Doctor of Philosophy
Brain parcellation studies are fundamental for neuroscience as they serve as a bridge between anatomy and function, helping researchers interpret how functions are distributed across different brain regions. However, two substantial challenges exist in current imaging-based brain parcellation studies: large variations in the functional organization across individuals and the intrinsic spatial dependence which causes nearby brain locations to have a similar function. This thesis presents a series of projects aimed to tackle these challenges from different perspectives by using advanced machine learning techniques.
To handle the challenge of individual variability in building precise individual parcellations, Chapter 3 introduces a novel hierarchical Bayesian brain parcellation framework. This framework learns a brain probabilistic parcellation by integrating across diverse datasets. For single individuals, the framework optimally combines the limited individual data with the group probability map, resulting in improved individual maps. We found that the resultant individual parcellation based on only 10 minutes of imaging scans can achieve an equivalent performance to the ones using 100 minutes of data alone. These improved individual parcellations are essential to accurately capture functional variations across studied populations.
The intrinsic spatial dependence between brain locations poses a significant challenge in both evaluating and generating brain parcellations. To address this, Chapter 2 presents a bias-free method for evaluating different brain parcellations, the distance-controlled boundary coefficient (DCBC). Compared to existing evaluation metrics that bias toward finer and spatial contiguous parcellations due to spatial smoothness, DCBC provides a fair evaluation by controlling the distance of brain location pairs, ensuring a direct comparison of parcellations in different resolutions. To address the intrinsic spatial dependence when learning parcellations, I propose a new model in Chapter 4, the multinomial restricted Boltzmann machine (m-RBM), that can be incorporated into the learning framework in Chapter 3. This model captures spatial structure between brain locations in its architecture. While simulations showed the utility of this type of model in estimating individual parcellations, we could not demonstrate better performance using real imaging data.
Together, this thesis significantly advances the technical toolkit for deriving brain parcellations from functional imaging data. The developments open up new avenues for future research into human brain organization.
Summary for Lay Audience
Understanding how the human brain is organized and how different parts of the brain interact with each other is a critical part of neuroscience research. Just like we usually use maps when exploring a new city, brain parcellation subdivides the brain into different regions based on their functional or structural properties, helping scientists make sense of this intricate organ. But, this is not an easy task. Two main challenges stand in the way by the nature of the brain: the fact (1) that every person's brain has a slightly different organization and (2) that brain locations close to each other tend to function more similarly.
This thesis uses advanced machine learning techniques to tackle these issues. Just like how no two cities are alike, no two brains are identical, so it is important that brain parcellations reflect that. To address the first challenge, we developed a hierarchical Bayesian framework that learns the probability that a brain location belongs to a specific map, rather than learning a fixed map. This enables the model to produce individual brain parcellations by combining the group probability map with data from an individual following the Bayes rule, resulting in a more accurate personalized brain map. The framework is also able to learn combined knowledge from different datasets, thereby making full use of different experiments, each studying a different aspect of brain function.
The thesis also addresses the issue of intrinsic smoothness of the brain, the fact that spatially-nearby brain locations have higher functional similarity than the far-away pairs. This biological characteristic is an important consideration, not only when generating but also when evaluating brain parcellations. Without taking it into account, the boundaries between different functional regions in the trained brain maps might be noisy and difficult to identify, or the evaluation methods could unfairly favor those parcellations that are finer-grained and spatially continuous. To solve these problems, we designed a new computational model based on restricted Boltzmann machines (RBM) that explicitly models the spatial smoothness in the brain. The resultant maps capture the spatial structures between brain locations, helping in a more precise mapping of brain functions to individual brain regions. In addition, we also developed a new way of evaluating these brain maps, which reduces the bias in existing evaluation methods caused by the intrinsic spatial smoothness of the brain, making it a bias-free evaluation approach.
In essence, this thesis presents a series of projects to advance the methodological toolkit for producing brain parcellation maps. The tools may lay the foundation for clinical applications in personalized medicine, and be useful for future studies that address fundamental neuroscience questions.
Zhi, Da, "Machine Learning Techniques for Improved Functional Brain Parcellation" (2023). Electronic Thesis and Dissertation Repository. 9436.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 4.0 License.