Enhanced multi-core beamformer algorithm for sensor array signal processing by combining data from magnetoencephalography
Inventors
Huang, Ming-Xiong • Lee, Roland R. • Diwakar, Mithun • Tal, Omer • Liu, Thomas T.
Assignees
US Department of Veterans Affairs • Office of General Counsel of VA • University of California San Diego UCSD
Publication Number
US-9883812-B2
Publication Date
2018-02-06
Expiration Date
2031-06-28
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
Techniques and systems are disclosed for implementing multi-core beamforming algorithms. In one aspect, a method of implementing a beamformer technique includes using a spatial filter that contains lead-fields of two simultaneous dipole sources rather than a linear combination of the two to directly compute and obtain optimal source orientations and weights between two highly-correlated sources.
Core Innovation
This invention relates to devices and techniques using medical imaging technologies, specifically improving magnetoencephalography (MEG) analysis by implementing a multi-core beamformer (MCBF) lead-field based inverse-modeling technique. The MCBF is designed to simultaneously reconstruct highly-correlated and uncorrelated neuronal sources from noisy sensor array data, providing accurate reconstruction of source positions, time-courses, and correlations.
The background identifies the problem that traditional beamformer methods have significant challenges when localizing highly-correlated neuronal sources from noisy MEG data, as these conventional approaches assume uncorrelated source time-courses, leading to suppression of correlated source power estimates. Existing methods like the coherently combining signal-to-interference plus noise ratio beamformer and constant modulus algorithm beamformer achieve moderate success but have limitations such as requiring a priori information. Furthermore, conventional methods cannot easily reconstruct multiple unknown neural pathways due to suppression of correlated sources.
The core innovation addresses these problems by using a spatial filter that concatenates lead-fields of two simultaneous dipole sources to form a multi-core spatial filter matrix, rather than a linear combination of lead-fields, allowing direct computation of optimal source orientations and weights between correlated sources. This approach reduces the computational cost significantly, avoiding the need for non-linear optimization and exhaustive searches. The method includes enhanced dual-core beamformer (eDCBF) and enhanced multi-core beamformer (eMCBF) techniques, which accurately compute weight matrices for multiple sources, detect signal orientations to separate multiple interfering sources, and reconstruct individual source time-courses and correlations even at low signal-to-noise ratios.
Claims Coverage
The patent contains one independent claim defining a method for implementing a multi-core beamformer lead-field based inverse-modeling technique that reconstructs correlated and uncorrelated sources from a sensor array, along with several dependent claims expanding on specific features.
Integration of multiple lead-field vectors into a multi-core spatial filter matrix
Combining a plurality of lead-field vectors for multiple source dipoles, each containing sensor information in three orientations, into a multi-core spatial filter matrix without prior knowledge of source locations.
Construction and decomposition of a matrix inversely proportional to source power
Using the multi-core spatial filter matrix and sensor signal covariance matrix to construct a second matrix inversely proportional to source power, then decomposing this matrix to obtain beamformer power and weighting in three orientations of multiple source dipoles.
Reconstruction of source time-courses and correlation analysis
Constructing multiple time-courses for source dipoles using the obtained matrices, beamformer power, covariance matrix, and spatial filter matrix, including reconstructing correlations based on source power covariance matrices.
Use of magnetoencephalography data for diagnosis
Obtaining data from magnetoencephalography (MEG) and combining it with the reconstructed time-courses to enable diagnosis of neurological and psychiatric disorders.
Construction of a matrix inversely proportional to signal-to-noise ratio (SNR) for relative source activity measurement
Constructing a third matrix inversely proportional to SNR and using it to measure relative source activity to enhance regional connectivity analysis for diagnosis.
Utilization of singular value decomposition in matrix decomposition
Applying singular value decomposition (SVD) as part of decomposing the constructed matrix for obtaining beamformer power and weightings.
Optimization of dipole combination search using independent coordinate axes
Performing a search process for best dipole combinations by independently searching along two coordinate axes to efficiently optimize source localization.
The independent claim and its dependent claims collectively cover a method that innovatively combines multiple lead-fields into a spatial filter matrix, mathematically processes sensor signals via inverse source power matrix decomposition, reconstructs source time-courses and correlations efficiently without requiring prior knowledge of source locations, uses MEG data for diagnosis, and employs advanced search and decomposition algorithms to reduce computational complexity and improve accuracy.
Stated Advantages
Significant reduction in computational time for dual-beamformer technique by a factor of 100 due to direct calculation of optimal amplitude-weighting, source orientations, and correlations.
Capability to handle both highly correlated and uncorrelated sources, which conventional beamformers and previous dual-beamformers cannot achieve.
Accurate reconstruction of individual source time-courses and correlations even at low signal-to-noise ratios.
Enhanced sensitivity for neuroimaging-based diagnosis and monitoring of neurological and psychiatric disorders, surpassing conventional neuroimaging techniques.
Fast and viable MEG source localization method enabling exploration of complex neuronal networks and multiple correlated sources without requiring a priori information.
Robustness of the enhanced multi-core beamformer methods to various levels of source correlations, SNRs, and source locations, with precise amplitude and orientation reconstruction.
Documented Applications
Localization of abnormal neuronal networks using MEG data for sensitive diagnosis of neurological and psychiatric disorders such as traumatic brain injury (TBI), stroke, Post-Traumatic Stress Disorder (PTSD), schizophrenia, Alzheimer's dementia, and Autism.
Recovery of source information from various sensor arrays beyond MEG, including radar, sonar, astronomical telescopes, magnetotelluric sensors, optical and other electromagnetic arrays.
Performing regional connectivity analysis for clinical diagnosis by comparing patient MEG regional connectivity maps with normative databases to detect abnormal neuronal connectivity.
Application in human auditory stimulation studies and right median nerve stimulation experiments to reconstruct correlated neural sources and functional pathways.
Interested in licensing this patent?