In this paper we construct an atlas that summarizes functional connectivity characteristics of a cognitive process from a population of individuals. The atlas encodes functional connectivity structure in a low-dimensional embedding space that is derived from a diffusion process on a graph that represents correlations of fMRI time courses. The functional atlas is decoupled from the anatomical space, and thus can represent functional networks with variable spatial distribution in a population. In practice the atlas is represented by a common prior distribution for the embedded fMRI signals of all subjects. We derive an algorithm for fitting this generative model to the observed data in a population. Our results in a language fMRI study demonstrate that the method identifies coherent and functionally equivalent regions across subjects. The method also successfully maps functional networks from a healthy population used as a training set to individuals whose language networks are affected by tumors.
Presurgical language mapping for patients with lesions close to language areas is critical to neurosurgical decision-making for preservation of language function. As a clinical noninvasive imaging technique, functional MRI (fMRI) is used to identify language areas by measuring blood-oxygen-level dependent (BOLD) signal change while patients perform carefully timed language vs. control tasks. This task-based fMRI critically depends on task performance, excluding many patients who have difficulty performing language tasks due to neurologic deficits. On the basis of recent discovery of resting-state fMRI (rs-fMRI), we propose a "task-free" paradigm acquiring fMRI data when patients simply are at rest. This paradigm is less demanding for patients to perform and easier for technologists to administer. We investigated the feasibility of this approach in right-handed healthy control subjects. First, group independent component analysis (ICA) was applied on the training group (14 subjects) to identify group level language components based on expert rating results. Then, four empirically and structurally defined language network templates were assessed for their ability to identify language components from individuals' ICA output of the testing group (18 subjects) based on spatial similarity analysis. Results suggest that it is feasible to extract language activations from rs-fMRI at the individual subject level, and two empirically defined templates (that focuses on frontal language areas and that incorporates both frontal and temporal language areas) demonstrated the best performance. We propose a semi-automated language component identification procedure and discuss the practical concerns and suggestions for this approach to be used in clinical fMRI language mapping.