Quantification of hepatic steatosis with T1-independent, T2-corrected MR imaging with spectral modeling of fat: blinded comparison with MR spectroscopy

Abstract

Purpose: To prospectively compare an investigational version of a complex-based chemical shift-based fat fraction magnetic resonance (MR) imaging method with MR spectroscopy for the quantification of hepatic steatosis.

Materials and methods: This study was approved by the institutional review board and was HIPAA compliant. Written informed consent was obtained before all studies. Fifty-five patients (31 women, 24 men; age range, 24-71 years) were prospectively imaged at 1.5 T with quantitative MR imaging and single-voxel MR spectroscopy, each within a single breath hold. The effects of T2 correction, spectral modeling of fat, and magnitude fitting for eddy current correction on fat quantification with MR imaging were investigated by reconstructing fat fraction images from the same source data with different combinations of error correction. Single-voxel T2-corrected MR spectroscopy was used to measure fat fraction and served as the reference standard. All MR spectroscopy data were postprocessed at a separate institution by an MR physicist who was blinded to MR imaging results. Fat fractions measured with MR imaging and MR spectroscopy were compared statistically to determine the correlation (r(2)), and the slope and intercept as measures of agreement between MR imaging and MR spectroscopy fat fraction measurements, to determine whether MR imaging can help quantify fat, and examine the importance of T2 correction, spectral modeling of fat, and eddy current correction. Two-sided t tests (significance level, P = .05) were used to determine whether estimated slopes and intercepts were significantly different from 1.0 and 0.0, respectively. Sensitivity and specificity for the classification of clinically significant steatosis were evaluated.

Results: Overall, there was excellent correlation between MR imaging and MR spectroscopy for all reconstruction combinations. However, agreement was only achieved when T2 correction, spectral modeling of fat, and magnitude fitting for eddy current correction were used (r(2) = 0.99; slope ± standard deviation = 1.00 ± 0.01, P = .77; intercept ± standard deviation = 0.2% ± 0.1, P = .19).

Conclusion: T1-independent chemical shift-based water-fat separation MR imaging methods can accurately quantify fat over the entire liver, by using MR spectroscopy as the reference standard, when T2 correction, spectral modeling of fat, and eddy current correction methods are used.

Figure 1:

Representative example of a fat fraction MR image and its corresponding spectrum obtained with multiecho STEAM MR spectroscopy in a 67-year-old woman with no known liver disease. This patient was slightly overweight (weight, 65.3 kg; body mass index, 27.2 kg/m²). Values for MR imaging fat fraction reconstructed without and with spectral modeling (MP) and without and with T2* correction are shown. Good subjective agreement between MR imaging fat fraction (17.7%) and T2-corrected MR spectroscopy fat fraction (19.1%) was achieved when spectral modeling and T2* correction were performed. TE = echo time.

Figure 2a:

Scatterplots show fat fractions obtained with MR imaging plotted against those obtained with MR spectroscopy. The effects of T2* correction and multipeak modeling of fat were investigated by performing four fat-fraction image reconstructions, as follows: (a) without T2* correction and with single peak modeling of fat; (b) without T2* correction and with multipeak modeling of fat; (c) with T2* correction and with single peak modeling of fat; and (d) with T2* correction and with multipeak modeling of fat. MR imaging fat fraction was reconstructed from the same source data and measured from the same ROIs that were co-localized with the MR spectroscopy voxel. Although all combinations demonstrate excellent correlation, agreement between MR imaging and MR spectroscopy was statistically significant only when both T2* correction and multipeak modeling were used. Magnitude fitting was used to avoid the effects of eddy currents for all four reconstructions. These results demonstrate the necessity of both T2* correction and multipeak modeling of fat.

Figure 2b:

Scatterplots show fat fractions obtained with MR imaging plotted against those obtained with MR spectroscopy. The effects of T2* correction and multipeak modeling of fat were investigated by performing four fat-fraction image reconstructions, as follows: (a) without T2* correction and with single peak modeling of fat; (b) without T2* correction and with multipeak modeling of fat; (c) with T2* correction and with single peak modeling of fat; and (d) with T2* correction and with multipeak modeling of fat. MR imaging fat fraction was reconstructed from the same source data and measured from the same ROIs that were co-localized with the MR spectroscopy voxel. Although all combinations demonstrate excellent correlation, agreement between MR imaging and MR spectroscopy was statistically significant only when both T2* correction and multipeak modeling were used. Magnitude fitting was used to avoid the effects of eddy currents for all four reconstructions. These results demonstrate the necessity of both T2* correction and multipeak modeling of fat.

Figure 2c:

Scatterplots show fat fractions obtained with MR imaging plotted against those obtained with MR spectroscopy. The effects of T2* correction and multipeak modeling of fat were investigated by performing four fat-fraction image reconstructions, as follows: (a) without T2* correction and with single peak modeling of fat; (b) without T2* correction and with multipeak modeling of fat; (c) with T2* correction and with single peak modeling of fat; and (d) with T2* correction and with multipeak modeling of fat. MR imaging fat fraction was reconstructed from the same source data and measured from the same ROIs that were co-localized with the MR spectroscopy voxel. Although all combinations demonstrate excellent correlation, agreement between MR imaging and MR spectroscopy was statistically significant only when both T2* correction and multipeak modeling were used. Magnitude fitting was used to avoid the effects of eddy currents for all four reconstructions. These results demonstrate the necessity of both T2* correction and multipeak modeling of fat.

Figure 2d:

Scatterplots show fat fractions obtained with MR imaging plotted against those obtained with MR spectroscopy. The effects of T2* correction and multipeak modeling of fat were investigated by performing four fat-fraction image reconstructions, as follows: (a) without T2* correction and with single peak modeling of fat; (b) without T2* correction and with multipeak modeling of fat; (c) with T2* correction and with single peak modeling of fat; and (d) with T2* correction and with multipeak modeling of fat. MR imaging fat fraction was reconstructed from the same source data and measured from the same ROIs that were co-localized with the MR spectroscopy voxel. Although all combinations demonstrate excellent correlation, agreement between MR imaging and MR spectroscopy was statistically significant only when both T2* correction and multipeak modeling were used. Magnitude fitting was used to avoid the effects of eddy currents for all four reconstructions. These results demonstrate the necessity of both T2* correction and multipeak modeling of fat.

Figure 3:

Representative examples of fat fraction images obtained with MR imaging in six patients. Fat fractions obtained with MR imaging (MRI) and MR spectroscopy (MRS) are also shown. There was good subjective agreement between MR imaging and MR spectroscopy (performed in the ROI drawn in the posterior segment). A wide range of fat fractions was experienced in this patient population. In addition, many patients had heterogeneous fat (eg, areas of focal sparing [arrows]).

Figure 4:

Scatterplot shows R2* plotted against fat fraction determined with MR imaging from the identical ROI in the posterior segment of the right lobe. Fat fraction and R2* were calculated by using the reconstruction that uses T2* correction, multipeak modeling of fat, and eddy current correction with magnitude fitting. No correlation was demonstrated with linear regression, indicating that R2* is not dependent on fat fraction. The mean R2* value was 38.3 seconds⁻¹ (range, 12.6–140.0 seconds⁻¹), which corresponds to a mean T2* of 26 msec ± 12 (range, 7–79 msec).