Knee dGEMRIC at 7 T: comparison against 1.5 T and evaluation of T₁-mapping methods

Osteoarthritis is a common, painful, and disabling condition characterized by degradation and loss of cartilage. Although the disease progresses slowly, early detection is critical for development of treatment strategies which may prevent or slow down degradation before the cartilage is irreversibly lost.

The delayed Gadolinium Enhanced Magnetic Resonance Imaging of Cartilage (dGEMRIC) technique is a well-established method for early assessment of cartilage quality in osteoarthritis [1]. Using this technique, the distribution of Gd-(DTPA)²⁻ contrast agent in cartilage after intravenous injection is assessed with quantitative T₁ mapping. The estimated T₁ is assumed to be indirectly related to the content of glycosaminoglycan (GAG) which is known to decrease early in osteoarthritis. The method is robust and has proved to sensitively detect early degenerative cartilage processes [2, 3] and loss of cartilage quality [4]. However, the technique requires injection of contrast agent, which in addition to a cumbersome examination procedure may also lead to long-term gadolinium deposits [5]. Thus, current development of magnetic resonance imaging (MRI) methods for assessment of cartilage quality is focused on methods that do not require contrast agent injection (e.g. GAG Chemical Exchange Saturation Transfer (gagCEST), ²³Na imaging, T₂ mapping, and T_1ρ mapping [6]). In the development process of these new techniques there is still a real need for an established method for cartilage quality evaluation to use as a reference. For this purpose dGEMRIC may still be the most suitable choice.

gagCEST and ²³Na imaging benefit from the use of an ultra-high field strength, such as 7 T [7, 8]. Most dGEMRIC studies have so far been conducted at clinical field strengths. To enable the use of dGEMRIC as a reference tool during the development of the new techniques, there is a need to first validate dGEMRIC also at 7 T.

Translating the dGEMRIC technique to an ultra-high field strength may have some advantages but there are also several challenges. Increasing the field strength increases the signal-to-noise ratio (SNR), which may be used to improve either the measurement precision or imaging resolution. However, a higher field strength also increases the expected T₁ values [9] and decreases the relaxivity of Gd-(DTPA)²⁻ [10]. These effects may require an altered dGEMRIC protocol and could reduce the sensitivity of the dGEMRIC experiment.

T₁ mapping is a core component of the dGEMRIC technique and several methods have been suggested in the literature. The gold standard approach is the 2D inversion recovery (IR) technique, but also 3D approaches such as the variable flip angle (VFA) [11] and Look-Locker (LL) techniques have been increasingly used over the last years [12‐14]. Several challenges for accurate T₁ measurements are expected when moving to a higher field strength. First, the longer T₁ values likely require longer inversion and repetition times which increases the acquisition time. Second, the B₁ field is likely more inhomogeneous at ultra-high field strengths compared to clinical field strengths. This may affect the quality of the inversion pulse for the IR and LL experiments, but may also make B₁ correction approaches necessary for the VFA and LL techniques [14, 15]. For dGEMRIC at 7 T, IR [16] and VFA [17, 18] have previously been used for T₁ mapping, but as no quantitative comparison between the methods has been performed, further investigation is needed to find the optimal T₁-mapping approach at ultra-high field strength.

The aim of this study was to evaluate the feasibility of T1 mapping for knee dGEMRIC at 7 T by comparison against 1.5 T in human subjects in vivo. In order to identify a preferred choice of T₁-mapping approach at the ultra-high field strength, we additionally aim to compare and evaluate three T₁-mapping techniques – IR, VFA, and LL.

The cartilage in all included subjects were deemed of adequate thickness for ROI evaluation in deep as well as superficial regions, based on the images from the diagnostic acquisition and also directly from the images used for T1 evaluation. No cases of unexpected pathology was found.

Visually, the T₁ values within ROIs of T₁ maps obtained with the IR sequence were regarded precise and homogenous, both from 1.5 T and 7 T (Fig. 1). In contrast, T₁ varied considerably within the ROI for the 3D sequences (VFA and LL) in the healthy subject as well as the patient image examples.

The median T₁ values obtained with IR at 1.5 T and 7 T for the various ROI and subject groups are presented in Table 1. As expected, longer T₁ values were estimated at 7 T with an average factor of 1.8 (1.4–2.3) times longer 7 T T₁ s compared to those at 1.5 T. The same overall patterns were observed at both field strengths with shorter T₁ values in superficial compared to deep cartilage, and shorter T₁ in the medial compared to the lateral condyle for both patients and healthy subjects. At both field strengths, patient T₁ values were slightly shorter and with a larger spread of values compared to healthy subjects. The difference between patient T1 values and healthy subjects T1 values were however not statistically significant. In the superficial medial region the difference was close to significant both at 1.5 T (P = 0.11) and at 7 T (P = 0.09), whereas for all other ROI’s the test resulted in higher P-values (P > 0.25).

To be able to compare the T₁ values at the two field strengths, they were normalized to the median T₁ in healthy subjects at each field strength (Fig. 2). The differences in normalized T₁ values between 1.5 T and 7 T for the various ROIs were all small, with a largest relative difference of 10% in the superficial lateral region in patients. The normalized T₁ values were not statistically different in most ROIs in neither patient (medial deep: P = 0.36, medial superficial: P = 0.16, lateral superficial: P = 0.13) nor healthy subjects (medial deep: P = 0.77, medial superficial: P = 0.70, lateral superficial: P = 0.70). The only exception was the deep lateral ROI were the difference in normalized T₁ between field strengths was statistically significant in patients (P = 0.04) and had a low P value also in healthy subjects (P = 0.06).

The median (range) time between the medial IR sequences at the two field strengths was 40 (28–63) minutes for healthy subjects and 40 (28–49) minutes for patients. To determine if this time difference would have impact on the comparison of the dGEMRIC results between the field strengths, the ratio of T₁ at 7 T and 1.5 T was compared between healthy subjects first scanned at 1.5 T and at 7 T, respectively. The median (range) T₁ ratios when starting at 1.5 T / 7 T was 1.92 (1.51–2.05) / 1.67 (1.47–2.26) for the superficial ROIs and 1.82 (1.71–2.20) / 1.85 (1.72–1.99) for the deep ROIs. The difference was larger for superficial ROIs, but not statistically different for neither superficial (P = 0.16) nor deep ROIs (P = 0.62). For this reason, we do not discriminate between in which order the measurements at the two field strengths were performed in the results presented here.

A linear correlation was found between the T₁ values at 1.5 T and 7 T with a Pearson correlation coefficient of 0.80 (Fig. 3). The coefficient of variation at 1.5 T/7 T T₁ s was 0.50/0.54 for healthy subjects and 0.57/0.88 for patients, thus indicating a slightly larger spread of the 7 T T₁ values, especially for patients. The inter-reader variability of the IR T₁s was 5.71 ± 49.2 ms at 1.5 T and − 4.25 ± 96.1 ms at 7 T. The adiabatic pulse quality observed at 7 T was significantly lower compared to at 1.5 T with median (range) k factors equal to 1.85 (1.75–1.93) and 1.71 (1.11–1.89) at 1.5 and 7 T, respectively (P = 4·10^− 13). Of the 38 acquired IR data sets at each field strength, three data sets at 7 T had too low SNR for a voxel-by-voxel T₁ estimation due to a poor quality adiabatic pulse. For these data sets, an ROI-based T₁ estimation was performed.

Using the same six TIs at 7 T as used at 1.5 T results in lower T₁ values (bias ± limits of agreement − 127 ± 234 ms) and a lower quality fit with significantly higher SEE (median SEE = 138 (20.9–1180) compared to SEE = 77.0 (29.4–524), P = 2·10^− 6) compared to using an adapted sampling scheme with an additional longer TI at 7 T (Fig. 4). Thus, the presentation of 7 T T₁ results is based on the adapted sampling scheme.

A poor agreement was observed between the T₁ values measured with VFA and IR at 7 T (Fig. 5 and Table 2). With no B₁ correction the VFA technique severely underestimated T₁. Although the accuracy was improved using a B₁ correction, the precision worsened with wider limits of agreement. The inter-slice variability for the B₁-corrected case was − 2.55 ± 433 ms and the inter-reader variability was − 51.9 ± 327 ms. Both of these estimates of variability indicate a poor precision of the VFA method. Out of the 10 acquired VFA data sets, one was excluded due to technical difficulties during imaging.

The LL T₁ values agrees well with IR T₁ values at 7 T (Fig. 6 and Table 2). Compared to VFA, LL both with and without B₁ correction is more accurate and precise. The agreement also for LL is improved using B₁ correction, but the correction is not as vital as for VFA. For the B₁-corrected LL data, the inter-slice variability was 1.33 ± 172 ms and the inter-reader variability was 11.5 ± 63.7 ms, thus indicating a smaller variability of LL data compared to VFA. Out of the 14 acquired LL data sets, five were excluded due to insufficient B₀ shim causing a failed adiabatic inversion pulse, and two were excluded due to technical difficulties during imaging.

This study compared dGEMRIC at 1.5 T and 7 T after a single (one) contrast-agent injection in both healthy subjects and patients. A randomized alternation of the 1.5 T and 7 T examinations of healthy subjects indicated that it was possible to obtain both measurements after a single contrast-agent administration without significant bias due to the difference in time delay after injection. Using an adapted sampling scheme with an additional longer TI, dGEMRIC based on T₁ measurements with IR is feasible at 7 T. The estimated T₁ values at the two field strengths were strongly correlated, although there was a slightly wider distribution of the 7 T T₁ s. Similar normalized T₁ values were found using the two field strengths with significant, yet small, difference between the two only for the deep lateral region. Out of the investigated 3D options at 7 T, LL showed a higher agreement to IR results compared to VFA. For both LL and VFA, B₁ correction is necessary at 7 T. Careful B₀ shimming is crucial, especially for the IR and LL methods.

The feasibility of dGEMRIC at 7 T has previously been studied for knee [16] and hip applications [17, 18]. Our estimated postcontrast femoral T₁ values are slightly longer than those presented by Welsh et al. in healthy volunteers [16]. In expectation of longer T₁ values at 7 T, we chose to use an additional longer TI, compared to what was used in the Welsh study. The results from our study indicate that this choice is necessary to avoid underestimation of T₁ at 7 T.

dGEMRIC has previously been compared between 7 T and a clinical field strength in repaired cartilage tissue of the hip, where dGEMRIC at 7 T resulted in an unexpected T₁ decrease compared to dGEMRIC at 3 T [18]. Our study, however, resulted in the expected markedly longer T₁ values at 7 T compared to 1.5 T in native knee cartilage of both healthy volunteers and patients. In the previous study [18], the VFA technique was used, while we in contrast chose an IR T₁-mapping method for the field strength comparison. Based on the evaluation of these techniques at 7 T presented in our current study, we believe that this resulted in a more accurate comparison of T₁ values between field strengths.

VFA, LL, and IR have previously been compared at clinical field strengths [25]. Similarly to the results in this study, both the accuracy and precision of LL was superior to VFA also at 1.5 T. Previous studies using dGEMRIC at 7 T has used IR [16] and VFA [17, 18] for T₁ mapping. In a few healthy hips, these two techniques have also been semi-quantitatively compared at 7 T and their respective measures were considered comparable [17]. Success of the VFA technique is dependent on an optimal choice of flip angle pair and B₁-inhomogeneity correction at high field strengths [11, 15]. Thus, at 7 T VFA is expected to be especially challenging as B₁ inhomogeneity is likely high enough to also affect the optimal choice of flip angles. This issue may explain why VFA showed poor agreement with IR even after B₁ correction in our quantitative comparison.

The inversion pulses used in IR sequences were of better quality at 1.5 T compared to at 7 T in this study. This might be explained by the use of a transmit/receive coil at 7 T compared to a receive-only coil at 1.5 T. In addition, the adiabatic-type pulses rely on a successful B₀ shim which is more challenging at 7 T. Especially, this issue was apparent for the LL technique of which several data sets had to be excluded for this reason. However, also a few IR data sets suffered from poor SNR due to this problem. The B₀-shim procedure was improved during the course of the study, and after the volume-based first-order shim technique first used was replaced by use of the Shimtool [26] (an image-based second-order shim technique) the shim was sufficient in all the remaining LL data sets and the SNR of the remaining IR data sets were consistently high.

In our implementation, dGEMRIC using IR required a longer scan time at 7 T compared to at 1.5 T. The reason is that we, as we expected longer T₁s at the higher field strength, chose to increase the repetition time and add an extra acquisition with longer TI at 7 T. Our comparison of using six and seven TIs for the T₁ estimation demonstrates that this was necessary to achieve an accurate T₁ estimation at 7 T. However, we also noticed that the longer acquisition time made the sequences more sensitive to patient motion as motion correction was more frequently needed in the 7 T scans as compared to the shorter 1.5 T scans. In a practical case when designing a study protocol it is important to take both the benefits and the potential disadvantages of a longer scan time into consideration.

Focus of this work was on the T₁-mapping techniques used for the dGEMRIC method at 7 T, and patients were primarily recruited to increase the expected range of T₁ values. For this reason, a full comparison of dGEMRIC indices and diagnostic performance was beyond the scope of this study. However, although the difference in post-contrast T₁ values between patients and healthy volunteers was small (not statistically significant) in this study, it was similarly small at both 1.5 T and 7 T. At both field strengths it was the same region (superficial medial) that was closest to a significant difference, which is also the region were cartilage changes had been observed in the patients. The normalized T₁ values were also similar at the two field strengths. This hence implies that the methods perform similarly at the two field strengths. As possible explanation for the small differences found, we speculate that the difference in timing between the protocols starting at 1.5 T and 7 T may have increased the spread of the data making comparisons between groups more difficult.

dGEMRIC using Gd-(DTPA)^2- (Magnevist) will probably be performed less frequently in the future given the fact that its use will be restricted based on the recent reports about long-term gadolinium deposits [5]. However, dGEMRIC could potentially also be used with other contrast agents such as gadoterate meglumine (Dotarem, Guerbet, Villepinte, France). To date almost all dGEMRIC studies have been performed using Magnevist, and future dGEMRIC studies with other contrast agents would of course first need careful and in depth validation studies. After such validation, the results from our evaluation of T1 mapping methods and field strength comparison would still be valuable for future studies with dGEMRIC at 7 T.

There were mainly two limitations of the study design of this work: the time difference between the examinations at the two field strengths and that no precontrast T₁ values were measured. Both for ethical and study design purposes, the examinations at the two field strengths were conducted after a single (one) contrast agent injection. Thus they could not be performed at identical post-injection time delays. Efforts were made to minimize the time delay between examinations and the subjects were moved in a wheel chair between the examination rooms to avoid loading the knee and thus redistributing the contrast agent between measurements. The achieved time delay between examinations are within the previously reported plateau of dGEMRIC values between 2 h and 3 h after injection [27]. Although no statistically significant difference in T₁ was found due to timing differences in healthy volunteers, they may have increased the spread of the data as mentioned in the paragraph above. Precontrast T₁ values were not measured neither at 1.5 T nor 7 T. This choice was made as the addition of these measurements would make the visit and scan time unbearably long for the study subjects. Instead, it was prioritized to make it feasible to perform the examinations at the two field strengths in a single visit and after a single contrast agent injection. Previous work indicates that the precontrast T₁ value contributes little additional information compared to postcontrast values at both 1.5 T and 3 T in native cartilage [3, 28]. In repaired cartilage tissue, measurements of precontrast T₁ may be more important [29]. The importance of a precontrast T₁ value may need to be investigated further also at 7 T.

In conclusion, T₁ mapping for use in the dGEMRIC method is feasible at 7 T with similar normalized T₁ values compared to at 1.5 T and with a strong correlation between T₁ values at 1.5 T and 7 T. However, the IR protocol at 7 T needs to be adapted to the longer T₁ values at this field strength. As a 3D alternative to IR at 7 T, LL is preferred and VFA is not recommended without further optimization of the method. For both 3D methods, B₁ correction is necessary for an accurate T₁ estimation. For LL and IR, careful B₀ shimming is crucial at 7 T.