Researcher develops new statistical statistical methodologies to process MRI-dissemination data

In her doctoral dissertation at the University of Jyväskylä, M.Sc. Jia Liu developed and implemented a series of new computational statistical methodologies to process MRI-dissemination data. The statistical problem of the thesis stems from clinical needs for the diagnosis of brain diseases such as Lewy body dementia.

Water, the molecule of life, is ubiquitous in all organisms. Diffusion plays an important role in all molecular interactions. Using modern magnetic resonance imaging (MRI) techniques, it is possible to measure, non-invasively, the characteristics of the probability distribution of the random paths taken by the molecules of water diffused to the Inside our cells. By understanding the geometry of diffusion in the brain, we can see the invisible: distinguishing between white matter and gray matter, mapping the bundles of nerve fibers and extracting statistical information about microscopic cellular structures.

In his doctoral dissertation, MSc Jia Liu developed and implemented a series of new computational statistical methodologies to process MRI-diffusion data. The statistical problem of the thesis stems from clinical needs for the diagnosis of brain diseases such as Lewy-binding dementia.

MRI-diffusion data are "big data" with typically 100,000 volume elements (voxels) in a complete brain scan and hundreds of data points for each voxel. Although the new generation of commercial MRI scanners has dramatically shortened data acquisition times and prices are also falling, performing a patient's MRI is expensive and time consuming. the MR scanner for hours to acquire more detailed data.

On the other hand, the time and cost of data processing has decreased, and there is a strong demand for advanced statistical techniques capable of producing accurate results without increasing the amount of data. Any possible improvement in the modeling and calculation of MRI-diffusion statistics may have an impact on the diagnosis of brain diseases and other neuroscience domains.

In his doctoral dissertation, M.Sc. Jia Liu introduces new data augmentation schemes to simplify the likelihood function by including new latent observations in the model, thus enabling more efficient and accurate calculations. These ideas have been implemented in different algorithms, such as expectation-maximization (EM), the Monte-Carlo chain Markov chain (McMC) and Bayesian variation (VB) under various diffusion parameterizations. MRI. The statistical methods developed are more efficient than the existing methods.