SEARCH

SEARCH BY CITATION

Keywords:

  • dGEMRIC;
  • knee cartilage;
  • osteoarthritis;
  • T1 mapping;
  • variable flip angle

Abstract

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. SPECIFIC EXPERIMENTAL DETAILS
  5. RESULTS
  6. DISCUSSION AND CONCLUSION
  7. Acknowledgements
  8. REFERENCES

Delayed gadolinium-enhanced MRI of cartilage is a technique, which involves T1 mapping to identify changes in the structural integrity of cartilage associated with osteoarthritis. Currently, the gold standard is 2D inversion recovery turbo spin echo, which suffers from long acquisition times and limited coverage. Three-dimensional variable flip angle (VFA) is an alternate technique, which has been shown to be accurate when an estimate of T1 is available a priori. This study validates the variable flip angle method for delayed gadolinium-enhanced MRI of cartilage of the femoro-tibial knee cartilage. When amplitude of (excitation) radiofrequency field inhomogeneities were minimized using nonselective pulses and amplitude of (excitation) radiofrequency field correction using an additional acquisition of a amplitude of (excitation) radiofrequency field map, the accuracy of T1 measurements were improved, and slice-to-slice variations over the 3D volume were minimized. In conclusion, fast 3D T1 mapping using the variable flip angle method with amplitude of (excitation) radiofrequency field correction appears to be an efficient and accurate method for delayed gadolinium-enhanced MRI of cartilage of the knee. Magn Reson Med, 2011. © 2010 Wiley-Liss, Inc.

Osteoarthritis (OA) is the most common form of arthritis resulting in the loss of cartilage (1). Radiography is routinely used in the evaluation of OA (2) to detect the loss of cartilage, but to-date there is no evidence that it can detect changes during early stages of the disease. MRI is becoming the modality of choice for the evaluation of cartilage and provides information on both anatomy and composition (3). Delayed gadolinium-enhanced MRI of cartilage (dGEMRIC) is a novel method with the potential to identify early changes in the composition of cartilage associated with OA (4, 5). dGEMRIC uses a negatively charged MRI contrast agent Gd(diethylene triamine pentaacetic acid)2− [Gd(DTPA)2−] as a mobile ion probe. When this contrast agent is allowed to penetrate into the cartilage, its spatial distribution shows an inverse relationship with the concentration of negatively charged glycosaminogylcan. The contrast agent uptake is relatively low in normal cartilage and high in degraded, osteoarthritic or glycosaminogylcan-depleted cartilage. A measure of postgadolinium relaxation time T1 (Gd) or ideally the change in R1 relaxation rate (ΔR1) is thought to be a reflection of glycosaminogylcan content.

Most of the reported work on dGEMRIC has used the standard two-dimensional (2D) inversion recovery turbo spin echo (IR-TSE) technique to acquire quantitative T1 mapping. Recently, three-dimensional (3D) sequences based on IR spoiled gradient recalled echo (6) and Look–Locker techniques (7, 8) have been shown to provide quantitative T1 mapping with sufficient accuracy compared with 2D IR-TSE. The typical acquisition times for these different approaches are: 10 min for one slice with 2D IR-TSE, 20 min for a 3D IR spoiled gradient recalled echo sequence, and 10 min for a 3D Look–Locker sequence for full joint coverage with comparable in-plane resolution. Three-dimensional variable flip angle (VFA) is an attractive approach with shorter acquisition times, e.g., approximately 6 min for a two flip angle acquisition with similar spatial resolution. It has been shown to be accurate when an estimate of T1 is available a priori (9). The technique has been applied to whole brain T1 mapping (10, 11) and more recently to dGEMRIC of the hip (12) and knee (13, 14) joints. The primary objective of the study was to verify the need for amplitude of (excitation) radiofrequency field (B1) correction to improve accuracy and uniformity over the 3D volume. Specifically, the use of nonselective excitation pulses and correction for inherent B1 inhomogeneities due to coil and electromagnetic properties of the object were evaluated.

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. SPECIFIC EXPERIMENTAL DETAILS
  5. RESULTS
  6. DISCUSSION AND CONCLUSION
  7. Acknowledgements
  8. REFERENCES

Scanner and Subjects

All data were acquired on a 32-channel 1.5 T MR scanner (Magnetom Avanto; Siemens, Erlangen, Germany) using a commercial transmit/receive quadrature extremity knee coil (Siemens). Two-dimensional IR-TSE and 3D VFA sequences were used in both phantoms and in a small number of human subjects. In vivo MR imaging was performed postcontrast [90–120 min following administration of 0.2 mmol/kg Magnevist (Bayer Healthcare, NJ)]. Four subjects with OA (one male and three female; mean age, 59 years) and two healthy subjects who were asymptomatic with no history of documented knee injury or disease (one male and one female; mean age, 25 years) participated in the study. Subjects with OA had been previously diagnosed clinically and their Kellgren–Lawrence scores were available (grade 1, N = 1; grade 2, N = 3). The study was performed with the approval of the institutional review board, and informed consent was obtained from all subjects.

SPECIFIC EXPERIMENTAL DETAILS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. SPECIFIC EXPERIMENTAL DETAILS
  5. RESULTS
  6. DISCUSSION AND CONCLUSION
  7. Acknowledgements
  8. REFERENCES

Comparison of 3D VFA (With and Without B1 Correction) Versus 2D IR-TSE

To allow comparable coverage used in vivo, a phantom was prepared consisting of nine 10 cm long tubes with 2% agar gel doped with varying concentrations of Gd(DTPA)2−. Using this phantom, slice-to-slice variations over the entire slab thickness due to B1 inhomogeneity were investigated. Additionally, data were acquired in six subjects (four OA and two healthy subjects) using 2D IR-TSE and 3D VFA with and without B1 correction. Data were acquired on one knee and in the sagittal orientation. Comparisons were made between 2D IR-TSE and 3D VFA T1 values to determine T1 accuracy with and without B1 correction.

Sequence Parameters

The parameters for the 3D VFA sequence included pulse repetition time/echo time = 15/1.7 msec, matrix size = 256 × 256 × 32, field of view = 160 mm, slice thickness = 3 mm, and bandwidth = 210 Hz/pixel. A two flip angle acquisition and inline T1 mapping with an acquisition time of 4 min 7 sec was used. Based on a preliminary analysis of different T1 values, the flip angles were determined to be 6° and 33° corresponding to an estimated T1 of 500 msec. This was based on estimating those angles resulting in 71% of signal compared with the maximum signal achieved at the Ernst angle (9). This combination resulted in the least mean squared error (similar to a method previously described; Ref.12).

The slice position for the 2D IR-TSE sequence was matched with one corresponding slice from the 3D volume obtained with the 3D VFA method. The scan parameters were pulse repetition time/echo time = 2200/13 msec, inversion time = 1680, 650, 350, 150, and 50 msec, matrix size = 384 × 384, slice thickness = 3 mm, field of view = 16 cm. The total acquisition time for five inversion times was 10 min 25 sec (2 min 5 sec for each inversion time). T1 maps were calculated offline using MRI Mapper (copyright Beth Israel Deaconess Medical Center and Massachusetts Institute of Technology, 2006; Ref.15).

B1 Correction

To minimize slab profile effects when using selective excitation pulses, use of nonselective pulses were investigated. For minimizing the effects of inherent B1 inhomogeneities, a B1 correction scheme was used. The method involves B1 mapping by comparing spin echo and corresponding stimulated echo signal intensities (16, 17). Based on this measurement, a correction term for the local flip angles is estimated. B1 mapping was performed at low spatial resolution and using a field of view that is greater than that used for T1 mapping. The sequence parameters for the 2D multislice acquisition were pulse repetition time/echo time/mixing time =1368/14/14 msec, matrix size = 64 × 64, field of view = 250 mm, slice thickness = 5 mm, and 36 slices with 150% slice gap. The acquisition time for the B1 map was 86 sec. Relative flip angle distribution was calculated for the 3D volume following postprocessing and excluding voxels with low signal to noise ratio. Interpolated values were then used for T1 mapping. In the ideal case, the flip angle should be 90° at all locations within the imaging volume. A correction factor [c(x,y,z)] was calculated as the actual value of the flip angle at position (x,y,z) divided by 90°. The T1 value corrected for B1 influence is calculated on a per pixel basis using Eq. 1

  • equation image(1)

where Q = S1/S2. S1 and S2 are pixel values from each of the images with different flip angles at each spatial location (x,y,z).

Equation 1 is the standard T1 calculation formula for the 3D VFA with the added correction factors c(x,y) applied to the prescribed flip angles based on the measured flip angle map.

Data Analysis

In phantoms, regions of interest (16 × 16 pixels) were defined at the center of each tube. For human subjects, regions of interest were defined at the central regions of the femoral and tibial cartilage in both medial and lateral condyles similar to a previous report (7). Linear correlation and Bland–Altman plots were used for statistical analysis to determine the degree of agreement between the two techniques.

RESULTS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. SPECIFIC EXPERIMENTAL DETAILS
  5. RESULTS
  6. DISCUSSION AND CONCLUSION
  7. Acknowledgements
  8. REFERENCES

The slab profiles of the 3D T1 maps from three representative tubes of the phantom obtained using selective and nonselective pulses are shown in Fig. 1. As evident from these plots, the use of nonselective pulses removed the drop in T1 values along the edges of the slab. However, the T1 values (especially for longer T1s) are underestimated with both selective and nonselective pulses when compared with the corresponding 2D IR-TSE measurement. This suggests the need for additional B1 inhomogeneity correction. Figure 2 portrays the flip angle map obtained with the phantom, illustrating the presence of B1 inhomogeneity within the slice as well as across the slices. The slab profiles following B1 correction show improved agreement with 2D IR-TSE (Fig. 1c,d).

thumbnail image

Figure 1. Slab profile effects. Plots of T1 values as a function of slice numbers obtained in three representative tubes (shown as three different line styles) of the phantom are shown when using (a) selective pulses, (b) nonselective pulses, (c) selective pulses with B1 correction, and (d) nonselective pulses with B1 correction. Two-dimensional IR-TSE-derived T1 values for reference purpose (horizontal lines) are included. While the use of nonselective pulses improves the slab profile effects, the quantitative agreement with 2D IR-TSE is improved only with B1 correction.

Download figure to PowerPoint

thumbnail image

Figure 2. B1 (relative flip angle) map obtained with the phantom. Note the B1 inhomogeneity (variation in intensity) within a slice [sagittal view (a)] and across the slices [coronal view (b)]. Ideally the entire image should be uniform with a color representing 90°. The lines show the position of slices with respect to each other.

Download figure to PowerPoint

Figure 3 illustrates the slice-to-slice variation with and without B1 correction. While use of nonselective pulses improves the correlation of T1 values across the slices, the improvement is further enhanced following B1 correction. Also note the improvement in the quantitative agreement following B1 correction (b and d), especially for the long T1 values.

thumbnail image

Figure 3. Slice-to-slice variation in T1 measurements in phantom are presented. Plots show T1 measurements obtained with VFA from representative off-center slices as a function of T1 values in the central slice using (a) selective pulses, (b) selective pulses and B1 correction, (c) nonselective pulses, and (d) nonselective pulses and B1 correction. Note the offsets from different slices are similar at each of the tubes with relatively long T1 in (b) suggesting that the residual disagreement is primarily from the slice profile effect. Note further improvement in (d) which includes use of nonselective pulses.

Download figure to PowerPoint

Figure 4 illustrates the correlation between T1 measurements with the standard 2D IR-TSE technique and 3D VFA without and with B1 correction in the phantom, both using nonselective pulses. There is a substantial improvement in the level of agreement in the T1 values following B1 correction as demonstrated by slope (1.06 vs. 0.82), offset (−2.4 vs. 37.9), and r (0.99 vs. 0.96). The associated Bland–Altman plots illustrate the considerable reduction in errors following B1 correction as demonstrated by the 95% confidence limits (−69.3 to 23.2 vs. −73.6 to 139.8).

thumbnail image

Figure 4. Correlation and Bland–Altman plots for T1 measurements by 3D VFA and standard 2D IR-TSE techniques in the phantom. (a) and (c) are measurements without B1 correction and (b) and (d) are for measurements with correction. The data points are from different slice locations as shown in Figure 3. Note the relative improvement in the level of agreement and reduction in errors following B1 correction.

Download figure to PowerPoint

Correlation plots for the in vivo data are shown in Fig. 5 illustrating substantial improvement in the level of agreement in the T1 values following B1 correction as demonstrated by slope (0.99 vs. 0.63), offset (−9.4 vs. 122.8), and r (0.81 vs. 0.54). The associated Bland–Altman plots show reduced errors following B1 correction as demonstrated by the 95% confidence limits (−65.2 to 91.2 vs. −77.5 to 154.3).

thumbnail image

Figure 5. Correlation and Bland–Altman plots for T1 measurements by 3D VFA and standard 2D IR-TSE techniques in vivo are shown. (a) and (c) are measurements without B1 correction and (b) and (d) are measurements with correction. Note the improvement in the level of agreement and reduction in errors following B1 correction.

Download figure to PowerPoint

DISCUSSION AND CONCLUSION

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. SPECIFIC EXPERIMENTAL DETAILS
  5. RESULTS
  6. DISCUSSION AND CONCLUSION
  7. Acknowledgements
  8. REFERENCES

Two-dimensional IR-TSE is currently considered as the gold standard for T1 measurements to perform dGEMRIC. However, it has relatively long acquisition times and limited 2D coverage. The VFA method using two flip angles is an attractive alternative that can provide complete joint coverage and reduce total acquisition time. The results presented here support the feasibility of using this approach for evaluating femoro-tibial cartilage, provided B1 correction is included.

As demonstrated, minimizing B1-related variations improved the level of agreement both spatially within the sample and quantitatively when compared with the 2D IR-TSE measurements. The level of improvement on the Bland–Altman plot was lower in vivo compared with the phantoms. We believe this may be related to the heterogeneity within the cartilage resulting in a wide range in T1 values across the regions of interest.

The slab profile effect associated with the use of selective pulses can be avoided by the use of nonselective pulses. Use of nonselective pulses does not notably increase the total acquisition time for knee imaging because the entire joint is normally included in the volume of interest. For imaging other joints such as the hip, one can alternately increase the slab thickness so that the profile effects are outside the volume of interest. Our data demonstrate considerable improvement in the level of agreement between slices when using nonselective pulses (Fig. 3c). However, use of nonselective pulses alone does not provide adequate agreement in T1 values compared with the 2D IR TSE method (Fig. 4a). The disagreement at long T1 values could be due to contributions from two potential sources: (i) the flip angles not being what were prescribed due to B1 inhomogeneity and (ii) the VFA approach having limited accuracy at long T1s owing to the relatively small signals with each flip angle. However, the level of agreement observed following B1 correction (Fig. 4b) suggests predominant contribution from the former. It is conceivable that with the availability of B1 map and the known characteristics of a slab-selective pulse, B1 could be predicted for the slab-selective pulse in the future. This may be useful in some applications where the use of nonselective pulses may not be feasible.

The B1 mapping was performed using a single acquisition technique based on comparing the spin echo and the corresponding stimulated echo signal intensities (16–18). However, any other method for acquiring B1 maps such as the double angle method (19) could be used. The inherent advantages of a single acquisition method include shorter acquisition time and avoiding any potential misregistration errors. The disadvantage with this approach is the potential for T1-related differences between the spin echo and the stimulated echo. This is generally minimized by shortening the interval between these echoes (18). Further improvements can be achieved by using multiple measurements with varying flip angles.

Our data suggesting the need for B1 correction for T1 mapping by VFA approach is consistent with previous reports using either selective (20) or nonselective pulses (21) in the brain. There are recent reports on the use of 3D VFA-based T1 mapping (same as the one used in this study) for dGEMRIC (12, 14) but without the use of B1 correction. A more recent report has shown improvement in T1 quantitation following B1 correction for dGEMRIC of the patellar cartilage (13) using three or four flip angles and offline analysis. The novelty of this work, especially when compared with Refs.12 and14, is that it has specifically evaluated the variation of T1 over the 3D volume in the phantom and shows that the uniformity improves with the use of both nonselective excitation pulses and B1 correction. Also, unlike Ref.14, our results show less ideal response of VFA-estimated T1s compared with 2D IR-TSE in phantoms (Fig. 4a) but improved only following B1 correction (Fig. 4b). Ref.12 was a study in the hip joint and the slices evaluated were centrally located. Further the use of body coil for excitation may have resulted in less B1 variation over the volume of interest.

The availability of inline calculation of T1 maps allows for use of 3D visualization tools readily available on all scanner platforms.

In conclusion, the fast 3D T1 mapping by 3D VFA is an efficient method and appears to be accurate for imaging knee cartilage provided correction for B1 inhomogeneities is included. The additional time required to acquire the B1 map is albeit a small price compared to using IR-based acquisitions requiring multiple images with different inversion times.

Acknowledgements

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. SPECIFIC EXPERIMENTAL DETAILS
  5. RESULTS
  6. DISCUSSION AND CONCLUSION
  7. Acknowledgements
  8. REFERENCES

The authors thank Dr. Pippa Storey for useful suggestions regarding use of nonselective pulses, Ms. Hongyan Du for statistical consult, Ms. Joann Carbray for her technical assistance in constructing the phantom, and Ms. Sally Gartman for editorial help with the manuscript.

REFERENCES

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. SPECIFIC EXPERIMENTAL DETAILS
  5. RESULTS
  6. DISCUSSION AND CONCLUSION
  7. Acknowledgements
  8. REFERENCES