AUTHOR=Xiang Jing , Luo Qian , Kotecha Rupesh , Korman Abraham , Zhang Fawen , Luo Huan , Fujiwara Hisako , Hemasilpin Nat , Rose Douglas F. 

TITLE=Accumulated source imaging of brain activity with both low and high-frequency neuromagnetic signals

JOURNAL=Frontiers in Neuroinformatics

VOLUME=Volume 8 - 2014

YEAR=2014

URL=https://www.frontiersin.org/journals/neuroinformatics/articles/10.3389/fninf.2014.00057

DOI=10.3389/fninf.2014.00057

ISSN=1662-5196

ABSTRACT=Recent studies have revealed the importance of high-frequency brain signals (>70 Hz).  One challenge of high-frequency signal analysis is that the size of time-frequency representation of high-frequency brain signals could be larger than 1 terabytes (TB), which is beyond the upper limits of a typical computer workstation’s memory (<196 GB).  The aim of the present study is to develop a new method to provide greater sensitivity in detecting high-frequency magnetoencephalography (MEG) signals in a single automated and versatile interface, rather than the more traditional, time-intensive visual inspection methods, which may take up to several days.  To address the aim, we developed a new method, accumulated source imaging, defined as the volumetric summation of source activity over a period of time.  This method analyzes signals in both low- (1~70 Hz) and high-frequency (70~200 Hz) ranges at source levels.  To extract meaningful information from MEG signals at sensor space, the signals were decomposed to channel-cross-channel matrix (CxC) representing the spatiotemporal patterns of every possible sensor-pair.  A new algorithm was developed and tested by calculating the optimal CxC and source location-orientation weights for volumetric source imaging, thereby minimizing multi-source interference and reducing computational cost.  The new method was implemented in C/C++ and tested with MEG data recorded from clinical epilepsy patients.  The results of experimental data demonstrated that accumulated source imaging could effectively summarize and visualize MEG recordings within 12.7 hours by using approximately 10 GB of computer memory.  In contrast to the conventional method of visually identifying multi-frequency epileptic activities that traditionally took 2-3 days and used 1-2 TB storage, the new approach can quantify epileptic abnormalities in both low- and high-frequency ranges at source levels, using much less time and computer memory.