T₂ is one of the most commonly used MRI contrasts for non-invasive diagnosis and characterization of pathologies. Quantitative evaluation of T₂ has been shown to be valuable for various applications including stroke¹, multiple sclerosis², cardiac imaging³, cancer detection⁴, and musculoskeletal imaging⁵. Assessment of T₂ contrast, however, is usually performed in a visually qualitative manner making it observer dependent and preventing utilization of the full dynamic range that can be generated by T₂ contrast. This limitation is due to the challenges of quantifying T₂ in vivo including the very long scan times associated with full spin-echo (SE) acquisitions (tens of minutes), or, in the case of multi-SE (MSE) protocols, the inherent deviation of the T₂ decay curve from the theoretical exponential model S(t)=S₀exp[–t/T₂] due to stimulated and indirect echoes, non-rectangular slice profiles, and inhomogeneous transmit (B₁⁺) field profiles. Strategies for overcoming these effects include omitting the first echo-time (TE) point, excluding odd TE data points, or incorporating the above imperfections into a more realistic signal model^6-11. Recently, a new technique – the Echo Modulation Curve (EMC) algorithm – has been introduced^12,13, which relies on precise Bloch simulations of the MSE pulse-sequence, to produce the true T₂ values of the tissue with excellent correlation to values acquired using gold-standard single spin-echo. In this work we present the results of an in vivo stability benchmark of the EMC technique, performed on healthy volunteers and evaluating accuracy, precision and reproducibility over a wide range of parameter settings, and in comparison to conventional monoexponential fitting.

Data acquisition: Data were collected from 30 healthy volunteers on whole-body 3 T scanners (Siemens Skyra and Trio) using a standard MSE protocol. Common scan parameters were {TR=2500 ms, N-echoes=10, res=1.7x1.7mm², T_acq=2:44min (2x GRAPPA acceleration)}.

Reconstruction: T₂ maps were generated using (1) the EMC algorithm¹², and (2) by fitting the set of DICOM time series to a conventional monoexponential model.

Inter-scan and parameter-set stability tests: A series of MSE scans were performed on the brain of a single healthy volunteer using 6 different parameter sets (see Table 1). Each scan was repeated twice on each of the two scanners to a total of 24 scans. The volunteer was, moreover, taken out and back in the scanner between each pair of scans. Mean, standard-deviation (SD), and coefficient of variation (CV) were calculated for 5 region-of-interests (ROIs): genu, splenium of corpus callosum, caudate nucleus head, frontal white matter, and periventricular white matter (GNU, SPL, CDN, FWM, PWM).

Inter subject variability tests: MSE data was collected for 30 healthy volunteers, ages [25…52] (16 males) using parameter-set #2 in Table 1. Average values of the mean, SD, CV over the entire group, were calculated for 6 ROIs: GNU, SPL, CDN, FWM, putamen (PTM), and thalamus (TLM).

Inter-scan stability: Table 2 summarizes the mean, SD and CV values averaged across all 24 stability scans. The EMC technique offers better accuracy (mean value), higher precision (lower SD value), and lower relative spread (lower CV), over the assayed scanners and parameter-sets. The variability between each set of two identical scans is decreased from 4.6% for exponential fit to 1.6% for EMC fit. Inter-subject variability tests are listed in Table 3: consistently higher CV is seen for the exponentially fitted values for all ROIs suggesting that the lower precision (higher SD) of this fitting approach, is not just due to scaling of the mean value but reflects an inconsistent spread of the values’ distribution.Fitting techniques which do not account for the full set of coherence pathways occurring in multi echo protocols suffer from inherent bias and increased spread (i.e., SD and CV) of the measured T₂ values. This deficiency is moreover exacerbated by the fact that the fitting error will depend on the baseline T₂ values, the protocol implementation and the parameter set being used. By incorporating all the experimental factors into the signal model, the EMC algorithm is able to overcome these limitations and extract the true T₂ values of a tissue, in a fashion that is protocol- and scanner-independent, while performing in clinically relevant scan times. The EMC framework can be further accelerated using radial sampling strategies¹³ and extended to model other contrasts (e.g. T₁, diffusion, T₂^*), to derive multi-component T₂ distributions, and to support arbitrary acquisition schemes.