At the Center for Lifespan Changes in Brain and Cognition (LCBC, https://www.oslobrains.no/), we study how the human brain develops over the whole lifespan. We’re particularly interested in how physical changes in the brain are related to changes in cognitive function (e.g., decline in memory), and if it’s possible for people to improve their own brain health, e.g., through memory training or physical exercise. All of our work requires large datasets and complex statistical models, and this document describes some possible master thesis projects which would be of high interest to us.
Project description
When studying how the brain develops, we use magnetic resonance imaging (MRI). MRI of a single brain generates 3D images we >300,000 pixels, giving a fine-grained map of the density of brain cells. We then use these images to understand how peoples’ brains differ, e.g., between a group having a mental disease and healthy controls, or between people of different age. This gives maps highlighting the regions where we find differences between the brains, as in Figure 1.
Typical datasets contain tens of thousands of people, and many of them have been followed over time, so that we have multiple measurements of the same person. This requires mixed-effect models, since repeated measures of the same person are more correlated than measurements of different persons. We also need to account for the spatial correlation, since the number of brain cells in neighboring points are correlated. This requires spatiotemporal mixed models, and commonly used method is described in (Bernal-Rusiel et al., 2013). Unfortunately, no R package exists for analyzing MRI data using these spatiotemporal models. The whole process of importing data, defining and running models, and finally plotting the results, requires writing scripts in four different programming languages(!!). The available implementation of the spatiotemporal models is also slow and inflexible.
The goal of this project is to help improving this, by creating R scripts which allow running the whole analysis in R: importing data, defining models, fitting models, and plotting the results. The main part of the work will be to extend the R package lme4 (Bates et al., 2015) so that it can fit the spatiotemporal models used for MRI data. The scripts developed by the student will later be part of an R package that we will develop. The key tasks will be to (1) understand the spatiotemporal models used in MRI analysis, (2) understand the modular fitting functions provided by lme4, and then (3) try to figure out how the spatiotemporal models can be fitted using lme4.
Some experience with R and numerical linear algebra will be useful, but more important is that the student is motivated to learn more about sparse matrix methods and parallel computing, which will be important tools in this work. You will also learn a lot about developing statistical software, including version control (GitHub), unit testing, and documentation.
References
Bates, D. M., Mächler, M., Bolker, B., & Walker, S. (2015). Fitting Linear Mixed-Effects Models Using lme4. Journal of Statistical Software, 67(1), 1–48. https://doi.org/10.18637/jss.v067.i01
Bernal-Rusiel, J. L., Reuter, M., Greve, D. N., Fischl, B., & Sabuncu, M. R. (2013). Spatiotemporal linear mixed effects modeling for the mass-univariate analysis of longitudinal neuroimage data. NeuroImage, 81, 358–370. https://doi.org/10.1016/j.neuroimage.2013.05.049