, the Netherlands) positioned in a 7-T MRI scanner (Achieva, Philips, Best, the Netherlands). The silent gradient coil was fitted with a 32-channel receive array (Nova Medical, Wilmington, MA, USA).

The silent gradient coil consists of a resonant single-coil head insert gradient coil combined with an audio amplifier that enables ~20 kHz switching with adequate power [14]. This gradient insert operates in the z-direction (feet-head), features an integrated radiofrequency transmit coil and can be switched off between examinations (Fig. 1a). The silent gradient coil can in principle operate at a maximum gradient amplitude and slew rate of 40 mT/m and 5,200 T/m/s. In comparison, a conventional whole-body gradient system operates at a gradient amplitude of around 40 mT/m and is limited by peripheral nerve stimulation to a maximum slew rate of 200 T/m/s. However, the small size of the silent gradient coil produced no noticeable peripheral nerve stimulation despite the order of magnitude higher slew rate. In this work, the silent gradient coil was driven at a gradient amplitude of 28.6 mT/m to limit heating of the audio amplifier due to the high duty cycle of the MPRAGE sequence.

a The silent gradient coil used in this work (indicated by the red arrow). b Sequence diagrams of the readout of the quiet and conventional MPRAGE. The quiet MPRAGE features lower slew rates and amplitudes to limit sound from the audible gradients and incorporates an extra silent gradient during the readout to improve imaging efficiency. MPRAGE Magnetisation prepared rapid gradient-echo 

Imaging protocol

Both sequences featured a field of view of 240 × 240 × 172 mm³ and 1.0 mm isotropic resolution. The quiet sequence used optimised imaging parameters and a gradient mode to reduce sound, while the conventional sequence used standard clinical parameters and gradient mode. The sequences differed primarily in their TE and repetition time (TR) which were 8.9 ms and 17.6 ms for the quiet sequence and 1.9 ms and 4.2 ms for the conventional sequence, respectively. The acquisition time was 2:44 min:s for the quiet sequence and 2:24 min:s for the conventional sequence. Other imaging parameters can be found in Table 1. Images were reconstructed offline in MATLAB (MathWorks, Natick, MA, USA) using an iterative sensitivity encoding, SENSE, reconstruction. For the quiet sequence, the spatiotemporal behaviour of the oscillating gradient field was characterised using a field camera (Skope, Zürich, Switzerland) and used as an input for the reconstruction.

The quiet sequence featured a silent readout module consisting of a silent 20 kHz readout gradient that was applied simultaneously with the whole-body encoding gradients of a conventional MPRAGE sequence. The acoustic noise is reduced by using a reduced slew rate and gradient amplitude for the whole-body encoding gradients, which generally leads to longer repetition time and acquisition time. However, the silent readout provides extra spatial encoding during each readout without introducing extra acoustic noise, leading to fewer encoding steps to form an image and therefore reducing the total acquisition time. In summary, this approach reduces the acoustic noise while minimally affecting the acquisition time. The silent readout module was combined with a controlled aliasing in parallel imaging, CAIPI, sampling pattern to limit image noise enhancement from variations in sample density introduced by the rapidly oscillating silent gradient [16]. A schematic representation of the sequence is displayed in Fig. 1b.

Importantly, the slow switching of the whole-body gradient still resulted in a longer TE and TR during the quiet sequence, which, when not addressed, results in suboptimal grey-white matter contrast and cerebrospinal fluid (CSF) nulling. Therefore, we performed signal simulations using extended phase graphs, EPG, which allowed us to simulate the grey-white matter contrast and CSF nulling for a range of TEs, TRs, and flip angles. The quiet sequence was simulated for a range of flip angles between 1 and 90°. The flip angle that generated the contrast that most closely matched the contrast in the conventional sequence was then chosen (assuming no radiofrequency inhomogeneities).

The inversion pulse determines the nulling of the CSF signal and grey-white matter contrast. In particular, the spatial homogeneity of the transmit radiofrequency field (B₁) strongly influences the effectiveness of the inversion pulse. At higher magnetic field strengths (> 7 T), the B₁ field becomes more inhomogeneous, resulting in a spatially varying image contrast. To ensure a more homogenous image contrast, we have implemented a time-resampled frequency-offset corrected inversion, TR-FOCI, inversion pulse [17], which is less sensitive to B₁ field inhomogeneities than conventional inversion pulses. This inversion pulse was used for both the quiet and conventional sequence.

Objective sound level measurements

The sound level during the quiet and conventional sequence was measured using an MRI safe condenser microphone (ECM8000, Behringer, Willich, Germany) connected to a computer, which recorded the sound directly using MATLAB. A 94 dB noise source (sound level calibrator type 4231, Bruel & Kjaer, Nærum, Denmark) was used to calibrate this microphone. During the sound measurements, the microphone was placed in the gradient insert without a subject being present and at a position that mimicked the position of the ears during the examination. The measurement data was processed in MATLAB, and exponential filtering and A-weighting were applied to correspond to the fast response setting and output of a sound level metre [18].

Quantitative image assessment

For each subject, the images were skull-stripped using optiBET [19] allowing the registration of the quiet sequence images to the images of the conventional sequence using a rigid-body registration (FLIRT FSL toolbox) [20]. The grey-white matter contrast of the sequences was quantified using the signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR) and tissue signal histograms. The grey and white matter were segmented from the skull-stripped images using the FAST automated segmentation tool from the FSL toolbox [21]. Importantly, bias field-corrected images were used to remove signal variations due to inhomogeneous B₁. The segmentation output was analysed in MATLAB (MathWorks, Natick, MA, USA), and the SNR and CNR were calculated using Eqs. 3 and 4 from Oliveira et al. [22]:

Here, the SNR was determined by calculating the ratio of the average signal μ_foreground and standard deviation (SD) σ_foreground in the combined grey and white matter, which was then scaled to the number of voxels (n) to allow for comparison between scans. The CNR was determined by calculating the absolute difference between the average signal in the grey (μ_grey-matter) and white matter (μ_grey-matter). The noise was estimated by combining the SDs in the grey (σ_grey-matter) and white matter (σ_grey-matter).

Qualitative image assessment

Blind assessment of the registered skull-stripped images of both the quiet and the conventional sequences was performed by two neuroradiologists to determine the image quality: one with eleven years of experience in 7-T neuroimaging and a neuroradiology fellow with three years of neuroimaging experience. Overall image quality, visibility of anatomical details, grey-white matter contrast and delineation of vascular structures were scored using a 5-point Likert scale from 1 (very poor) to 5 (excellent). Visibility of anatomical details and grey-white matter contrast were divided into the following subcategories, i.e, areas of the brain: frontal, temporal, parietal and occipital lobe, limbic system, and basal ganglia. Additionally, flow, susceptibility, bounce point and truncation artifacts, if present, were scored from 1 (severe) to 4 (mild). An average score per category was determined for the quiet and conventional sequence.

Subject experience

Subjects were given adequate hearing protection, i.e., earplugs combined with earmuffs. Each subject underwent both the quiet and the conventional sequence twice to determine consistency in reporting; the order of the sequences differed between subjects to rule out any order effects. Immediately after each sequence and after the whole MRI examination (delayed), subjects were asked to rate the sound level of each sequence on a scale from 0 to 10, with 0 being absolutely silent and 10 being the loudest sound they could imagine. In addition, after the whole MRI examination, subjects completed a questionnaire in which they rated their level of comfort, overall experience and willingness to undergo the sequence again on a scale from 0 to 10. For level of comfort, 0 meant being extremely uncomfortable and 10 the most comfortable they could imagine; for overall experience, 0 meant not being satisfied at all and 10 extremely satisfied, and for willingness to undergo the sequence again, 0 meant being absolutely not willing and 10 very much willing to undergo this MRI sequence again in the future.

Statistical analysis

For the quantitative image assessment, a Wilcoxon signed-rank test was used to assess the differences in SNR and CNR with a significance level of p < 0.05. A Cohen’s κ was calculated to determine the interobserver agreement for the qualitative image assessment scores for both the quiet and the conventional sequence.

Since all subjects underwent each sequence twice, differences in ratings between the first and second time were assessed first, after which an average rating per category was calculated. For each of the experience measures, differences in the experience ratings of both the quiet and the conventional sequence were assessed using Wilcoxon signed-rank tests with a significance level of p < 0.05.