29915
EEG Integrated Platform (EEG-IP): Repository of Pre-Processed EEG Data Optimized for Signal Retention and Harmonization across Projects.
Poster Presentation
Thursday, May 2, 2019: 11:30 AM-1:30 PM
Room: 710 (Palais des congres de Montreal)
J. Desjardins1, S. van Noordt2, S. Huberty3, L. A-Abbas4, &. The BUCANL Team5, A. C. Evans2, M. H. Johnson6, C. A. Nelson7, S. J. Webb8, E. J. Jones9, H. Tager-Flusberg10, S. Jeste11, A. R. Levin12, S. J. Segalowitz13, &. The BASIS Team14, M. Elsabbagh15 and T. Charman16, (1)SHARCNET, St Catharines, ON, Canada, (2)Montreal Neurological Institute, McGill University, Montreal, QC, Canada, (3)Mcgill University, Montreal, QC, Canada, (4)Research Institute - McGill University Health Centre, Montreal, QC, Canada, (5)Brock University, St Catherine, QC, Canada, (6)Centre of Brain and Cognitive Development, Birkbeck College, University of London, London, United Kingdom, (7)Boston Children's Hospital, Boston, MA, (8)Psychiatry and Behavioral Sciences, University of Washington, Seattle, WA, (9)Centre for Brain and Cognitive Development, Birkbeck, University of London, London, United Kingdom, (10)Psychological and Brain Sciences, Boston University, Boston, MA, (11)University of California, Los Angeles, Los Angeles, CA, (12)Neurology, Boston Children's Hospital, Boston, MA, (13)Psychology, Brock University, St Catharines, ON, Canada, (14)Centre for Brain and Cognitive Development, Birkbeck University of London, London, United Kingdom, (15)McGill University, Montreal, PQ, Canada, (16)Institute of Psychiatry, Psychology and Neuroscience, King’s College London, London, United Kingdom
Background: EEG is a non-intrusive method for capturing brain behavior dynamics at a high temporal resolution. Because of the high availability of EEG it is particularly well suited for accumulating large samples of data from diverse participant populations including infants at varying risk of ASD. The substantial recent advancements in analytic methods for studying brain dynamics in EEG recordings have increased the impact of this modality in integrated neuroscience. Although this technology holds great potential for large samples, the acquisition is prone to artifacts. Further, there are few standards for data acquisition that would facilitate harmonizing data across projects. Finally, substantial pre-processing of EEG data is required to isolate the cortical signals from the various artifact sources. There are no accepted standards for the pre-processing and many of the current strategies are highly reductionistic resulting in substantial loss of potentially important data. Identifying neural correlates to autism in infant EEG could have a large impact on treatment outcomes, however, the lack of standards and methods for harmonizing data across projects for large scale analyses is still a limitation in the field.
Objectives: Implement tools required to produce a transferable “lossless” state for harmonized pre-processed EEG data which places minimal restrictions on post processing procedures. Apply this pre-processing strategy at large scales on HPC resources to contributed data sets of infant EEG from independent research institutions to produce an integrated repository for the study of ASD risk factors.
Methods: Contributing Research groups from Birbeck University of London, University of Washington and the Boston Children’s Hospital shared over 1450 EEG sessions consisting of 446 unique participants spanning 3 age retest trajectories (London: 7, 14 months; Washington: 6, 12, 18 months; Boston: 3, 6, 9, 12, 18, 24, 36 months). The contributed EEG recording were pre-processed using the lossless pipeline that employs robust measures to isolate spatially non-stationary channels and periods of time in the recording (bad channels, movement artifacts, etc), then performs multiple Adaptive Mixture Independent Component Analysis (AMICA) to isolate stationary noise factors from cortical signal (eye movements, blinks, ECG, EMG, etc), as well as isolate the activation of specific cortical generators.
Results: Descriptive statistics of the pre-processed data indicate that the lossless pipeline performed well in isolating cortical signal from noise in all three sites. Although the three contributing sites had varying acquisition properties the resulting cleaned data has similar properties across the sites.
Conclusions: With increasing computational resources at researchers disposal large scale analysis have the potential for significant impact on our understanding of ASD risk factors. While EEG data is well suited for contributing to large scale neuroimaging efforts because of its non-invasive nature and high accessibility, its vulnerability to artifacts and lack of acquisition/processing standards make it difficult to harmonize data across projects for combined analysis or replication. The lossless pipeline provides a method for maximizing the retention of cortical signal from EEG recordings and establishes a state of data that is harmonized across acquisition parameters.
