Image-based Modeling of Biological Shape and Function

Polina Golland
Computer Science and Artificial Intelligence Laboratory

Understanding how the brain develops and how diseases affect its anatomy and function continue to be among the most intriguing questions in neuroscience and clinical research. Magnetic Resonance Imaging (MRI) provides high resolution scans of the brain structure and its activity. With appropriate interpretation techniques, these images can shed light on how the brain changes in normal development and under the influence of disease, providing significant benefits to both science and medicine. Developing mathematical models and computational methods that enable such descriptions from MRI images is at the core of our research.

Recently, we developed an approach to modeling a population of images that leads to an efficient algorithm for clustering images based on their similarity while also accounting for possible deformations across individuals. The output of the algorithm is a small number of templates that represent homogeneous sub-populations in the image set. By comparing the groupings with demographic and clinical information, such as age and diagnosis, we can characterize and understand the changes induced by neurological diseases and detect sub-types of the disorders.

Figure 1 shows three templates identified by the algorithm as representative of the entire population. It also shows the histograms of ages associated with each templates for a model based on two templates (left) and a model based on three templates (right). In the three-template model, the probable Alzheimer’s patients were significantly more likely associated with the third template than the healthy controls of matched age. (photo: Polina Golland)

Figure 1 shows three templates identified by the algorithm as representative of the entire population. It also shows the histograms of ages associated with each templates for a model based on two templates (left) and a model based on three templates (right). In the three-template model, the probable Alzheimer’s patients were significantly more likely associated with the third template than the healthy controls of matched age. (photo: Polina Golland)

As an example, when applied to a set of 400 brain scans from the aging and Alzheimer’s study (provided by Prof. Randy Buckner, Harvard University), the algorithm identified templates that clearly represented changes in the brain anatomy that correspond to aging. Increasing the number of requested templates caused the algorithm to focus on the older subjects more than their representation in the data set would warrant. The grouping also correlated strongly with the results of behavioral and memory tests. In particular, the subjects diagnosed with probable Alzheimer’s were more likely to belong to one of the identified groups, while the normal controls of similar ages clustered separately. This finding suggests higher variability in the older population, potentially related to the onset of Alzheimer’s disease.

In a different project, we demonstrated a novel approach to computationally modeling the patterns of brain activity that can be seen in fMRI. Our algorithm analyzes the imaging data and groups locations in the brain, where activation patterns are similar, into anatomically defined regions which are hierarchically organized. These detected hierarchies represent a robust and anatomically meaningful model for patterns of co-activation in fMRI.

Our work offers new computational tools for exploring functional organization of the brain and discovery of novel systems whose coordinated action represents brain activity and is captured by non-invasive fMRI imaging. It also promises to help identify perturbations of the functional systems due to disorders such as schizophrenia and Alzheimer’s disease, as well as pathologies, such as brain tumors.

Leave a Reply

Replies which add to the content of this article in the EECS Newsletter are welcome. The Department reserves the right to moderate all comments. If you would like to provide any updated information please send an email to newsletter@eecs.mit.edu.