Magnetic resonance fingerprinting (MRF) is a new technique promising the efficient quantification of tissue parameters¹. A fundamentally unresolved issue of MRF, however, is the appropriate choice of sequence parameters to achieve a high sensitivity to the tissue parameters of interest. In particular, a fast and robust quality measure to determine this encoding capability is desirable to identify the ideal set of sequence parameters for a specific application. In this contribution, we compare and evaluate different candidates for such an MRF sequence performance measure.Five different methods were considered to measure the sequence encoding capability. For the evaluation of these approaches, dictionaries of SSFP-like sequences² (T₁ range: 200-1500 ms, ΔT₁=10 ms; T₂ range: 50-400 ms, ΔT₂=5 ms) with an inversion pulse (TI=20 ms) followed by randomly generated flip angle (FA) patterns (range: 0‑70°, 1000 sequence steps, Fig. 1) were calculated. To reduce the dimensionality of the sequence parameter space, a fixed repetition time (TR) of 15 ms was employed. The five methods were: (i) The average ‘local’ dot product of two consecutive dictionary entries with ΔT₂=0 ms and ΔT₁=10 ms calculated at fifty equidistant locations in the dictionary. For a high encoding capability to T₁, this average dot product should be as low as possible. (ii) The same as method (i) using two consecutive entries with ΔT₁=0 ms and ΔT₂=5 ms. (iii) A matrix containing all ‘global’ dot products of fifty equidistant entries in the dictionary was generated, and the maximum of this matrix was calculated. (iv) ‘Full‘ Monte Carlo (MC) method: Gaussian noise (SNR~10) was added to a dictionary entry, and the resulting signal was matched to the dictionary; this was repeated 2000 times with randomly resampled noise for the same entry. This method was applied for fifty equidistant entries, and the dictionary-wide average error was calculated. (v) Finally, acceleration of the full MC method was realized by using local dictionaries consisting only of the closest neighbors (ΔT₁=10 ms and/or ΔT₂=5 ms) of the examined entries. These local dictionaries were expanded in T₁ and/or T₂ direction, until the cumulative matching probability of the outermost entries was below 2%. To evaluate the performance of these measures, an MRF measurement of a gel phantom containing 12 substances with different T₁/T₂ combinations in the range of the dictionary was performed on a Philips Achieva 1.5 T scanner with an 8-channel head receiver array using the FA pattern in Fig. 1. Afterwards, reduced sequence lengths were simulated by shortening both the acquired data and the dictionary in time.Figures 2a and b depict the mean T₁ and T₂ values in the 12 tubes of the phantom as a function of sequence length, respectively. The results of methods (i-iii) for the same sequence, as depicted in Fig. 2c, reveal a fundamental problem of the dot product based techniques: due to the normalization of the dictionary entries, absolute signal values of the entries generally decrease with increasing sequence length. Consequently, methods (i) and (iii) predict a decreasing accuracy of the MRF sequence towards higher sequence lengths, which is inconsistent with the experimental results. In contrast, the relative matching errors that were obtained using the full MC method (iv) (Fig. 2d) were consistent with the phantom measurements: the convergence to the final T₁ and T₂ values (Figs. 2a,b) coincided with sharp decreases of the relative matching errors as indicated by the vertical lines. Figure 3 shows a comparison of the considered methods for the different sequences. While the local dot product based methods (i/ii) yield measures somewhat similar to the full MC method both for T₁ and T₂, the curve shape for the global dot product method (iii) was markedly different. The accelerated MC method successfully reproduced the matching errors of the full MC method (max. deviation < 0.1 percentage points). On a 6-core Intel Xeon 2.67 GHz system, the evaluation of a single sequence using this method required about 3 to 4 min. Compared with the full MC method, this represented a reduction of the computation time by a factor > 12 (Fig. 3d).Fast and accurate evaluation of the quality of a given sequence in MRF is generally challenging. While interpretation of the dot product quality measures can be difficult in some cases, the presented Monte Carlo method is able to accurately predict the minimum required duration of the sequence. By restricting this method to small local dictionaries, substantial acceleration could be achieved. This technique may hence be applied to increase the sensitivity of MRF to tissue parameters by optimizing the employed sequence.