**1. Introduction**

Magnetic resonance imaging (MRI) is an effective method for structure investigation of biological samples [1] or human body parts, such as a head [2,3], a thorax [4], etc. In addition to the standard closed-bore MRI scanner, the open-air one is increasingly used in special cases, e.g., in claustrophobic patients [5]. In this MRI device, two parallel permanent magnets form a static magnetic field between them [6]. A gradient system consisting of two internal planar coils parallel to the magnets is used to select slices in three dimensions. A tested object is placed in the magnetic field, together with an external radio frequency (RF) receiving/transmitting coil.

The coil current changes quickly during gradient switching, resulting in undesirable vibration of the whole structure [7], and subsequent acoustic noise disturbing the speech recorded during articulation and concurrent three-dimensional (3D) MRI scanning for examination of dynamic changes in the shape of the vocal tract and vocal folds [8]. In this case, a speech denoising method must be applied on the recorded signal. Another solution includes the recording of a speech signal by a special fiber optical microphone that can be located in the MRI scanning area [9]. Here, real-time speech processing is enabled during MR scanning at the expense of rather complicated practical realization involving synchronization of both processes, special hardware for an MRI device, etc. A cheaper type of an optical microphone (e.g., the first or the second generation of the Optoacoustics FOMRI microphone) has a limited frequency response in the range between 50 and 4000 Hz that is insufficient for our purpose.

**<sup>\*</sup>** Correspondence: Jiri.Pribil@savba.sk; Tel.: +421-2-59104543

The third generation of the Optoacoustics FOMRI microphone system using fiber-optic technology and active noise reduction [10] can solve this problem; however, this solution is more expensive.

Acoustic noise interference is not only a technical problem, but a physiological and psychological one as well. These negative effects on humans are well known in industrial environments with long-term noise exposure [11]; however, they evolve gradually and can be observed first on a short-term scale. The intensity of the vibration and the resulting acoustic noise and the time duration of its exposure are crucial factors affecting the degree of their physiological, as well as psychological impact.

Motivation of our work was an exploration of energetic relations between the vibration and the noise in the scanning area of the low-field open-air MRI device and its surroundings to find a proper choice of a scan sequence and its parameters—sequence type, repetition time (TR), echo time (TE), slice orientation, etc. In addition, the volume inserted in the scanning area influenced the intensity of the vibration and noise produced in the scanning area of the MRI device. Finally, a tested object with its mechanical properties caused the change in the overall mechanical impedance when it loaded the system of lower gradient coils placed in the patient's bed.

The experimental part has the following structure: In the preliminary phase, the sensitivity and the frequency response of the used vibration sensor were determined. Time-domain vibration and noise signals were recorded and processed, along with the measured sound pressure level (SPL), to find energetic features that were then evaluated statistically. Next, analysis of the influence of scan parameters on the time duration of the executed MR sequence and on the quality factor of the obtained MR images was carried out. Finally, the energetic features determined from the measured data were compared with the results of the subjective evaluation based on the listening test method.

### **2. Subject and Methods**

As the open-air MRI is used primarily in medical practice, planes perpendicular to three axes of the Cartesian coordinate system are called according to medical terminology [12]. Anterior and posterior parts of a human body are divided by a frontal (coronal) plane. Another vertical plane divides a body to its left and right sides and is called a sagittal plane. The third plane dividing a body horizontally into superior and inferior parts is called a cross-sectional (transverse) plane. The scan orientation is selected by activation of the corresponding gradient coils, so the current flowing through them, as well as the resulting vibration and acoustic noise during execution of the scan sequence, depend on this selection. Two of the fundamental pulse sequences are preferably used in this MRI device: A spin echo (SE), being an excitation pulse followed by one or more refocusing pulses, and a gradient echo (GE), produced by conjunction of an excitation pulse with a gradient field reversal [13]. For optimal operation of the MRI unit, different parameters of the used scan sequence (the field of view, the number of slices, the thickness of a slice, etc.) are used for different scanned objects. The intensity of the vibration and noise produced by the MRI system depends not only on the setting of these parameters, but on the volume and the weight of the examined object as well. The tested object (a person, a sample, or a phantom) loads the lower gradient coil structure and thus becomes a part of the mechanical vibration system with its mass, stiffness, and damping.

If the vibration is to be picked up while the MRI sequence is executed, the vibration sensor placed in the MRI scanning area must not contain any ferromagnetic part to prevent its interaction with the static magnetic field, which may decrease the quality of the acquired image. It is essential that the sensor has good sensitivity with as small a dependence on frequency as possible. Using the reference sensitivity *B*a0 at a chosen reference frequency, the frequency-dependent sensitivity (frequency response) of an accelerometer *B*<sup>a</sup> [mV/ms<sup>−</sup>2] may be expressed in [dB] by a relation:

$$G\_{a\log}(f) = 20 \cdot \log\_{10}(B\_a(f) / B\_{a0}).\tag{1}$$

The sensor's frequency range should cover harmonic frequencies of the vibration and noise signals concentrated in the low band due to limited frequency spectrum of the gradient pulse sequence. These requirements can be fulfilled by sensors constructed for acoustic pickup in musical instruments, or other ones based on the piezoelectric principle. First usage of all these vibration sensors must be preceded by measurement of their sensitivity and frequency response.

For obtaining high-quality MR images without artifacts, the acoustic sensors (the measuring microphone and the sound level meter) containing ferromagnetic parts must be placed beyond the influence of this static magnetic field, although, adequately close to the noise source. The sensitivity of the recording microphone and its directional pattern are selected in regard to effective rejection of the ambient noise. The sound level meter enables choice of the type of frequency weighting to match human perception of silent sounds (A weighting with more suppressed low and high frequencies) and loud sounds (C weighting with much less suppression of low frequencies than A weighting). Due to high measured sound pressure levels, we chose C weighting.

Several approaches can be applied for determination of the signal energy. Three of them represent the basis of our comparisons:

1. The standard root mean square (RMS) is calculated from a signal *x*(*n*) in a defined frame (window) with the length of *M* samples:

$$Signal\_{RMS} = \sqrt{\frac{1}{M} \sum\_{n=1}^{M} \left| \mathbf{x}(n) \right|^2},\tag{2}$$

2. Another energetic parameter is determined from the Teager–Kaiser energy operator *O*TK [14]:

$$O\_{TK} = \mathbf{x}(n)^2 - \mathbf{x}(n-1) \cdot \mathbf{x}(n+1), \qquad En\_{TK} = abs \left(\frac{1}{M-2} \sum\_{n=1}^{M-2} O\_{TK}(n)\right). \tag{3}$$

3. The third approach uses the short-term *N*FFT-point fast Fourier transform (FFT) to compute the power spectrum |S(k)| 2, and in each frame, the energy is assessed from the first cepstral coefficient *c*<sup>0</sup> or from the autocorrelation coefficient *r*0:

$$En\_{\rm c0} = \sqrt{\left| \prod\_{k=1}^{N\_{FFT}/2} |\mathcal{S}(k)|^2 \right|^{\frac{2}{N\_{FFT}}}},\\ En\_{\rm r0} = \frac{2}{N\_{FFT}} \sum\_{k=1}^{N\_{FFT}/2} \left| \mathcal{S}(k) \right|^2. \tag{4}$$

Next, the basic and supplementary spectral properties are determined from the recorded noise and vibration signals. Methods similar to those used in speech signal analysis can be applied for processing of these signals whose spectral content lies within the standard audio frequency range. The basic spectral properties (basic resonance frequencies *F*V1 and *F*V2 and their ratios, spectral decrease, etc.) are usually determined from the spectral envelope. The supplementary spectral features (spectral centroid (SC), harmonic-to-noise ratio (HNR), spectral entropy, etc.) describe the shape of the power spectrum of the analyzed signal.
