Diffusion Tensor Imaging: on the assessment of data quality - a preliminary bootstrap analysis

In the field of nuclear magnetic resonance imaging, diffusion
tensor imaging (DTI) has proven an important method for the
characterisation of ultrastructural tissue properties. Yet various
technical and biological sources of signal uncertainty may prolong
into variables derived from diffusion weighted images and thus
compromise data validity and reliability. To gain an objective
quality rating of real raw data we aimed at implementing the
previously described bootstrap methodology (Efron, 1979) and
investigating its sensitivity to a selection of extraneous
influencing factors. We applied the bootstrap method on real DTI
data volumes of six volunteers which were varied by different
acquisition conditions, smoothing and artificial noising. In
addition a clinical sample group of 46 Multiple Sclerosis patients
and 24 healthy controls were investigated. The response variables
(RV) extracted from the histogram of the confidence intervals of
fractional anisotropy were mean width, peak position and height.
The addition of noising showed a significant effect when exceeding
about 130% of the original background noise. The application of an
edge-preserving smoothing algorithm resulted in an inverse
alteration of the RV. Subject motion was also clearly depicted
whereas its prevention by use of a vacuum device only resulted in a
marginal improvement. We also observed a marked gender-specific
effect in a sample of 24 healthy control subjects the causes of
which remained unclear. In contrary to this the mere effect of a
different signal intensity distribution due to illness (MS) did not
alter the response variables.