Standardisation and optimisation of magnetic resonance techniques for multicentre studies

AIM: An approach to measuring physical quantities such as lesion load with MRI in multicentre studies is presented. METHOD: Examples are given of imperfections in current techniques: (1) a step change in a serial trial, giving an apparent (but artefactual) decrease in total lesion volume in untreated patients with multiple sclerosis; (2) inaccuracy (systematic error) in lesion volume, found by measuring a phantom with lesions of known volumes; (3) spatial non-uniformity in the radiofrequency coil sensitivity, giving gross image shading. When using the magnetic resonance imaging (MRI) scanner as a scientific instrument to measure physical quantities, accuracy (closeness to the truth, or lack of systematic error), and precision (reproducibility, or lack of random error) are the keys to success. Quality assurance procedures can utilise phantoms, normal control subjects, and stable patients, and have to be included in serial studies and trials. Between scanner agreement can perhaps be improved by attempting to replicate an inaccurate procedure at each site; but it is more preferable to seek accuracy and precision (as a perfectly accurate and precise procedure must give the same results at all sites). CONCLUSION: Before being included in a serial study, a measurement procedure should ideally demonstrate:(1) accuracy in a phantom; (2) precision in repeated measurements on a phantom; (3) precision in repeated measurements on human subjects.