Characterization of normal brain tissue using seven calculated MRI parameters and a statistical analysis system

A statistical analysis system for classifying normal brain tissue has been applied to the analysis of MRI scans on 45 volunteers to achieve a discrimination accuracy of 84% for 13 tissue types among three age group sets, with classification accuracies for individual regions ranging from 50 to 100%.

Abstract

A statistical analysis system for classifying normal brain tissue has been applied to the analysis of MRI scans on 45 volunteers. The Bayes Maximum Likelihood method was used to achieve a discrimination accuracy of 84% for 13 tissue types among three age group sets, with classification accuracies for individual regions ranging from 50 to 100%. In order to attain this level of discrimination a set of seven derived relaxation-type parameters was used to categorize the tissue types. Values for these experimentally estimated parameters were derived from the MRI intensities of eight images in the following pulse sequences: (1) a Carr-Purcell-Meiboom-Gill (CPMG) four-echo train, (2) a single-echo inversion recovery, and (3) three single-echo sequences with varying repetition times, TR, and echo delays, TE. The T2 values derived from ratios of single-echo intensities showed better discrimination power than those from the four-echo CPMG train. The general precision of the seven estimated parameters was excellent, with percentage standard deviations ranging from 4 to 18% for the various regions studied. The tissue discrimination achieved by use of just three relaxation parameters, T1, T2, and proton density, calculated from intensities of images from a four-echo sequence, an inversion recovery sequence, and a short TR single-echo sequence, was not as good, being only 55%.