Multivariate feature selection for fMRI analysis
Eralp Kumbasar, Dr. Tülay Adali, Department of Computer Science and Electrical Engineering.
fMRI has become a widely used imaging tool for exploring the normal neural functions as well as disordered brain functions like schizophrenia. Among all fMRI data analysis strategies, data-driven-based methods have a unique advantage of capturing the whole picture of available information since they effectively minimize assumptions imposed on the brain activity. With the increasing number of multimodal data and multisite data, the problem of balancing the computation cost and analysis performance is becoming more important than ever before. In this project, our interest is in identifying the most informative multivariate features when analyzing multiple fMRI datasets. Our goal is the development of flexible new decomposition methods as well as identifying the best feature extraction strategy for a given problem.
Constrained and seeded independent vector analysis for steady state visually evoked potential processing and detection
Tulay Adali, CSEE
Zois Boukouvalas, Mathematics & Statistics
Rami Mowakeaa, CSEE
Darren Emge, CSEE
The detection of steady state visual evoked potentials (SSVEPs) has been identified as an effective solution for brain computer interface (BCI) systems as well as for neurocognitive investigations of visually related tasks. SSVEPs are induced at the same frequency as the visual stimuli being presented and can be observed in the scalp-based recordings of electroencephalogram signals, though they are one component buried amongst the normal brain signals and complex noise. Variations in individual hemodynamic responses as well as the presence of multiple biological artifacts complicate the use of direct frequency analysis making blind source separation methods, e.g., independent component (ICA) and independent vector analysis (IVA), desirable solutions for these applications. IVA has been shown capable of enhancing and improving detection of SSVEPs by exploiting the complimentary information that exists across EEG channels. In this work, we present a novel implementation of IVA called seeded IVA, which incorporates a priori information to seed or bias the source estimation, thus improving convergence and reducing the computational time associated with standard IVA implementations
Independent Component Analysis by Entropy Maximization with Kernels
Tulay Adali, Department of Electrical and Computer Engineering
Zois Boukouvalas, Department of Mathematics and Statistics
Rami Mowakeaa, Department of Electrical and Computer Engineering
Independent component analysis (ICA) is the most popular method for blind source separation (BSS) with a diverse set of applications, such as biomedical signal processing, video and image analysis, and communications. Maximum likelihood (ML), an optimal theoretical framework for ICA, requires knowledge of the true underlying probability density function (PDF) of the latent sources, which, in many applications, is unknown. ICA algorithms cast in the ML framework often deviate from its theoretical optimality properties due to poor estimation of the source PDF. Therefore, accurate estimation of source PDFs is critical in order to avoid model mismatch and poor ICA performance. In this paper, we propose a new and efficient ICA algorithm based on entropy maximization with kernels, (ICA-EMK), which uses both global and local measuring functions as constraints to dynamically estimate the PDF of the sources with reasonable complexity. In addition, the new algorithm performs optimization with respect to each of the cost function gradient directions separately, enabling parallel implementations on multi-core computers. We demonstrate the superior performance of ICA-EMK over competing ICA algorithms using simulated as well as real-world data.
Study of Brain Function
Tulay Adali, Department of Computer Science and Electrical Engineering
Sai Ma
Pedro Rodriguez
Hualiang Li
Vince Calhoun
The brain is an incredibly complex organ of interrelated structural and functional connectivity. Data-driven analysis methods, such as ICA, have proven very useful in the study of brain function, in particular when the dynamics are hard to model and underlying assumptions about the data have to be minimized. We work on the development of a rich array of methods to analyze and to fuse the
information from multiple modalities so as to understand the brain function in both the healthy and the diseased brain.
Optimization of multi-set data analysis and fusion for Big Data
Zhongqiang Luo, Sichuan University of Science and Engineering (SUSE)
Tulay Adali, Department of Computer Science and Electrical Engineering, UMBC
Optimization of multi-set analysis methods is of utmost importance especially when dealing with big datasets. Our goal is to study and develop algorithms for this purpose by evaluating their performance with increasing dimensionality so that they can be reliably applied to problems in data fusion for medical image analysis and beyond.
Multivariate data-driven solutions for static and dynamic brain connectivity
Tulay Adali, Department of Computer Science and Electrical Engineering, UMBC.
There is considerable evidence that disruption of time-varying connectivity in mental illness is more sensitive than static connectivity. We develop and test multivariate data-driven methods to estimate the changes of interest in functional magnetic resonance imaging (fMRI) data a robust manner.
Multimodal Data Fusion
Tulay Adali, Department of Computer Science and Electrical Engineering
Ruchir Saheba, Department of Computer Science and Electrical Engineering
Data collection from different sensors or modalities is becoming increasingly popular in neurological studies, since each modality is expected to provide unique, yet complementary, information about the brain function. Maximizing the utilization of the joint information available in such interrelated modalities is, therefore, the fundamental motivation for performing a fusion on multimodal data. Since the relationship among modalities are is not well understood, it is important to reduce the assumptions placed on the data and let the modalities fully interact with each other. To that end, the emphasis of this research project is to develop data-driven fusion methods based on blind source separation (BSS) techniques such as independent component analysis (ICA), its multi-dataset version independent vector analysis (IVA) and canonical correlation analysis (CCA) to jointly analyze multimodal neurological data. Data-driven fusion methods minimize modelling assumptions placed on the data, and thus enables full interaction among the modalities. This results in meaningful decompositions of the multimodal data that can be either used as informative features or interpretable biomarkers of disease.
Optimization of multi-set data analysis and fusion for big data
Dr.Tulay Adali, Department of Computer Science and Electrical Engineering.
Optimization of multi-set analysis methods is of utmost importance, especially when dealing with big datasets. Our goal is to study and develop algorithms for this purpose by evaluating their performance with increasing dimensionality so that they can be reliably applied to problems in data fusion for medical image analysis and beyond.