As the nuclear magnetic resonance signal itself can be noisy and its meaning challenging to interpret, mathematical models are generally fitted to these measurements to obtain the more accurate characterization of the brain microstructure. This, of course, requires that the mathematical model itself is sound in respect to the measurement setup.
The doctoral thesis “From Diffusion to Tracts” by Viljami Sairanen from University of Helsinki focuses on the extensively used tensor models as they have been shown to unravel details of the physical diffusion phenomena along with various applications in the basic neuroscience, the clinical research, and even in the neurosurgery.
"One of the greatest challenges in the diffusion weighted MRI measurements is subject motion during the image acquisition as that can cause a complete loss of the measurement which is especially highlighted in ill or uncooperative patients’ studies," says Viljami Sairanen.
A tool developed and distributed internationally
Due to the used acquisition technique, this loss extends to multiple measurements simultaneously resulting in an enormous gap in the sampling. Such gaps can be problematic for any model fitting, even for the currently available robust means developed to exclude outlier measurements from affecting the estimate.
Hence in the dissertation studies, a tool coined as SOLID was developed to detect these outliers and to robustly process them during the tensor-based model estimation. SOLID was implemented as a part of the widely used ExploreDTI toolbox to allow the rapid international distribution of the tool.
Unfortunately, any reduction in the measurement sampling will lead to increasing error propagation during the model estimation. Mathematically this is detailed in terms of a condition number for the matrix inversion in the linear least squares fitting.
Novel quality control
Previously, the condition number has been used to optimize the diffusion weighted MRI acquisition gradient scheme but in the work by Viljami Sairanen it was renovated into a novel quality control tool.
The condition number of the matrix inversion that provides the model estimate can be calculated after the outliers are excluded to assess spatially and directionally varying error propagation to obviate any bias in subject or population studies. To motivate the importance of the robust methods and diffusion weighted MRI at large, neurocognitive studies with neonates’ visual abilities and bilinguals’ acquisition age of the second language were conducted as a part of this thesis.
The findings in these studies indicated that premature birth affects the white matter structures across the brain whereas the age of acquisition of the second language affects only the speech related brain structures.
Viljami Sairanen,”From Diffusion to Tracts” on E-thesis website.