[Definition criteria for a magnetic resonance quality assurance program: multicenter study]

We report the preliminary results of a multicenter trial aimed at defining methods, reference values and frequency of measurements for an MR quality assurance program. In particular, we stress the definition of two attention levels (investigation and intervention) for image uniformity and signal-to-noise ratio (SNR) by means of short- and long-term measurements.

The short-term protocol consisted of 10 successive spin-echo (SE, 2 echoes) acquisitions and was repeated after 24 hours. Measurements were made with the same test phantom which was circulated among all the participating centers. Image uniformity and SNR were evaluated using the software available on each MR unit. The long-term protocol, consisting of a single SE acquisition with the same parameters used for the short-term one, was performed once a month for 12 months. A mid term protocol was also performed twice a week for 5 weeks, and the analysis of the results is still in progress.

The short-term protocol results allowed absolute comparison of system performance. Uniformity and SNR were significantly different among centers (p < 0.05), also in the statistical comparison of two MR units of the same model (Siemens Magnetom SP63-1.5 T). Overall, three 1.5 T systems provided similar SNR values, while the results obtained for the 1 T system were markedly lower (51% of the maximum). This result can be explained by the dependence of the analytical expression of SNR on the magnetic field. The other 1.5 T system performed more poorly than the others operating at the same magnetic field. This difference can be explained by the specific characteristics of the coil and by technological aging. Because of the small sample size (5 units), the maximum variation coefficients (3% for the first echo and 3.5% for the second one) were assumed as a reference value for the both parameters (SNR and uniformity). These values were used for the long-term analysis: at every measurement the evaluated parameter was statistically compared with the result of the previous month. We propose to set an investigation level at p = 0.05: when the newly measured parameter differs from the previous value (p < 0.05), we should investigate if this is due to a normal long-term variation or to a system fault. The intervention level is then defined as the 95% prediction interval of the evaluated parameter regression vs time. Measurements that do not fall within the prediction interval are not used for future statistics.

Some preliminary results concerning SNR and uniformity were obtained in the investigation performed on 5 MR systems to define methods and references for a Quality Assurance program. We introduced an "investigation level" and an "intervention level" related to short-term and long-term variability. The investigation levels could be a useful reference value to predict the short-term variability of an MR system with similar characteristics, thus avoiding a long and onerous series of measurements. Concerning the optimization of measurement frequencies, the preliminary analysis of these results showed that a daily measurement frequency is excessive in stable equipment conditions, while monthly measurements showed that the investigation level was often exceeded. However, frequency optimization will be investigated after the analysis of the mid-term measurements, which is still in progress.