A method for multitask fMRI data fusion applied to schizophrenia

Abstract

Abstract:It is becoming common to collect data from multiple functional magnetic resonance imaging (fMRI) paradigms on a single individual. The data from these experiments are typically analyzed separately and sometimes directly subtracted from one another on a voxel-by-voxel basis. These comparative approaches, although useful, do not directly attempt to examine potential commonalities between tasks and between voxels. To remedy this we propose a method to extract maximally spatially independent maps for each task that are "coupled" together by a shared loading parameter. We first compute an activation map for each task and each individual as "features," which are then used to perform joint independent component analysis (jICA) on the group data. We demonstrate our approach on a data set derived from healthy controls and schizophrenia patients, each of which carried out an auditory oddball task and a Sternberg working memory task. Our analysis approach revealed two interesting findings in the data that were missed with traditional analyses. First, consistent with our hypotheses, schizophrenia patients demonstrate "decreased" connectivity in a joint network including portions of regions implicated in two prevalent models of schizophrenia. A second finding is that for the voxels identified by the jICA analysis, the correlation between the two tasks was significantly higher in patients than in controls. This finding suggests that schizophrenia patients activate "more similarly" for both tasks than do controls. A possible synthesis of both findings is that patients are activating less, but also activating with a less-unique set of regions for these very different tasks. Both of the findings described support the claim that examination of joint activation across multiple tasks can enable new questions to be posed about fMRI data. Our approach can also be applied to data using more than two tasks. It thus provides a way to integrate and probe brain networks using a variety of tasks and may increase our understanding of coordinated brain networks and the impact of pathology upon them. Hum Brain Mapp 27:598-610, 2006.

Feature-Based Fusion of Medical Imaging Data

“…An advantage of ICA over variance-based approaches like SVD or principal component analysis (PCA) is the use of higher order statistics to reveal hidden structure [19], [20]. We have recently done work showing the value of combining multitask fMRI data [21], fMRI and sMRI data [22], and fMRI and ERP data [23]. One important aspect of the approach is that it allows for the possibility that a change in a certain location in one modality is associated with a change in a different location in another modality (or, in the case of ERP, one is associating time in ERP with space in fMRI) as we demonstrate with a number of examples in this paper.…”

“…Next, we present a feature-based fusion approach that provides a general framework for fusing information from multiple data types, such as multitask fMRI data, or fMRI and event-related potential (ERP) data. The extracted features for each data type are fused using a data-driven analysis technique, ICA, which has proved quite fruitful for medical image analysis [23]- [27]. The fusion framework we present thus enables the discovery of relationships among data types for given samples, for example, at the group level, to study variations between patients and controls.…”

BrainZoom: High Resolution Reconstruction from Multi-modal Brain Signals

“…The latter include popular multi-modal fusion techniques that rely on factor analysis tools, such as Independent Component Analysis (ICA) which aims to recover statistically independent latent sources of brain activation; and Canonical Correlation Analysis (CCA) which aims to unravel sources of co-variation in multiple 'views' of the brain stemming from the different modalities. Variants of PCA [6], ICA [5], and CCA [8] have been studied for multimodal fusion of various brain imaging modalities. See [2,3,9,10,13,20,21] for additional background on existing localization and common component extraction methods.…”

Joint Decompositions with Flexible Couplings

“…Although the framework described above dates back to the coupled (or linked) decomposition model described in [11], it was reexplored in independent component analysis approaches [6,7,9] and repopularized recently in data science in [17], where the problem of joint nonnonegative matrix factorization was considered under the constraint that one of the factors is shared by all matrices. In all these cases, the coupling occurs through equality constraints on latent factors.…”