Multiexponential proton spin-spin relaxation in MR imaging of human brain tumors

In vivo measurements of proton relaxation processes in human brain tumors have been performed by magnetic resonance (MR) imaging using a whole-body superconductive MR scanner, operating at 1.5 T. The T1 and T2 relaxation time measurements were based on a combined Carr-Purcell/Carr-Purcell-Meiboom-Gill sequence with two interleaved repetition times and 32 echoes. First, comparative measurements in the imager and with the spectrometer of relaxation times were performed on phantoms containing fluids of different T1 and T2 to evaluate accuracy. A maximum deviation of approximately 10% was found. Multislicing with a gap width of one slice thickness influenced the accuracy of T1 relaxation measurement. A gap width of at least two times the slice thickness was necessary for reliable determination of T1. No influence on T2 values was observed by multislicing. Second, in human head imaging the multiexponential behavior of the T2 decay curves has been analyzed in each pixel, where the mean square deviation has been used as a criterion to discriminate between mono- and biexponential behavior. Mean values of monoexponential T1 and multiexponential T2 relaxation data for white matter, gray matter, CSF, edema, and tumor were sampled in 12 patients with brain tumors. T2 showed monoexponential behavior in white and gray matter, whereas CSF, edema, and tumor showed distinct biexponentiality. The biexponential analysis generally yields "fast" and "slow" components with T2f = 80 +/- 17 ms and T2s = 2,030 +/- 210 ms for CSF (partial volume effect), T2f = 104 +/- 25 ms and T2s = 677 +/- 152 ms for edematous tissues, T2f = 97 +/- 19 ms and T2s = 756 +/- 99 ms for tumor tissues, respectively. Using a stepwise discriminant analysis by forward selection, the two best discriminating parameters of the multiexponential relaxation analysis for each pair of classification groups have been selected. For the discrimination of edematous and tumor tissues a retrospective overall accuracy of 94% has been found.