Linked independent component analysis for multimodal data fusion

FMRIB (Oxford University Centre for Functional MRI of the Brain), Department Clinical Neurology, University of Oxford, Oxford, UK.
NeuroImage (Impact Factor: 6.36). 10/2010; 54(3):2198-217. DOI: 10.1016/j.neuroimage.2010.09.073
Source: PubMed


In recent years, neuroimaging studies have increasingly been acquiring multiple modalities of data and searching for task- or disease-related changes in each modality separately. A major challenge in analysis is to find systematic approaches for fusing these differing data types together to automatically find patterns of related changes across multiple modalities, when they exist. Independent Component Analysis (ICA) is a popular unsupervised learning method that can be used to find the modes of variation in neuroimaging data across a group of subjects. When multimodal data is acquired for the subjects, ICA is typically performed separately on each modality, leading to incompatible decompositions across modalities. Using a modular Bayesian framework, we develop a novel "Linked ICA" model for simultaneously modelling and discovering common features across multiple modalities, which can potentially have completely different units, signal- and contrast-to-noise ratios, voxel counts, spatial smoothnesses and intensity distributions. Furthermore, this general model can be configured to allow tensor ICA or spatially-concatenated ICA decompositions, or a combination of both at the same time. Linked ICA automatically determines the optimal weighting of each modality, and also can detect single-modality structured components when present. This is a fully probabilistic approach, implemented using Variational Bayes. We evaluate the method on simulated multimodal data sets, as well as on a real data set of Alzheimer's patients and age-matched controls that combines two very different types of structural MRI data: morphological data (grey matter density) and diffusion data (fractional anisotropy, mean diffusivity, and tensor mode).

49 Reads
  • Source
    • "Dimensionality selection was automatically performed by LICA (Groves et al., 2011), yielding 66 components (83 components for the additional analysis with less smoothing). In LICA, each component has a shared individual subject weight (loading) across modalities, indicating the degree to which each subject contributes to that component. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Schizophrenia (SZ) is a psychotic disorder with significant cognitive dysfunction. Abnormal brain activation during cognitive processing has been reported, both in task-positive and task-negative networks. Further, structural cortical and subcortical brain abnormalities have been documented, but little is known about how task-related brain activation is associated with brain anatomy in SZ compared to healthy controls (HC). Utilizing linked independent component analysis (LICA), a data-driven multimodal analysis approach, we investigated structure–function associations in a large sample of SZ (n = 96) and HC (n = 142). We tested for associations between task-positive (fronto-parietal) and task-negative (default-mode) brain networks derived from fMRI activation during an n-back working memory task, and brain structural measures of surface area, cortical thickness, and gray matter volume, and to what extent these associations differed in SZ compared to HC. A significant association (p b .05, corrected for multiple comparisons) was found between a component reflecting the task-positive fronto-parietal network and another component reflecting cortical thickness in fronto-temporal brain regions in SZ, indicating increased activation with increased thickness. Other structure–function associations across, between and within groups were generally moderate and significant at a nominal p-level only, with more numerous and stronger associations in SZ compared to HC. These results indicate a complex pattern of moderate associations between brain activation during cognitive processing and brain morphometry, and extend previous findings of fronto-temporal brain abnormalities in SZ by suggesting a coupling between cortical thickness of these brain regions and working memory-related brain activation.
  • Source
    • "atrophy and white matter diffusion disintegration with a model-free approach [116]. In separate independent components, white matter diffusion alterations were found with and without grey matter atrophy. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Current therapies for Alzheimer's disease (AD) offer partial symptomatic relief and do not modify disease progression. There is substantial evidence indicating a disease onset years before clinical diagnosis, at which point no effective therapy has been found. In this study, we investigated the efficacy of a new multi-target drug, M30, at relatively early stages of the AD-like amyloid pathology in a robust rat transgenic model. McGill-R-Thy1-APP transgenic rats develop the full AD-like amyloid pathology in a progressive fashion, and have a minimal genetic burden. McGill rats were given 5 mg/kg M30 or vehicle per os, every 2 days for 4 months, starting at a stage where the transgenic animals suffer detectable cognitive impairments. At the completion of the treatment, cognitive functions were assessed with Novel Object Location and Novel Object Recognition tests. The brains were then analyzed to assess amyloid-β (Aβ) burden and the levels of key inflammatory markers. Longterm treatment with M30 was associated with both the prevention and the reversal of transgene-related cognitive decline. The effects on cognition were accompanied by a shift of the Aβ-immunoreactive material toward an amyloid plaque aggregated molecular form, diminished molecular signs of CNS inflammation and a change in microglia morphology toward a surveying phenotype. This study is the first to demonstrate the therapeutic potential of M30 in a rat model of the AD amyloid pathology. It provides a rationale for further investigations with M30 and with potential multi-target approaches to delay, prevent or reverse the progression the AD pathology at early disease-stages.
  • Source
    • "A limitation of this early approach is that it will not be applicable to modalities that exhibit different numbers of sources or for non-perfect correlations among mixing matrices. In [2], the authors present a probabilistic approach based on a modular Bayesian framework. This method has two configurations, one forces a common modal map (sources) and the other shares the same mixing matrix among modalities. "
    [Show abstract] [Hide abstract]
    ABSTRACT: In the biomedical field, current technology allows for the collection of multiple data modalities from the same subject. In consequence, there is an increasing interest for methods to analyze multi-modal data sets. Methods based on independent component analysis have proven to be effective in jointly analyzing multiple modalities, including brain imaging and genetic data. This paper describes a new algorithm, three-way parallel independent component analysis (3pICA), for jointly identifying genomic loci associated with brain function and structure. The proposed algorithm relies on the use of multi-objective optimization methods to identify correlations among the modalities and maximally independent sources within modality. We test the robustness of the proposed approach by varying the effect size, cross-modality correlation, noise level, and dimensionality of the data. Simulation results suggest that 3p-ICA is robust to data with SNR levels from 0 to 10 dB and effect-sizes from 0 to 3, while presenting its best performance with high cross-modality correlations, and more than one subject per 1,000 variables. In an experimental study with 112 human subjects, the method identified links between a genetic component (pointing to brain function and mental disorder associated genes, including PPP3CC, KCNQ5, and CYP7B1), a functional component related to signal decreases in the default mode network during the task, and a brain structure component indicating increases of gray matter in brain regions of the default mode region. Although such findings need further replication, the simulation and in-vivo results validate the three-way parallel ICA algorithm presented here as a useful tool in biomedical data decomposition applications.
    2014 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBC 2014; 08/2014
Show more


49 Reads
Available from