Respiratory motion-associated movement of the liver can potentially cause MR data to originate from different locations within liver parenchyma, a vessel or even outside the liver. A study of abdominal motion showed the diaphragm typically moved 15–20 mm during free breathing [30
]. Even with a metabolic liver disease like NAFLD in which steatosis has been shown histologically to be diffusely and equally distributed [31
], respiration will affect the MR spectra, causing phase shifts [32
]. In liver spectroscopy studies, typically a number of spectra are acquired and automatically averaged, with only the final averaged spectrum saved. Without taking into account respiratory motion, the final averaged spectrum may be inaccurate.
Performing individual-spectral correction before averaging the acquired spectra significantly increased the S/N in the liver spectra in this study. The individual spectral phase variations in the spectra were likely due to intrahepatic motion (as opposed to extrahepatic motion) during the spectral acquisition. This approach to dealing with the phase shifts caused by respiratory motion was proposed by Zhu et al. [23
] and further shown by others [24
], though this correction was only demonstrated in a limited number of subjects. Additionally, only one of these studies acquired spectra in the liver, in which the single subject was a healthy volunteer [24
]. However, this technique of phasing individual spectra before combining them has also been applied in the brain to correct for smaller motion [26
This correction cannot eliminate artifacts beyond frequency and phase variations, such as might result from shifts into different tissues. The approach in this study to deal with such extrahepatic motion was to identify spectra with large changes in the water peak area (50% different from the median) and then remove them before averaging the spectra. The same approach can and has been applied to identify such changes in the fat peak area rather than or in addition to changes in the water peak area. Prescot et al. [27
] also applied a similar technique to eliminate individual spectra from a time series of single voxel spectra obtained from the gallbladder. They used the presence of a large (CH2
)n lipid peak at 1.2 ppm, based upon in vitro studies, as their criteria for elimination [27
]. In the current study, 41% of subjects had large changes in their lipids/water ratio during the spectral acquisition, likely due to motion-induced shifting from hepatic to extrahepatic tissue, indicating that such motion is common in liver MRS and that neighboring tissue can have very different water MRS signals. This motion caused, in general, a 2.6% change in the resultant lipids/water ratios. Such a change may not be very significant when simply identifying fat in the liver; however, it can be variable and may be significant when assessing subtle changes in hepatic steatosis that might result from therapeutic intervention. In 11% of subjects, such respiratory motion caused greater than 5% change in the lipids/water ratio, with differences ranging as high in magnitude as 290%. In this extreme case, the lipids/water ratio was 0.27 after only individual phase corrections, suggestive of Grade 2 or 3 steatosis [19
] (steatosis grading relies on the percentage of hepatocytes containing fat: Grade 0: <5%, Grade 1: 5–33%, Grade 2: 33–66%, Grade 3: >66% [8
]). In contrast, the corrected lipids/water ratio (0.07) indicated a low Grade 1 steatosis [19
]. Such artifactual variation could cause significant misinterpretation of the liver MRS result, particularly if assessing the effects of an intervention. Additionally, as the size of this artifactual effect on the lipids/water ratio is variable, study design sample size calculations could be impacted significantly if this correction was not applied.
Other potential approaches to dealing with respiratory motion in liver MRS include breath-holding and respiratory gating. Breath-held liver MRS has been attempted and did reduce the variability of measurements [32
], but cannot provide the higher S/N of longer sequences. Respiratory gating, often used in MR, would be expected to reduce motion artifacts, particularly of small amounts of motion. However, variable intensity of breathing could lead to different tissues being sampled at the same phase of the respiratory cycle. Such large motion artifact would not be corrected with respiratory gating. Also, respiratory gating would lead to individual scans being acquired at differing intervals, leading to data that would be dependent upon the tissue MR relaxation times, the acquisition scan times, the subjects’ breathing patterns and the actual MRS compound concentrations, which would greatly confound interpretation of the MRS data. As additional spectra with respiratory gating were not acquired in this study, a direct comparison of the two techniques is not possible as part of this study.
While large respiratory motion from spectrum to spectrum was corrected in this study, any motion within the acquisition of a single spectrum was not corrected. As the time between excitation and measurement was only 30 ms, such motion was assumed to be small and not of significance. While respiratory motion caused a median magnitude difference in the lipids/water ratio of 2.6%, this was due in part to the long time series of spectra acquired (128), such that artifacts in just a few spectra would not have a large impact on the final average. However, shorter acquisitions would be much more susceptible to motion effects. One such example is the unsuppressed water acquisition performed in this study, with only eight acquisitions. In one case, one of these eight spectra had virtually no signal, leading to a 12.5% difference in the water signal measured. When artifacts did occur in the spectra, the lipids/water ratios in those artifactual spectra were generally quite different from the remainder, so final averaged ratios would also be different if the artifactual spectra were included and lipids/water ratios from the same acquisition were calculated.
A limitation of this technique was that the large changes in the water peak area during the measurement were presumed to be due to extrahepatic motion and corresponding spectra were eliminated. This would not be suitable if dynamic biological changes were expected in the tissue of interest. Additionally, this correction retains spectra with similar water peak areas and reports them as the desired spectra. In actuality, the presented extrahepatic motion-correction scheme yields spectra from the most commonly sampled tissue, which may not be the desired location. As the current study is not investigating a dynamic process during the acquisition and as only 4.4% of the spectra in this study had extrahepatic motion, this limitation is not expected to be a concern in this study.
Our assessment and correction of respiratory motion indicate that the presented, longer, free breathing, water suppressed acquisition can be used to evaluate the liver lipids/water ratios free of respiratory motion artifacts. Also, this acquisition has the potential to provide increased sensitivity due to higher S/N and increased characterization due to increased spectral information, which may be useful for measuring subtle changes in steatosis and for assessing liver disorders beyond steatosis.
This technique was fully automatic and, as such, could be implemented on scanners. However, the algorithm was based upon characteristics of the spectra studied. First, individual spectra were required to be stored and were required to have sufficient water S/N to allow individual phasing. In low S/N applications, a modified water suppression scheme could be utilized to retain sufficient water S/N. Second, the spectral peaks were expected to have the same phasing as the water peak. Third, the water peak region was known and used in the automatic frequency- and phase-correction routine. Additionally, as mentioned above, the motion was expected to be a small fraction of the acquisition, such that the final averaged spectrum represented a tissue type most frequently sampled and not necessarily the desired tissue. Automatic implementation of this algorithm for applications beyond the one presented would require assessment and confirmation of these criteria or could lead to misinterpretation of the resultant spectra.
In conclusion, this study showed that individual spectral frequency- and phase correction of liver MRS increased the S/N’s for all subjects. It is recommended that this be standard practice for all MRS studies susceptible to respiratory or other similar motion artifact. Extrahepatic motion (such as into a vessel) occurred frequently and in some cases caused significant change in the lipids/water ratios with the potential to adversely impact the accurate interpretation of the liver MRS results. Therefore, to obtain a sensitive, high S/N MRS measure of hepatic steatosis, we advocate both intrahepatic and extrahepatic respiratory motion correction of a time series of MRS data. This may be particularly important with increasing MRS use to evaluate the effects of diet, exercise and other therapeutic interventions in ameliorating hepatic steatosis.