Merging images with different central frequencies reduces banding artifacts in balanced steady‐state free precession magnetic resonance cisternography

Abstract Purpose The aim of this study was to evaluate the utility of merged balanced steady‐state free precession (bSSFP) magnetic resonance cisternography images. Materials and Methods Twenty ears of 10 healthy volunteers (six men, four women; mean age ± standard deviation, 26.7 ± 1.6 yr) and 10 patients (two men, eight women; mean age, 46.3 ± 10.9 yr) with neoplasm around the sella turcica were included. Two different devices (A and B) were used to confirm the versatility of our method for MR devices with different local magnetic field homogeneity. Images with different central frequencies (±10, ±20, ±30, ±40, and ±50 Hz) were merged with the maximum magnitude of corresponding pixels from the images acquired using both devices. Two neuroradiologists visually graded the image quality of 11 sites in the inner ear and three sites around the sella turcica (scale: 0–2) and compared the quality with that of the corresponding basic image (0 Hz). Results The image quality was better in merged images of the vestibule, superior semicircular canal (SCC), posterior SCC, and horizontal SCC (P = 0.005 to 0.020 mainly at ±40 and ±50 Hz on devices A and B), as well as in merged images of the sella turcica and right cavernous sinus (±50 Hz, P = 0.003 and 0.020 on device B, respectively), than it was in the corresponding basic images. Conclusions The maximum magnitude merging of images with different central frequencies makes it possible to reduce banding artifacts on bSSFP images without the need for special pulse sequences and image processing programs.


| INTRODUCTION
Magnetic resonance (MR) cisternography using the three-dimensional (3D) balanced steady-state free precession technique 1 (bSSFP) allows the structures of the inner ear, cavernous sinuses, cisternal spaces, and ventricular system to be clearly visualized. [2][3][4] Alternatively, a 3D fast spin-echo technique can also be used to visualize the inner ear. 5,6 Compared to the fast spin-echo method, the bSSFP technique can achieve a higher signal-to-noise ratio and higher resolution within a shorter acquisition time; 7 however, when using bSSFP, inhomogeneity of the local magnetic field owing to the aerated sinus results in banding artifacts. The bSSFP technique can also reflect gadolinium enhancement. Although this characteristic is useful for visualizing the nerves in the cavernous sinus and tumors near the cistern, 8 banding artifacts owing to the paranasal sinuses can impede visualization of the cavernous sinus and sella turcica. Hence, methods that can reduce banding artifacts are required to improve the visualization of these structures.
Several methods for reducing banding artifacts have been proposed, for instance, by merging multiple bSSFP acquisitions, each with a different radio frequency (RF) phase increment from excitation to excitation. [9][10][11][12][13][14][15] However, these methods cannot be used on all MR devices.
In bSSFP MR imaging, the locations of banding artifacts are defined according to the morphology of the imaging target and its surrounding structures. To reduce the banding artifacts in regions with low local magnetic field homogeneity, such as the skull base, we proposed a method of merging bSSFP MR cisternography images with different central frequencies. This idea has been discussed previously in the bSSFP community, 16 with the aim of obtaining stable functional MR imaging of the brain by extending its spatial coverage in areas with high local magnetic field homogeneity. Changing the central frequency is very similar to changing the phase cycling angle for bSSFP, except that it may excite the wrong position. Moreover, unlike other suggested methods for reducing banding artifacts, our proposed method is widely available, even on old MR machines.
In the present study, we examined the utility and versatility of the merged images in healthy individuals and patients with neoplasm around the sella turcica who underwent gadolinium-enhanced imaging.

| MATERIALS AND METHODS
This study was approved by the ethics committee at our institution, and written informed consent was obtained from each subject prior to his/her participation.

2.A | Scanning parameters for the healthy individuals
Twenty ears from 10 healthy adults (six men, four women; mean ± standard deviation age, 26.7 ± 1.7 yr) were evaluated. All MR imaging studies were performed on two 1.5-T MR devices (de-

2.C | Visual grading
Two board-certified neuroradiologists with 11 and 9 yr of experience independently evaluated the anatomical structures, including all four nerves in the internal auditory canal (facial, cochlear, superior vestibular, and inferior vestibular nerves), the cochlear turns, modiolus, spiral lamina of the cochlea, vestibule, and superior, posterior, and horizontal semicircular canals (SCCs) on the basic and merged images. 6 Each neuroradiologist visually graded the image quality of the anatomical structures as good (grade 2, no artifacts and the entire structure was visible), fair (grade 1, image contained artifacts, but the entire structure was visible), or poor (grade 0, visibility of the structures was impeded by artifacts). Any MATSUMOTO ET AL. | 235 discrepancies in grading between the two readers were resolved by consensus.

| 237
We measured the signal-to-noise ratio (SNR) of each image. Four rectangular regions of interest (ROI) were drawn on the phantom images. For the without air condition, the four ROIs were placed around the holes [ Fig. 2(b)]. The ROIs for the condition without air were the same shape as those in the condition with air but different locations [ Fig. 2(c)]. The size of each ROI was 120 × 9 mm. The SNRs of ROI 1, 2, 3, and 4 were calculated as follows: where Mean ROI and SD ROI are the mean and standard deviation of the signal intensity of ROI 1, 2, 3, and 4, respectively. Finally, the SNRs of the four ROIs were averaged for each condition.  acquired using device B were significantly higher than were those of the basic images (P = 0.005). In the superior SCC, the grades of the ±20 to ±50 Hz images acquired using device A and of the ±50 Hz images acquired using device B were significantly higher than were those of the basic images (P = 0.005 and P = 0.020, respectively) ( Fig. 5). In the posterior SCC, the grades of the ±40 Hz images acquired on device B were significantly higher than were those of the basic images (P = 0.020). In the horizontal SCC, the grades of the ±10 to ±50 Hz images acquired using device A and of the ±30 to ±50 Hz images acquired using device B were significantly higher than were those of the basic images (P = 0.020 and P = 0.020, respectively) (Fig. 4).

2.E |
Some cases had obvious wrapping artifacts in the slice-encoding direction on the merged images with large differences in the central frequencies (Fig. 6). For example, in the ±50 Hz image, wrapping artifacts were observed in six and one image(s) from the 10 individuals obtained using devices A and B, respectively.

3.B | Phantom experiment
As shown in Fig. 7, among the images offset from −100 Hz to +100 Hz, the phantom images offset at 20 Hz and −20 Hz showed the maximum SNR value for the with and without air conditions, respectively. Therefore, the −10 to +40 Hz and −50 to 0 Hz phantom images for the with and without air conditions, respectively, surpassed 90% of the maximum value.

3.C | Assessments of the sella turcica and
cavernous sinus from the clinical study Table 2 summarizes the findings from the sella turcica and cavernous sinus. No grading discrepancies were noted between the two readers when evaluating these structures (κ = 1.000). On the basic images, linear, or arcuate artifacts were clearly visualized near the sinuses of all 10 patients (Fig. 8). In 9 of the 10 patients, the sella turcica received a grade of 0 on the basic images because of artifacts from the sphenoid sinus. In the remaining patient, the sphenoid sinus was hypoplastic. In six and four of the 10 patients, the right and left cavernous sinuses, respectively, were classified as grade 0 on the basic images. In contrast, the merged images showed fewer artifacts and the structures were scored as grade 2. The grades of both the sella turcica and right cavernous sinus were significantly higher on the merged images than they were on the basic images (P = 0.003 and P = 0.020, respectively).

| DISCUSSION
In the present study, we merged the maximum intensities of each pixel corresponding to different central frequencies to determine Here, we noted that fewer banding artifacts were observed on high-frequency-shift merged images than on low-frequency-shift merged images. However, the greater the center frequency shift, the showed the necessity of using the minimum frequency shift while recognizing these two factors.
Since our method merges multiple images, it can be applied to a region where there is no influence of motion between the images, for instance during spinal cord MR myelography. 18 Banding artifacts are especially likely to occur when imaging the cervical spinal cord and thoracic cord owing to local magnetic field inhomogeneity, and therefore our method will likely be highly useful in clinical settings. Other applications include imaging of the musculoskeletal regions, particularly articular cartilage 19 and shoulder joint MR arthrography. 20 Several limitations of our study should be noted. First, the scan time of this study was extended to 3 min 50 s because both positive and negative images were needed, which is twice as long as the usual scan time of 1 min 55 s. Longer scan times might result in motion artifacts and mismatched images during merging. However, we did not encounter any problems during the image fusion procedures in this study, and sufficient images were obtained even for small structures of the inner ear. Moreover, the longer scan time in our study was similar to those used in other studies, as many SSFP techniques require long scan times to acquire multiple images using different RF phases. Second, no comparison was made to RF-based methods (constructive interference in the steady state-type sequences). However, since our study aimed to confirm the versatility of our method, we used two devices with different local magnetic field homogeneities. Device B was an MR device in which RF-based methods could not be applied. Third, the small sample size of the patient group might limit the generalizability of the results to clinical settings. However, our results showed statistically significant differences, implying that the statistical test had enough power. Finally, the developmental degrees of the aerated sinuses were not evaluated individually, and differences in these areas might have affected our results. However, we selected healthy volunteers and patients with disorders that do not interrupt the aerated sinuses, such as chronic otitis media. These participants should exhibit the artifacts that are typical of aerated sinuses. Therefore, we believe that the results of our study are robust.

| CONCLUSION S
The maximum magnitude merging of images with different central frequencies makes it possible to reduce banding artifacts on bSSFP MR cisternography images without the need for special pulse sequences and image processing programs.