Iterative Sparse Source Imaging for EEG Brain Activity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional EEG/MEG source imaging techniques face challenges in accurately determining the extent of underlying brain sources due to overly smooth solutions and the need for subjective thresholding to separate active from background activity, which limits the objective estimation of source extent.
Innovation Solution
The iteratively reweighted edge sparsity minimization (IRES) strategy uses a sensor array and processor to record electrical activity, generate an initial estimate, impose edge sparsity, and iteratively reweight optimization problems to converge on a more accurate estimation of underlying source distribution, eliminating background activity and creating clear edges between active and background sources without requiring subjective thresholds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Stability of the object's composition
If conventional EEG/MEG source imaging techniques use minimum norm or LORETA methods with L2 norm regularization, then the solution is smooth and stable, but the estimated source distribution becomes overly smoothed and extended, making it difficult to distinguish active cortical regions from background activity
Solution Approach 1:
The patent changes the regularization norm parameter from L2 (Euclidean) to L1 (Manhattan) norm. This parameter change transforms the optimization problem to produce sparse solutions with clear edges between active and background regions, while maintaining solution stability through the convex optimization framework.
Solution Approach 2:
Instead of minimizing the L2 norm to achieve smooth solutions, the patent inverts the approach by minimizing the L1 norm. This inversion leads to sparse solutions that naturally create clear boundaries between active cortical regions and background activity, eliminating the need for subjective thresholding.
2Adaptability or versatility
If distributed source models use a large number of dipoles distributed within the brain volume, then the model becomes more realistic for extended functional areas, but the problem becomes highly underdetermined with many more unknowns than measurements
Solution Approach 1:
The patent applies L1 norm regularization to the distributed source model, which induces sparsity in the solution. This parameter change effectively reduces the number of active unknowns by driving many dipole amplitudes to zero, thereby making the highly underdetermined problem more tractable while preserving the ability to model extended functional areas.
3Productivity
If equivalent dipole models assume electrical activity can be represented by a small number of dipoles, then the inverse problem becomes over-determined, but this leads to a nonlinear optimization problem requiring a priori determination of the number of dipoles
Solution Approach 1:
The patent uses L1 norm regularization in the distributed source model, which creates a convex optimization problem that is computationally efficient to solve. This approach eliminates the need for nonlinear optimization and a priori determination of dipole numbers, while still producing focused solutions that effectively represent the underlying sources.
4Measurement precision
If subjective thresholding is applied to separate pertinent source activity from background activity, then active cortical regions can be identified, but this introduces subjectivity and requires arbitrary cutoff values
Solution Approach 1:
The patent changes the regularization norm to L1, which naturally produces sparse solutions with clear edges between active and background regions. This parameter change eliminates the need for subjective thresholding by creating solutions where active sources are naturally distinguished from background activity through sparsity, providing an objective and automated method for source identification.
Data Source
AI summary
Three-dimensional electrical source imaging of electrical activity in a biological system (e.g., brain or heart) may involve sensor array recording, at multiple locations, of signals (e.g. electrical or magnetic signals) of electrical activity. An initial estimate of an underlying source is obtained. Edge sparsity is imposed on the estimated electrical activity to eliminate background activity and create clear edges between an active source and background activity. The initial estimate is iteratively reweighted and a series of subsequent optimization problems launched to converge to a more accurate estimation of the underlying source. Images depicting a spatial distribution of a source are generated based on the iteratively reweighted edge sparsity, and the time-course of activity for estimated sources generated. Iterative reweighting penalizes locations with smaller source amplitude based on solutions obtained in previous iterations, and continues until a desirable solution is obtained with clear edges. The objective approach does not require subjective thresholding.