Robert M. Frank

Development of NeuroElectroMagnetic ontologies(NEMO): a framework for mining brainwave ontologies

Event-related potentials (ERP) are brain electrophysiological patterns created by averaging electroencephalographic (EEG) data, time-locking to events of interest (e.g., stimulus or response onset). In this paper, we propose a generic framework for mining anddeveloping domain ontologies and apply it to mine brainwave (ERP) ontologies. The concepts and relationships in ERP ontologies can be mined according to the following steps: pattern decomposition, extraction of summary metrics for concept candidates, hierarchical clustering of patterns for classes and class taxonomies, and clustering-based classification and association rules mining for relationships (axioms) of concepts. We have applied this process to several dense-array (128-channel) ERP datasets. Results suggest good correspondence between mined concepts and rules, on the one hand, and patterns and rules that were independently formulated by domain experts, on the other. Data mining results also suggest ways in which expert-defined rules might be refined to improve ontologyrepresentation and classification results. The next goal of our ERP ontology mining framework is to address some long-standing challenges in conducting large-scale comparison and integration of results across ERP paradigms and laboratories. In a more general context, this work illustrates the promise of an interdisciplinary research program, which combines data mining, neuroinformatics andontology engineering to address real-world problems.

A single-trial analytic framework for EEG analysis and its application to target detection and classification

Modern neuroimaging technologies afford a non-invasive view into the functions of the human brain with great spatial (fMRI) and temporal resolution (EEG). However, common signal analytic methods require averaging over many trials, which limits the potential for practical application of these technologies. In this paper we advance a novel single-trial analysis method for EEG and demonstrate this approach with a target detection task. The method utilizes a framework consisting of multiple processing modules that can be applied in whole or in part, including noise mitigation, source-space transformation, discriminant analysis, and performance evaluation. The framework introduces an enhanced noise mitigation technology based on Directed Components Analysis (DCA) that improves upon existing spatial filtering techniques. Source-space transformation, utilizing a finite difference model (FDM) of the human head, estimates activity measures of the cortical sources involved in task performance. Such a source-space discrimination provides measurement invariance between training and testing sessions and holds the promise of providing a degree of classification not possible with scalp-recorded EEG. The frameworks discrimination modules interface with performance evaluation modules to generate classification performance statistics. When applied to EEG acquired during performance of a target detection task, this method demonstrated that neural signatures of target recognition correctly classified up to 87% of targets in a rapid serial visual presentation (RSVP) of target/non-target images. On average, the single-trial classification method resulted in greater than 60% improvement over behavioral performance for target detection.

Automated protocol for evaluation of electromagnetic component separation (APECS): Application of a framework for evaluating statistical methods of blink extraction from multichannel EEG

OBJECTIVE We present APECS (Automated Protocol for Evaluation of Electromagnetic Component Separation), a framework for evaluating the accuracy of blind source separation algorithms in removing artifacts from EEG data. APECS applies multiple, automated procedures to quantify the extent to which blinks are removed, and the degree to which nonocular activity is left intact. METHODS APECS was used to evaluate blink removal using three BSS algorithms: Second-Order Blind Inference (SOBI) and two Independent Component Analysis (ICA) implementations, FastICA and Infomax. The algorithms were applied to a series of blink-free EEG datasets, which were contaminated with real or simulated blinks. Extracted components were assumed to contain blink activity if correlation of their spatial projectors to a predefined blink template exceeded some threshold, and if polarity inverted above and below the eyes. Blink-related components were then subtracted to produce filtered data. The success of each data decomposition is evaluated through the use of multiple, automated metrics, to determine which decomposition best approximates the ideal solution (complete separation of blink from nonblink activity). RESULTS The outcomes for the evaluation measures were generally congruent, but also provided different and complementary information about the quality of each data decomposition. Under our testing framework, Infomax outperformed both FastICA and SOBI. Best results were achieved when blink activity loaded onto a single component. CONCLUSIONS Multiple metrics, both quantitative and qualitative, are important in evaluating algorithms for artifact extraction. SIGNIFICANCE Failure to achieve complete separation of blink from nonblink activity can affect experimental outcomes, as illustrated here, using an ERP study of word-nonword discrimination. This illustrates the importance of methods for evaluation of artifact extraction results.

A framework to support automated classification and labeling of brain electromagnetic patterns

This paper describes a framework for automated classification and labeling of patterns in electroencephalographic (EEG) and magnetoencephalographic (MEG) data. We describe recent progress on four goals: 1) specification of rules and concepts that capture expert knowledge of event-related potentials (ERP) patterns in visual word recognition; 2) implementation of rules in an automated data processing and labeling stream; 3) data mining techniques that lead to refinement of rules; and 4) iterative steps towards system evaluation and optimization. This process combines top-down, or knowledge-driven, methods with bottom-up, or data-driven, methods. As illustrated here, these methods are complementary and can lead to development of tools for pattern classification and labeling that are robust and conceptually transparent to researchers. The present application focuses on patterns in averaged EEG (ERP) data. We also describe efforts to extend our methods to represent patterns in MEG data, as well as EM patterns in source (anatomical) space. The broader aim of this work is to design an ontology-based system to support cross-laboratory, cross-paradigm, and cross-modal integration of brain functional data. Tools developed for this project are implemented in MATLAB and are freely available on request.

Minimal Information for Neural Electromagnetic Ontologies (MINEMO): A standards-compliant method for analysis and integration of event-related potentials (ERP) data

We present MINEMO (Minimal Information for Neural ElectroMagnetic Ontologies), a checklist for the description of event-related potentials (ERP) studies. MINEMO extends MINI (Minimal Information for Neuroscience Investigations)to the ERP domain. Checklist terms are explicated in NEMO, a formal ontology that is designed to support ERP data sharing and integration. MINEMO is also linked to an ERP database and web application (the NEMO portal). Users upload their data and enter MINEMO information through the portal. The database then stores these entries in RDF (Resource Description Framework), along with summary metrics, i.e., spatial and temporal metadata. Together these spatial, temporal, and functional metadata provide a complete description of ERP data and the context in which these data were acquired. The RDF files then serve as inputs to ontology-based labeling and meta-analysis. Our ultimate goal is to represent ERPs using a rich semantic structure, so results can be queried at multiple levels, to stimulate novel hypotheses and to promote a high-level, integrative account of ERP results across diverse study methods and paradigms.

Parallel ICA methods for EEG neuroimaging

HiPerSAT, a C++ library and tools, processes EEG data sets with ICA (independent component analysis) methods. HiPerSAT uses BLAS, LAPACK, MPI and OpenMP to achieve a high performance solution that exploits parallel hardware. ICA is a class of methods for analyzing a large set of data samples and extracting independent components that explain the observed data. ICA is used in EEG research for data cleaning and separation of spatiotemporal patterns that may reflect different underlying neural processes. We present two ICA implementations (FastICA and Info-max) that exploit parallelism to provide an EEG component decomposition solution of higher performance and data capacity than current MATLAB-based implementations. Experimental results and the methodology used to obtain them are presented. Integrating HiPerSAT with EEGLAB (A. Delorme and S. Makeig, 2004) is described, as well as future plans for this research.

A Semi-Automatic Framework for Mining ERP Patterns

Event-related potentials (ERP) are brain electrophysiological patterns created by averaging electroencephalographic (EEG) data, time-locking to events of interest (e.g., stimulus or response onset). In this paper, we propose a semi-automatic framework for mining ERP data, which includes the following steps: PCA decomposition, extraction of summary metrics, unsupervised learning (clustering) of patterns, and supervised learning, i.e. discovery, of classification rules. Results show good correspondence between rules that emerge from decision tree classifiers and rules that were independently derived by domain experts. In addition, data mining results suggested ways in which expert- defined rules might be refined to improve pattern representation and classification results.

Directed Components Analysis: An Analytic Method for the Removal of Biophysical Artifacts from EEG Data

Artifacts generated by biophysical sources (such as muscles, eyes, and heart) often hamper the use of EEG for the study of brain functions in basic research and applied settings. These artifacts share frequency overlap with the EEG, making frequency filtering inappropriate for their removal. Spatial decomposition methods, such as principal and independent components analysis, have been employed for the removal of the artifacts from the EEG. However, these methods have limitations that prevent their use in operational environments that require real-time analysis. We have introduced a directed components analysis (DCA) that employs a spatial template to direct the selection of target artifacts. This method is computationally efficient, allowing it to be employed in real-world applications. In this paper, we evaluate the effect of spatial undersampling of the scalp potential field on the ability of DCA to remove blink artifacts.

Ontology-Based mining of brainwaves: a sequence similarity technique for mapping alternative features in event-related potentials (ERP) data

In this paper, we present a method for identifying correspondences, or mappings, between alternative features of brainwave activity in event-related potentials (ERP) data The goal is to simulate mapping across results from heterogeneous methods that might be used in different neuroscience research labs The input to the mapping consists of two ERP datasets whose spatiotemporal characteristics are captured by alternative sets of features, that is, summary spatial and temporal measures capturing distinct neural patterns that are linked to concepts in a set of ERP ontologies, called NEMO (Neural ElectroMagnetic Ontologies) [3, 6] The feature value vector of each summary metric is transformed into a point-sequence curve, and clustering is performed to extract similar subsequences (clusters) representing the neural patterns that can then be aligned across datasets Finally, the similarity between measures is derived by calculating the similarity between corresponding point-sequence curves Experiment results showed that the proposed approach is robust and has achieved significant improvement on precision than previous algorithms.

Instrumentation and Signal Processing for Low-frequency Bounded-EIT Studies of the Human Head

In this paper we describe the instrumentation and signal processing we have implemented as part of our bounded-EIT experiments performed with human subjects. Our hardware/software solution facilitates accurate and low-cost data collection for noninvasive conductivity estimation of human head tissues in vivo.