Inter-vendor Harmonised T1-Mapping Without Proprietary Information: Preliminary In-Vivo Results at 3T

Background: T1-mapping is useful for quantitative tissue characterisation. A current challenge is reproducibility of normal ranges between centres, leading to requirements for scanning large cohorts of volunteers [1]. This abstract presents preliminary results from in-vivo testing of a harmonised T1-mapping method, developed without access to proprietary vendor information.

Methods: Five healthy volunteers (age 26, range 24-32, 1 female) were scanned on 3T MR scanners from two vendors. On each scanner, 9 left ventricular (LV) short-axis slices were positioned from the mitral valve annulus to apex, and T1-maps acquired of the central 5 slices. The chosen sampling pattern was ShMOLLI 5(1)1(1)1 with conditional reconstruction, to limit heart rate sensitivity. For recovery pauses, beats were used on A and seconds on B. The difference would only affect low T1 (liver) values for high heart rates. Sequence parameters were TE=1.4ms, FA=35°, FOV=340×255 mm, matrix=192×144, R=2, TD=532ms, min. TI=100ms and slice thickness 8mm. TRs were 2.1ms (vendor A) and 2.7ms (vendor B) due to underlying differences between proprietary approaches to SSFP. Phantom scans [2,3] identified a possible difference in inversion time (TI) definition and a small residual T1 difference, both of which were corrected. Additional in-vivo measurements with extended shim volume were performed to test agreement for large deviations from myocardial norm using spleen (high myocardial T1) and liver (lower myocardial T1) as models.

Results: The overall mean whole-slice myocardial T1 was 1110ms (range 1070–1190ms) for vendor A and 1080ms (range 1030–1160ms) for vendor B; the paired mean difference was -30±10ms (p=0.054, Fig. 1).
The mean septal T1 values agreed more closely between vendors: 1120ms on A (range 1080–1190ms) vs 1130ms (range 1060–1180ms) on B. The average difference was +13±13ms (p=0.54, Fig. 1, right). A larger difference was observed in lateral segments: -72±14ms (p=0.01), due to larger intersegmental variation on vendor B (118ms vs 54ms; Fig. 2).
The fitted correction based on the T1MES phantom [2] had a slope of 1.071±0.005 and intercept of -8±3ms (Fig. 3). After applying the phantom-based correction, the mean global T1 difference increased to +40±10ms (p=0.025); the application of the correction appears to have over-compensated. The difference in septal T1 increased to +86±13ms (p=0.005), though the difference in lateral wall T1 reduced to -5±13ms (p=0.72, Fig. 2).
In-vivo experiments showed similar sensitivity of both methods to wide range T1 variation, albeit with large variability. In particular, we observed higher blood T1 values for vendor B than for vendor A.

Conclusion: The observed LV T1 differences between vendors are of comparable size to differences across the LV myocardium in one vendor. The high intersegment variability may be due to different approaches to shimming, bore widths, unresolvable small differences in field strength and vendor optimisations. Further work is pending in a larger study.

Fig. 1 - Paired in-vivo myocardial T1-mapping differences between the vendors A and B at subject level.
Fig. 2 - Myocardial T1 differences between vendors (mean as lines and 95% confidence intervals as shading) according to myocardial segments averaged over all 5 mapping slices.
FIg. 3 - Overall agreement of T1-mapping values between two vendors using the T1MES phantom [3] and a wide range of tissue in-vivo.