To improve the understanding of the constituents of voxels beyond obtaining a single or a limited number of parameters per voxel.Quantitative relaxation parameters provide fundamental information about tissues. However, especially T1 mapping using inversion recovery (IR) is time consuming due to the requirements of waiting for T1 recovery and managing RF energy deposition (SAR) . Consequently, IR-based T1 mapping conventionally uses a limited set of samples (e.g., 1 or 2) during the recovery. While these samples can characterize mono-exponential decay, voxels with a spectrum of T1 values could be better described with increased sampling of the signal change during the inversion recovery. In this study we used a Multiple Inversion time Echo Planar Imaging sequence^1,2 (MI-EPI) to quickly obtain many samples from the inversion recovery, and use these data to estimate T1 distributions to better describe the voxels at the microscopic level. The rapid acquisition of EPI enables whole-brain coverage and dense sampling of the IR. Information of the T1 distribution within a voxel can potentially provide the elucidation sub-voxel composition and also to quantify dynamic partial volume effects due to tissue motion that cause signal variation in functional MRI^3,4.

Four healthy volunteers participated in this study. Data were acquired using a 7 Tesla whole-body scanner scanner (Siemens Healthcare) with a 32-channel array head coil for receive and a birdcage transmit coil. We acquired MI-EPI data where every inversion was followed by the readout of all slices, such that each slice was acquired at a different inversion time (TI). After each inversion recovery period, the slice acquisition order was permuted by one slice. After repeating the acquisition a number of times equaling the number of slices (Nslice), a set of data with Nslice equally spaced TI values were available for every slice. A dictionary of signal evolution was simulated using the Bloch equation with a range of values given to two parameters, T1 and a remainder T2*-weighted term (called S0), and the least-squares error (LSE) was computed between the data and the model.

Conventionally, the single T1/S0 assigned to a given voxel using this dictionary lookup procedure represents the global minimum of the LSE across the dictionary entries. Depending on which subset of the samples along the inversion recovery are used to model the measured signal, different T1/S0 values can be obtained. E.g., using only the low TI values could better represent the short T1 components whereas the higher TI values could result in longer T1 estimates as the dynamics of the short T1 constituents would already have elapsed. Here we randomly selected the samples along the IR curve (with replacement) into N sets and obtained N estimates of T1 and S0; we then assembled a smoothed histogram of these N parameter estimates for each voxel. Broad distributions could be regarded as representing the partial volume contribution of tissues with different T1/S0 values. E.g., a voxel containing both white matter (WM) and gray matter (GM) could have a “T1 spectrum” spanning the T1 values of those two tissue classes.

IR-EPI Protocol parameters: TR = 2.87 s, TE = 23 ms, excitation flip angle = 80°, field-of-view = 192 mm, matrix = 184 × 184, slice thickness = 1.04 mm, Nslice = number of TIs = number of TRs = 50, GRAPPA R = 4, effective echo spacing = 0.205 ms, spacing of different TI values = 56 ms. The sequence employed FLEET ACS reference acquisitions⁵ and trFOCI adiabatic inversions⁶. Fitting: 2 or 5 TI values were randomly sampled 500 times for every parameter histogram. A column of voxels spanning a cortical sulcus was selected (arbitrarily) for showing the results.

Figure 1 shows an axial T1 map generated from the whole data set and a close-up where the voxels of interest have been marked. The figure also includes the smoothed histograms resulting from the random sampling using 2 or 5 inversion times to obtain the T1 values.A wide and dense sampling at different TIs is required to characterize the internal T1 composition of a voxel. The speed of EPI and SNR provided by 7 T enabled investigation into individual voxels at a high spatial resolution. Using a large amount of data, here 5 samples, yielded robust T1 estimates with clear peak values at every WM and GM voxel shown (T1 of cerebrospinal fluid (CSF) was less well defined). However, the more wide-spread T1 distributions from the two-sample fitting provided potentially useful information about the alternative possibilities for the T1 values within the voxels.