Accuracy and precision in the measurement of relaxation times from nuclear magnetic resonance images.

Abstract

The accuracy (proximity to the true value) and precision (reproducibility) of relaxation times derived from nuclear magnetic resonance images were investigated. Two methods of deriving relaxation times were considered. A patient scanning protocol in which the minimum number of scans necessary for the calculation (three) were performed. Calculated T1 and T2 images were then formed. An animal (cat) protocol in which many more scans were performed. The data were read from the display and fitted by computer to the theoretical curves. The accuracy of the measurements was determined by an empirical method. A series of bottles with different concentrations of MnCl2 and CuSO4 in water were prepared and their relaxation times determined using the imager as a simple pulsed spectrometer. These values were compared with those derived from images. Over the normal range of tissue values (T1 less than 700 ms, T2 less than 200 ms) the animal protocol gave values of T1 up to 1% shorter than the true values. The T2 values were up to 5% shorter. Patient protocol values were up to 7% shorter for T1 and up to 20% shorter for T2. There was some difference between results for MnCl2 and for CuSO4 (particularly for patient T2s), suggesting that the results depend to a small extent on the T1/T2 ratio. The precision of the values was investigated by considering the standard deviations (SDs) of brain tissue measurements over populations of cats (animal protocols) and normal control subjects and multiple sclerosis patients (patient protocols). These were compared with the SDs of measurements of calibration bottles scanned with the patients. Standard deviations of 3% for T1 and 6% for T2 were found over 19 cats using the animal protocols; SDs of 7% for T1 and 14% for T2 were found over 15 normal control subjects using the patient protocols. Standard deviations of bottle measurements were similar to these figures. There are also variations between different subjects and different regions of the brain. There was no significant change between readings on the same patient in follow-up studies. Other sources of variation in the measurements made with the patient protocols were investigated by scanning phantoms. Noise in T1 and T2 images is about 2%. Spatial non-uniformity within slices is about 1% for T1 and 10% for T2. Non-uniformity between slices in multislice sets is 4% for T1 and 14% for T2. There is no long-term variation in measured values over 9 months; short-term variation is approximately 1%.