Small amounts of diglycerides (DAGs) in lipid depots are highly significant to metabolic health.¹ But diglycerides are difficult to distinguish from triglycerides (TAGs) with MR, since the small differences in chemical shift and J-coupling lead to nearly identical spectra. However, other studies of lipids have shown that even small differences in chemical shift, J-coupling and relaxation dynamics lead to large differences in the signal observed under spin echo trains with varied spacings.^2,3

This work shows that pseudo-random pulse sequences, like irregularly spaced spin echo trains, can rapidly generate signal codes that better distinguish spectrally similar molecules. The resulting codes (determined either experimentally or with product operator simulations) can be used to analyze data from a mixture of compounds with either dictionary based methods or linear decomposition approaches.^4,5 Though initial work is focused on demonstrating the feasibility of distinguishing DAG and TAG, the general approach may be useful for other metabolites or compounds that are spectrally too similar to distinguish through free evolution alone. It may also provide a new source of contrast for waters in structured media like tissue.

Spinach, a product operator based spin simulator, was used to model spin networks based on the glycerol backbones of DAG and TAG.⁶ These spin systems were used to generate free induction decays as well as signals resulting from trains of 16 spin echoes generated with random spacings of the 180 degree refocusing pulses. A number of pulse sequences were studied with random echo spacings chosen from different distributions. Each sequence was simulated on both compounds and one sequence was then used to analyze mixtures of various concentrations where the T2 of the measured signal differed from the modeled T2.

As preliminary experimental verification, signals were collected from diacetin, triacetin, and mixtures of these two. Though commercial technical grade diacetin is a mixture of several molecules, we regarded it as a pure compound, and bottles were prepared with diacetin/triacetin volume % of 100/0, 0/100, 50/50, 10/90 and 5/95. These were studied in a 3T Siemens Trio using both single voxel spectroscopy and multiecho acquisition with 1D spatial phase encoding.

Simulations suggest that families of pulse sequences with echo spacings chosen randomly from the same distribution can have very different ability to distinguish the glyceride backbones. A histogram of the correlation between diglyceride and triglyceride signals across all experiments is shown in Figure 1 and compares favorably against the correlation between simulated spectra (i.e. free evolution), which was found to be 0.79. A sequence with median performance (correlation = .29) was used to study the effect of erroneous T2 for a range of mixtures. As shown in Figure 2, errors of up to 30% in the presumed T2 have little effect on the quantification of diglycerides, though some bias can be detected.

Experimentally measured spectra and multiecho signals for “pure” triacetin and diacetin are shown in Figure 3. These codes for the “pure” liquids were then used to analyze mixtures of diacetin/triacetin using standard linear decomposition. The codes from this first experiment were then used to analyze the data acquired in a second study. Those results are shown in Figure 4, using volume % as the x-axis. (Molar fractions are difficult to calculate without a more thorough chemical analysis of the diacetin.) While neither method gives quantitative agreement with the ground truth, the multiecho approach shows considerably greater accuracy.